Multi_Lable_Text_Classification.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/amirkasaei/Multi-Lable-Text-Classification/blob/main/Multi_Lable_Text_Classification.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rG540utTBHVi"
      },
      "source": [
        "# **Libraries & Constants**\n",
        "\n",
        "---\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fo1HnzyLo6bh"
      },
      "source": [
        "## **Hazm**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 8,
      "metadata": {
        "id": "3S5FbX6h80Ff"
      },
      "outputs": [],
      "source": [
        "pip install hazm -q"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "!wget https://github.com/sobhe/hazm/releases/download/v0.5/resources-0.5.zip\n",
        "!unzip resources-0.5.zip -d resources"
      ],
      "metadata": {
        "id": "S1YeYnUnxfH0"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Q1qXvuLDsHy-"
      },
      "source": [
        "## **imports**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "U9giafB6qqAv"
      },
      "outputs": [],
      "source": [
        "import pandas as pd\n",
        "import tensorflow as tf\n",
        "import numpy as np\n",
        "from keras.models import Sequential, load_model\n",
        "from keras.layers import Dense, Input, Softmax, BatchNormalization, Dropout, Flatten, LSTM, GRU, SimpleRNN, Concatenate, concatenate, MaxPooling1D, GlobalMaxPool1D, Bidirectional, GlobalMaxPooling1D, GlobalAveragePooling1D\n",
        "# from keras.layers.merge import concatenate\n",
        "from keras.losses import CategoricalCrossentropy\n",
        "from keras.metrics import CategoricalAccuracy\n",
        "from keras.preprocessing.text import Tokenizer\n",
        "from tensorflow.keras.optimizers import RMSprop, Adam, Adamax, Nadam, SGD\n",
        "from tensorflow.keras import Model\n",
        "from tensorflow.keras.optimizers.schedules import InverseTimeDecay, ExponentialDecay\n",
        "from tensorflow.keras.utils import to_categorical, plot_model\n",
        "from keras.callbacks import ModelCheckpoint\n",
        "from keras.preprocessing.sequence import pad_sequences\n",
        "from sklearn.preprocessing import LabelEncoder\n",
        "from sklearn.model_selection import train_test_split, KFold\n",
        "from sklearn import metrics\n",
        "from matplotlib import pyplot as plt\n",
        "from sklearn.model_selection import KFold\n",
        "import os\n",
        "import string\n",
        "import cv2 as cv\n",
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
        "from copy import deepcopy\n",
        "from __future__ import unicode_literals\n",
        "import hazm"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# PerSick dataset\n",
        "!gdown --id 1f9eeHpkoyKmTsuZeNkXHrZK-4TP3BtOP\n",
        "# subject classification dataset\n",
        "!gdown --id 1ja5_brqB4kFhcjSCT_DhGqOmeEPfqGPc\n",
        "# dataset with subject and score classes (main for RNN)\n",
        "!gdown --id 1UJMcB24gRzj6-uyNR6y33Vv1R8QdvzTX"
      ],
      "metadata": {
        "id": "OM4H9httIqoK"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **Helper functions**"
      ],
      "metadata": {
        "id": "LXvHNESIZ_BV"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def encode6ClassToOneHot(arr):\n",
        "  tmp = np.zeros((arr.shape[0], 6))\n",
        "  tmp[np.where(arr==1)[0], :] = [1, 0, 0, 0, 0, 0]\n",
        "  tmp[np.where(arr==2)[0], :] = [0, 1, 0, 0, 0, 0]\n",
        "  tmp[np.where(arr==3)[0], :] = [0, 0, 1, 0, 0, 0]\n",
        "  tmp[np.where(arr==4)[0], :] = [0, 0, 0, 1, 0, 0]\n",
        "  tmp[np.where(arr==5)[0], :] = [0, 0, 0, 0, 1, 0]\n",
        "  tmp[np.where(arr==6)[0], :] = [0, 0, 0, 0, 0, 1]\n",
        "  return tmp\n",
        "\n",
        "def argmaxKeepDimensions(arr):\n",
        "  tmp = np.zeros_like(arr)\n",
        "  tmp[np.arange(len(arr)), arr.argmax(1)] = 1\n",
        "  return tmp\n",
        "\n",
        "def convertOneHotToClassNumber(arr):\n",
        "  tmp = np.zeros((arr.shape[0], 1))\n",
        "  tmp[np.arange(len(arr)), 0] = arr.argmax(1) + 1\n",
        "  return tmp"
      ],
      "metadata": {
        "id": "UTPdekKNZ9o9"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# create correct input form for SC NN\n",
        "def encodeSubjects(subjects):\n",
        "  subjects_encoded = []\n",
        "  for subject in subjects:\n",
        "    subjects_encoded.append(encodeWord(subject))\n",
        "\n",
        "  subjects_encoded = np.array(subjects_encoded)\n",
        "\n",
        "  subjects_encoded_pad = pad_sequences(subjects_encoded, maxlen=max_len, padding='post')\n",
        "  subjects_encoded_OH = to_categorical(subjects_encoded_pad, num_classes=len(char_index))\n",
        "  return subjects_encoded_OH"
      ],
      "metadata": {
        "id": "ePSzjYm6hvqf"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def showPlots(model):\n",
        "    plt.plot(model.history['accuracy'],\n",
        "              label='training accuracy', marker='.', color='green')\n",
        "    plt.plot(model.history['val_accuracy'],\n",
        "              label='test accuracy', marker='.', color='blue')\n",
        "    plt.xlabel('Epochs')\n",
        "    plt.ylabel('Accuracy')\n",
        "    plt.legend()\n",
        "    plt.show()\n",
        "\n",
        "    plt.plot(model.history['loss'],\n",
        "              label='training loss', marker='.', color='green')\n",
        "    plt.plot(model.history['val_loss'],\n",
        "              label='test loss', marker='.', color='blue')\n",
        "    plt.xlabel('Epochs')\n",
        "    plt.ylabel('Loss')\n",
        "    plt.legend()\n",
        "    plt.show()"
      ],
      "metadata": {
        "id": "kfL7Q0Tqn2-z"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **Subject Classsification NN (RNN)**\n",
        "\n"
      ],
      "metadata": {
        "id": "gzcKYBWePRVS"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Preprocessing**"
      ],
      "metadata": {
        "id": "LIA_l44HaJXD"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# read subject classification dataset\n",
        "scdataset = pd.read_excel('/content/subject classification dataset.xlsx')\n",
        "\n",
        "# replace nan values with 'NA'\n",
        "scdataset = scdataset.replace(to_replace = np.nan, value = 'NA')"
      ],
      "metadata": {
        "id": "N9jc-pM_PCnF"
      },
      "execution_count": 26,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# data augmentation\n",
        "\n",
        "classes = scdataset['Class']\n",
        "class1_indices = np.where(classes == 1)\n",
        "class2_indices = np.where(classes == 2)\n",
        "class3_indices = np.where(classes == 3)\n",
        "\n",
        "dataset_words = scdataset['Word'].values.tolist()\n",
        "\n",
        "def augmentData(dataset_words, class_indices, class_num):\n",
        "  type1 = [[dataset_words[i] + 'ی', class_num] for i in class_indices[0]]\n",
        "  type2 = [[dataset_words[i] + 'ها', class_num] for i in class_indices[0]]\n",
        "  type3 = [[dataset_words[i] + 'ان', class_num] for i in class_indices[0]]\n",
        "  return type1 + type2 + type3\n",
        "\n",
        "words = np.array(augmentData(dataset_words, class1_indices, 1) + augmentData(dataset_words, class2_indices, 2) + augmentData(dataset_words, class3_indices, 3))\n",
        "\n",
        "dataAug = pd.DataFrame(words, columns=['Word', 'Class'])\n",
        "\n",
        "scdataset_aug = pd.concat([scdataset, dataAug], axis=0, ignore_index=True)\n",
        "scdataset_aug.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "id": "6s8Uexd3C1wK",
        "outputId": "3d2a0e91-7598-4869-f92b-e8a8d374a817"
      },
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "       Word Class\n",
              "0       خرس     4\n",
              "1  گاو میش      4\n",
              "2   کروکدیل     4\n",
              "3      گوزن     4\n",
              "4     شاهین     4"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-142d8b87-f902-4e41-897b-0ada78de1270\" class=\"colab-df-container\">\n",
              "    <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Word</th>\n",
              "      <th>Class</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>خرس</td>\n",
              "      <td>4</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>گاو میش</td>\n",
              "      <td>4</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>کروکدیل</td>\n",
              "      <td>4</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>گوزن</td>\n",
              "      <td>4</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>شاهین</td>\n",
              "      <td>4</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "    <div class=\"colab-df-buttons\">\n",
              "\n",
              "  <div class=\"colab-df-container\">\n",
              "    <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-142d8b87-f902-4e41-897b-0ada78de1270')\"\n",
              "            title=\"Convert this dataframe to an interactive table.\"\n",
              "            style=\"display:none;\">\n",
              "\n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\" viewBox=\"0 -960 960 960\">\n",
              "    <path d=\"M120-120v-720h720v720H120Zm60-500h600v-160H180v160Zm220 220h160v-160H400v160Zm0 220h160v-160H400v160ZM180-400h160v-160H180v160Zm440 0h160v-160H620v160ZM180-180h160v-160H180v160Zm440 0h160v-160H620v160Z\"/>\n",
              "  </svg>\n",
              "    </button>\n",
              "\n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    .colab-df-buttons div {\n",
              "      margin-bottom: 4px;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "    <script>\n",
              "      const buttonEl =\n",
              "        document.querySelector('#df-142d8b87-f902-4e41-897b-0ada78de1270 button.colab-df-convert');\n",
              "      buttonEl.style.display =\n",
              "        google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "      async function convertToInteractive(key) {\n",
              "        const element = document.querySelector('#df-142d8b87-f902-4e41-897b-0ada78de1270');\n",
              "        const dataTable =\n",
              "          await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                    [key], {});\n",
              "        if (!dataTable) return;\n",
              "\n",
              "        const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "          '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "          + ' to learn more about interactive tables.';\n",
              "        element.innerHTML = '';\n",
              "        dataTable['output_type'] = 'display_data';\n",
              "        await google.colab.output.renderOutput(dataTable, element);\n",
              "        const docLink = document.createElement('div');\n",
              "        docLink.innerHTML = docLinkHtml;\n",
              "        element.appendChild(docLink);\n",
              "      }\n",
              "    </script>\n",
              "  </div>\n",
              "\n",
              "\n",
              "<div id=\"df-46519d1b-e069-438e-bb50-86e9a12a5d6a\">\n",
              "  <button class=\"colab-df-quickchart\" onclick=\"quickchart('df-46519d1b-e069-438e-bb50-86e9a12a5d6a')\"\n",
              "            title=\"Suggest charts\"\n",
              "            style=\"display:none;\">\n",
              "\n",
              "<svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "     width=\"24px\">\n",
              "    <g>\n",
              "        <path d=\"M19 3H5c-1.1 0-2 .9-2 2v14c0 1.1.9 2 2 2h14c1.1 0 2-.9 2-2V5c0-1.1-.9-2-2-2zM9 17H7v-7h2v7zm4 0h-2V7h2v10zm4 0h-2v-4h2v4z\"/>\n",
              "    </g>\n",
              "</svg>\n",
              "  </button>\n",
              "\n",
              "<style>\n",
              "  .colab-df-quickchart {\n",
              "      --bg-color: #E8F0FE;\n",
              "      --fill-color: #1967D2;\n",
              "      --hover-bg-color: #E2EBFA;\n",
              "      --hover-fill-color: #174EA6;\n",
              "      --disabled-fill-color: #AAA;\n",
              "      --disabled-bg-color: #DDD;\n",
              "  }\n",
              "\n",
              "  [theme=dark] .colab-df-quickchart {\n",
              "      --bg-color: #3B4455;\n",
              "      --fill-color: #D2E3FC;\n",
              "      --hover-bg-color: #434B5C;\n",
              "      --hover-fill-color: #FFFFFF;\n",
              "      --disabled-bg-color: #3B4455;\n",
              "      --disabled-fill-color: #666;\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart {\n",
              "    background-color: var(--bg-color);\n",
              "    border: none;\n",
              "    border-radius: 50%;\n",
              "    cursor: pointer;\n",
              "    display: none;\n",
              "    fill: var(--fill-color);\n",
              "    height: 32px;\n",
              "    padding: 0;\n",
              "    width: 32px;\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart:hover {\n",
              "    background-color: var(--hover-bg-color);\n",
              "    box-shadow: 0 1px 2px rgba(60, 64, 67, 0.3), 0 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "    fill: var(--button-hover-fill-color);\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart-complete:disabled,\n",
              "  .colab-df-quickchart-complete:disabled:hover {\n",
              "    background-color: var(--disabled-bg-color);\n",
              "    fill: var(--disabled-fill-color);\n",
              "    box-shadow: none;\n",
              "  }\n",
              "\n",
              "  .colab-df-spinner {\n",
              "    border: 2px solid var(--fill-color);\n",
              "    border-color: transparent;\n",
              "    border-bottom-color: var(--fill-color);\n",
              "    animation:\n",
              "      spin 1s steps(1) infinite;\n",
              "  }\n",
              "\n",
              "  @keyframes spin {\n",
              "    0% {\n",
              "      border-color: transparent;\n",
              "      border-bottom-color: var(--fill-color);\n",
              "      border-left-color: var(--fill-color);\n",
              "    }\n",
              "    20% {\n",
              "      border-color: transparent;\n",
              "      border-left-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "    }\n",
              "    30% {\n",
              "      border-color: transparent;\n",
              "      border-left-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "      border-right-color: var(--fill-color);\n",
              "    }\n",
              "    40% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "    }\n",
              "    60% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "    }\n",
              "    80% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "      border-bottom-color: var(--fill-color);\n",
              "    }\n",
              "    90% {\n",
              "      border-color: transparent;\n",
              "      border-bottom-color: var(--fill-color);\n",
              "    }\n",
              "  }\n",
              "</style>\n",
              "\n",
              "  <script>\n",
              "    async function quickchart(key) {\n",
              "      const quickchartButtonEl =\n",
              "        document.querySelector('#' + key + ' button');\n",
              "      quickchartButtonEl.disabled = true;  // To prevent multiple clicks.\n",
              "      quickchartButtonEl.classList.add('colab-df-spinner');\n",
              "      try {\n",
              "        const charts = await google.colab.kernel.invokeFunction(\n",
              "            'suggestCharts', [key], {});\n",
              "      } catch (error) {\n",
              "        console.error('Error during call to suggestCharts:', error);\n",
              "      }\n",
              "      quickchartButtonEl.classList.remove('colab-df-spinner');\n",
              "      quickchartButtonEl.classList.add('colab-df-quickchart-complete');\n",
              "    }\n",
              "    (() => {\n",
              "      let quickchartButtonEl =\n",
              "        document.querySelector('#df-46519d1b-e069-438e-bb50-86e9a12a5d6a button');\n",
              "      quickchartButtonEl.style.display =\n",
              "        google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "    })();\n",
              "  </script>\n",
              "</div>\n",
              "    </div>\n",
              "  </div>\n"
            ],
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "dataframe",
              "variable_name": "scdataset_aug",
              "summary": "{\n  \"name\": \"scdataset_aug\",\n  \"rows\": 612,\n  \"fields\": [\n    {\n      \"column\": \"Word\",\n      \"properties\": {\n        \"dtype\": \"string\",\n        \"num_unique_values\": 596,\n        \"samples\": [\n          \"\\u0622\\u0648\\u0627\\u0632\",\n          \"\\u0628\\u0646\\u0627\\u06cc\",\n          \"\\u062f\\u0634\\u0645\\u0646\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"Class\",\n      \"properties\": {\n        \"dtype\": \"category\",\n        \"num_unique_values\": 9,\n        \"samples\": [\n          \"2\",\n          1,\n          6\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    }\n  ]\n}"
            }
          },
          "metadata": {},
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "X_train_sc = np.array(scdataset['Word'])\n",
        "y_train_sc = np.array(scdataset['Class'])\n",
        "\n",
        "# one hot encoding for y_true\n",
        "y_train_sc = encode6ClassToOneHot(y_train_sc)\n",
        "\n",
        "print(X_train_sc.shape)\n",
        "print(y_train_sc.shape)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "g12TGUe4Rheu",
        "outputId": "2f9898fe-dc23-44f0-9e3f-cfc2f9ac259d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(465,)\n",
            "(465, 6)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# generate character encoding\n",
        "puncs = list(string.punctuation) + [' ', '1', '\\xa0', 'ِ', '\\u200c', 'ُ', 'َ', '٫', 'ّ', 'ْ', '۰', '۳', '٫']\n",
        "oov_token = 'not available char'\n",
        "char_index = {oov_token: 0}\n",
        "\n",
        "# encode input with encoding character set\n",
        "for i in range(len(X_train_sc)):\n",
        "    chars = list(X_train_sc[i])\n",
        "    for c in chars:\n",
        "        if c not in char_index and c not in puncs:\n",
        "            char_index[c] = len(char_index)\n",
        "\n",
        "len(char_index)\n",
        "print(char_index)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "uvbmQ4zCNrRN",
        "outputId": "0d30decd-71e1-470c-e644-b3d0a776c063"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "{'not available char': 0, 'خ': 1, 'ر': 2, 'س': 3, 'گ': 4, 'ا': 5, 'و': 6, 'م': 7, 'ی': 8, 'ش': 9, 'ک': 10, 'د': 11, 'ل': 12, 'ز': 13, 'ن': 14, 'ه': 15, 'آ': 16, 'ق': 17, 'ب': 18, 'ف': 19, 'ص': 20, 'غ': 21, 'ج': 22, 'ع': 23, 'ت': 24, 'چ': 25, 'پ': 26, 'ح': 27, 'ئ': 28, 'ط': 29, 'ذ': 30, 'ض': 31, 'N': 32, 'A': 33, 'ث': 34, 'ظ': 35, 'ژ': 36}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# encode words with character encoder\n",
        "def encodeWord(word, dic=char_index, oov=oov_token):\n",
        "    return [dic[c] if c in dic else dic[oov] for c in list(word)]\n",
        "\n",
        "X_train_sc_encoded = np.array([encodeWord(each) for each in X_train_sc])\n",
        "max_len = max([len(each) for each in X_train_sc_encoded])\n",
        "\n",
        "print('input shape: ', X_train_sc_encoded.shape)\n",
        "print('max length in input elements: ', max_len)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "VkDPd2VBUmP4",
        "outputId": "110c0784-f0b7-43ed-d437-77fae74fe95c"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "input shape:  (465,)\n",
            "max length in input elements:  11\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:5: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray.\n",
            "  \"\"\"\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# ADD padding to sample\n",
        "X_train_sc_pad = pad_sequences(X_train_sc_encoded, maxlen=max_len, padding='post')\n",
        "\n",
        "# convert to One Hot encoding\n",
        "X_train_sc_OH = to_categorical(X_train_sc_pad, num_classes=len(char_index))\n",
        "\n",
        "X_train_sc_OH.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rWtUa_ClVmvt",
        "outputId": "51f5a552-6749-4a65-bdee-51e7ed3d4c90"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(465, 11, 37)"
            ]
          },
          "metadata": {},
          "execution_count": 58
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Implementation**"
      ],
      "metadata": {
        "id": "9rY02m-adIsR"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "class SubjectClassificationNN:\n",
        "  def __init__(self, X_train, y_train):\n",
        "    self.X_train = X_train\n",
        "    self.y_train = y_train\n",
        "\n",
        "    input = Input(shape=X_train.shape[1:])\n",
        "    x = LSTM(128, return_sequences=True)(input)\n",
        "    x = GRU(128, return_sequences=True)(x)\n",
        "\n",
        "    flatten = Flatten()(x)\n",
        "\n",
        "    out = Dense(256, activation=\"tanh\")(flatten)\n",
        "    out = Dense(64, activation=\"tanh\")(out)\n",
        "    out = Dense(6, activation=\"softmax\")(out)\n",
        "\n",
        "    # Compile the Model\n",
        "    self.model = Model(inputs=input, outputs=out)\n",
        "    self.model.compile(loss='mean_squared_error', optimizer=Adam(0.001), metrics=['accuracy'])\n",
        "    # self.model.summary()\n",
        "\n",
        "\n",
        "  def train(self):\n",
        "    filepath = 'my_best_model.hdf5'\n",
        "    checkpoint = ModelCheckpoint(filepath=filepath,\n",
        "                              monitor='accuracy',\n",
        "                              verbose=0,\n",
        "                              save_best_only=True,\n",
        "                              mode='max')\n",
        "\n",
        "    self.history = self.model.fit(self.X_train, self.y_train, epochs=1000, batch_size=64, verbose=0, callbacks=[checkpoint]).history\n",
        "\n",
        "    self.model = load_model(filepath)\n",
        "\n",
        "  def getModelResult(self):\n",
        "    return self.model.evaluate(self.X_train, self.y_train)\n",
        "\n",
        "  def predictClass(self, subjects):\n",
        "    encoded_subjects = encodeSubjects(subjects) # encodes input\n",
        "    pred = self.model.predict(encoded_subjects) # calculate probablity\n",
        "    pred_OH = argmaxKeepDimensions(pred)  # One hot classification\n",
        "    pred_class = convertOneHotToClassNumber(pred_OH)  # class number\n",
        "    return pred_class\n"
      ],
      "metadata": {
        "id": "23LTrvoGvuHQ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rtWMFAy_QN7a"
      },
      "source": [
        "# **Preprocessing**\n",
        "\n",
        "---\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Generate Dataset**"
      ],
      "metadata": {
        "id": "8bAfP_I6BJEP"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "EpZwtddhuCIN"
      },
      "outputs": [],
      "source": [
        "path = 'https://raw.githubusercontent.com/Ledengary/COPER/main/Datasets/PerSICK.csv'\n",
        "# read csv data from github\n",
        "dataset = pd.read_csv(path)\n",
        "# save dataset file\n",
        "dataset.to_excel('raw_dataset.xlsx')\n",
        "# delete rows with nan elements\n",
        "dataset = dataset.dropna()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 14,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "id": "7LMLFOg4Qcvj",
        "outputId": "176d8af6-9be5-4626-9ec1-556054dd8609"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "  score                                          sentence1  \\\n",
              "0   4.5  گروهی از بچه ها در حیاط بازی می کنند و پیرمردی...   \n",
              "1   3.2  گروهی از کودکان در خانه مشغول بازی هستند و هیچ...   \n",
              "2   4.7  پسران جوان در فضای باز بازی می کنند و مرد در ه...   \n",
              "3   3.4  بچه ها در بیرون از خانه و در کنار یک مرد با لب...   \n",
              "4   3.7  پسران جوان در فضای باز بازی می کنند و مرد در ه...   \n",
              "\n",
              "                                           sentence2  \n",
              "0  گروهی از پسران در حیاط بازی می کنند و مردی در ...  \n",
              "1  گروهی از بچه ها در حیاط بازی می کنند و پیرمردی...  \n",
              "2  بچه ها در بیرون از خانه و در کنار یک مرد با لب...  \n",
              "3  گروهی از بچه ها در حیاط بازی می کنند و پیرمردی...  \n",
              "4  گروهی از بچه ها در حیاط بازی می کنند و پیرمردی...  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-30e87a97-2959-422e-9430-9d0a60df673c\" class=\"colab-df-container\">\n",
              "    <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>score</th>\n",
              "      <th>sentence1</th>\n",
              "      <th>sentence2</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>4.5</td>\n",
              "      <td>گروهی از بچه ها در حیاط بازی می کنند و پیرمردی...</td>\n",
              "      <td>گروهی از پسران در حیاط بازی می کنند و مردی در ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>3.2</td>\n",
              "      <td>گروهی از کودکان در خانه مشغول بازی هستند و هیچ...</td>\n",
              "      <td>گروهی از بچه ها در حیاط بازی می کنند و پیرمردی...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>4.7</td>\n",
              "      <td>پسران جوان در فضای باز بازی می کنند و مرد در ه...</td>\n",
              "      <td>بچه ها در بیرون از خانه و در کنار یک مرد با لب...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3.4</td>\n",
              "      <td>بچه ها در بیرون از خانه و در کنار یک مرد با لب...</td>\n",
              "      <td>گروهی از بچه ها در حیاط بازی می کنند و پیرمردی...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>3.7</td>\n",
              "      <td>پسران جوان در فضای باز بازی می کنند و مرد در ه...</td>\n",
              "      <td>گروهی از بچه ها در حیاط بازی می کنند و پیرمردی...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "    <div class=\"colab-df-buttons\">\n",
              "\n",
              "  <div class=\"colab-df-container\">\n",
              "    <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-30e87a97-2959-422e-9430-9d0a60df673c')\"\n",
              "            title=\"Convert this dataframe to an interactive table.\"\n",
              "            style=\"display:none;\">\n",
              "\n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\" viewBox=\"0 -960 960 960\">\n",
              "    <path d=\"M120-120v-720h720v720H120Zm60-500h600v-160H180v160Zm220 220h160v-160H400v160Zm0 220h160v-160H400v160ZM180-400h160v-160H180v160Zm440 0h160v-160H620v160ZM180-180h160v-160H180v160Zm440 0h160v-160H620v160Z\"/>\n",
              "  </svg>\n",
              "    </button>\n",
              "\n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    .colab-df-buttons div {\n",
              "      margin-bottom: 4px;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "    <script>\n",
              "      const buttonEl =\n",
              "        document.querySelector('#df-30e87a97-2959-422e-9430-9d0a60df673c button.colab-df-convert');\n",
              "      buttonEl.style.display =\n",
              "        google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "      async function convertToInteractive(key) {\n",
              "        const element = document.querySelector('#df-30e87a97-2959-422e-9430-9d0a60df673c');\n",
              "        const dataTable =\n",
              "          await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                    [key], {});\n",
              "        if (!dataTable) return;\n",
              "\n",
              "        const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "          '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "          + ' to learn more about interactive tables.';\n",
              "        element.innerHTML = '';\n",
              "        dataTable['output_type'] = 'display_data';\n",
              "        await google.colab.output.renderOutput(dataTable, element);\n",
              "        const docLink = document.createElement('div');\n",
              "        docLink.innerHTML = docLinkHtml;\n",
              "        element.appendChild(docLink);\n",
              "      }\n",
              "    </script>\n",
              "  </div>\n",
              "\n",
              "\n",
              "<div id=\"df-1a1039fe-ead6-4747-b006-20854b08fffa\">\n",
              "  <button class=\"colab-df-quickchart\" onclick=\"quickchart('df-1a1039fe-ead6-4747-b006-20854b08fffa')\"\n",
              "            title=\"Suggest charts\"\n",
              "            style=\"display:none;\">\n",
              "\n",
              "<svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "     width=\"24px\">\n",
              "    <g>\n",
              "        <path d=\"M19 3H5c-1.1 0-2 .9-2 2v14c0 1.1.9 2 2 2h14c1.1 0 2-.9 2-2V5c0-1.1-.9-2-2-2zM9 17H7v-7h2v7zm4 0h-2V7h2v10zm4 0h-2v-4h2v4z\"/>\n",
              "    </g>\n",
              "</svg>\n",
              "  </button>\n",
              "\n",
              "<style>\n",
              "  .colab-df-quickchart {\n",
              "      --bg-color: #E8F0FE;\n",
              "      --fill-color: #1967D2;\n",
              "      --hover-bg-color: #E2EBFA;\n",
              "      --hover-fill-color: #174EA6;\n",
              "      --disabled-fill-color: #AAA;\n",
              "      --disabled-bg-color: #DDD;\n",
              "  }\n",
              "\n",
              "  [theme=dark] .colab-df-quickchart {\n",
              "      --bg-color: #3B4455;\n",
              "      --fill-color: #D2E3FC;\n",
              "      --hover-bg-color: #434B5C;\n",
              "      --hover-fill-color: #FFFFFF;\n",
              "      --disabled-bg-color: #3B4455;\n",
              "      --disabled-fill-color: #666;\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart {\n",
              "    background-color: var(--bg-color);\n",
              "    border: none;\n",
              "    border-radius: 50%;\n",
              "    cursor: pointer;\n",
              "    display: none;\n",
              "    fill: var(--fill-color);\n",
              "    height: 32px;\n",
              "    padding: 0;\n",
              "    width: 32px;\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart:hover {\n",
              "    background-color: var(--hover-bg-color);\n",
              "    box-shadow: 0 1px 2px rgba(60, 64, 67, 0.3), 0 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "    fill: var(--button-hover-fill-color);\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart-complete:disabled,\n",
              "  .colab-df-quickchart-complete:disabled:hover {\n",
              "    background-color: var(--disabled-bg-color);\n",
              "    fill: var(--disabled-fill-color);\n",
              "    box-shadow: none;\n",
              "  }\n",
              "\n",
              "  .colab-df-spinner {\n",
              "    border: 2px solid var(--fill-color);\n",
              "    border-color: transparent;\n",
              "    border-bottom-color: var(--fill-color);\n",
              "    animation:\n",
              "      spin 1s steps(1) infinite;\n",
              "  }\n",
              "\n",
              "  @keyframes spin {\n",
              "    0% {\n",
              "      border-color: transparent;\n",
              "      border-bottom-color: var(--fill-color);\n",
              "      border-left-color: var(--fill-color);\n",
              "    }\n",
              "    20% {\n",
              "      border-color: transparent;\n",
              "      border-left-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "    }\n",
              "    30% {\n",
              "      border-color: transparent;\n",
              "      border-left-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "      border-right-color: var(--fill-color);\n",
              "    }\n",
              "    40% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "    }\n",
              "    60% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "    }\n",
              "    80% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "      border-bottom-color: var(--fill-color);\n",
              "    }\n",
              "    90% {\n",
              "      border-color: transparent;\n",
              "      border-bottom-color: var(--fill-color);\n",
              "    }\n",
              "  }\n",
              "</style>\n",
              "\n",
              "  <script>\n",
              "    async function quickchart(key) {\n",
              "      const quickchartButtonEl =\n",
              "        document.querySelector('#' + key + ' button');\n",
              "      quickchartButtonEl.disabled = true;  // To prevent multiple clicks.\n",
              "      quickchartButtonEl.classList.add('colab-df-spinner');\n",
              "      try {\n",
              "        const charts = await google.colab.kernel.invokeFunction(\n",
              "            'suggestCharts', [key], {});\n",
              "      } catch (error) {\n",
              "        console.error('Error during call to suggestCharts:', error);\n",
              "      }\n",
              "      quickchartButtonEl.classList.remove('colab-df-spinner');\n",
              "      quickchartButtonEl.classList.add('colab-df-quickchart-complete');\n",
              "    }\n",
              "    (() => {\n",
              "      let quickchartButtonEl =\n",
              "        document.querySelector('#df-1a1039fe-ead6-4747-b006-20854b08fffa button');\n",
              "      quickchartButtonEl.style.display =\n",
              "        google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "    })();\n",
              "  </script>\n",
              "</div>\n",
              "    </div>\n",
              "  </div>\n"
            ],
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "dataframe",
              "variable_name": "dataset",
              "summary": "{\n  \"name\": \"dataset\",\n  \"rows\": 3101,\n  \"fields\": [\n    {\n      \"column\": \"score\",\n      \"properties\": {\n        \"dtype\": \"category\",\n        \"num_unique_values\": 106,\n        \"samples\": [\n          \"3.95\",\n          \"3.5\",\n          \"3.7\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"sentence1\",\n      \"properties\": {\n        \"dtype\": \"string\",\n        \"num_unique_values\": 2260,\n        \"samples\": [\n          \"\\u062f\\u0648\\u0686\\u0631\\u062e\\u0647 \\u0633\\u0648\\u0627\\u0631 \\u062f\\u0631 \\u062e\\u06cc\\u0627\\u0628\\u0627\\u0646 \\u0647\\u0627\\u06cc \\u0634\\u0647\\u0631 \\u0633\\u0648\\u0627\\u0631 \\u0645\\u06cc \\u0634\\u0648\\u062f\",\n          \"\\u0645\\u0631\\u062f\\u06cc \\u062f\\u0631 \\u062d\\u0627\\u0644 \\u0628\\u0631\\u06cc\\u062f\\u0646 \\u0646\\u0648\\u0627\\u0631 \\u062c\\u0639\\u0628\\u0647 \\u0627\\u0633\\u062a\",\n          \"\\u0645\\u0642\\u062f\\u0627\\u0631\\u06cc \\u0627\\u0632 \\u063a\\u0630\\u0627 \\u062a\\u0648\\u0633\\u0637 \\u06cc\\u06a9 \\u0622\\u0634\\u067e\\u0632 \\u062a\\u0647\\u06cc\\u0647 \\u0645\\u06cc \\u0634\\u0648\\u062f\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"sentence2\",\n      \"properties\": {\n        \"dtype\": \"string\",\n        \"num_unique_values\": 2233,\n        \"samples\": [\n          \"\\u0628\\u06cc\\u0645\\u0627\\u0631 \\u062f\\u0631 \\u062d\\u0627\\u0644 \\u06a9\\u0645\\u06a9 \\u0628\\u0647 \\u062f\\u06a9\\u062a\\u0631 \\u0627\\u0633\\u062a\",\n          \"\\u0645\\u0631\\u062f\\u06cc \\u0631\\u0648\\u06cc \\u06cc\\u06a9 \\u062a\\u06a9\\u0647 \\u0646\\u0627\\u0646 \\u06a9\\u0631\\u0647 \\u06a9\\u0647 \\u0628\\u0632\\u0631\\u06af \\u0627\\u0633\\u062a \\u06a9\\u0631\\u0647 \\u067e\\u062e\\u0634 \\u0645\\u06cc \\u06a9\\u0646\\u062f\",\n          \"\\u0647\\u06cc\\u0686 \\u06af\\u0631\\u0628\\u0647 \\u0627\\u06cc \\u0628\\u0627 \\u062f\\u0633\\u062a\\u06af\\u0627\\u0647 \\u0628\\u0627\\u0632\\u06cc \\u0646\\u0645\\u06cc \\u06a9\\u0646\\u062f\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    }\n  ]\n}"
            }
          },
          "metadata": {},
          "execution_count": 14
        }
      ],
      "source": [
        "dataset.head()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 5,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "vwdwX2ZlgUv7",
        "outputId": "11a3bdfd-21f4-4d98-93db-72e925c5dbb6"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(3101, 2)\n",
            "(3101,)\n"
          ]
        }
      ],
      "source": [
        "inputs = np.array(dataset[['sentence1', 'sentence2']])\n",
        "scores = np.array(dataset['score'])\n",
        "print(inputs.shape)\n",
        "print(scores.shape)"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# hazm objects\n",
        "normalizer = hazm.Normalizer()\n",
        "lemmatizer = hazm.Lemmatizer()\n",
        "stemmer = hazm.Stemmer()\n",
        "tagger = hazm.POSTagger(model='/content/resources/postagger.model')\n",
        "parser = hazm.DependencyParser(tagger = tagger, lemmatizer = lemmatizer)"
      ],
      "metadata": {
        "id": "kxOxKWSI6Y5o"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# normalzing data\n",
        "for i in range(len(inputs)):\n",
        "  inputs[i] = [normalizer.normalize(inputs[i][0]), normalizer.normalize(inputs[i][1])]"
      ],
      "metadata": {
        "id": "GmQAHE4z6gyD"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def findSubject(input):\n",
        "  tokens = hazm.word_tokenize(input)\n",
        "  parseGraph = parser.parse(tokens)\n",
        "  # subjects of each sentence\n",
        "  subjects = []\n",
        "  for i in range(len(tokens)):\n",
        "    tokenDetails = parseGraph.get_by_address(i)\n",
        "    if tokenDetails['rel'] == 'SBJ':\n",
        "      subjects.append(tokenDetails['word'])\n",
        "  return subjects\n",
        "\n",
        "def findSubjects(input):\n",
        "  for i in range(len(input)):\n",
        "    print(i, end=', ', )\n",
        "    subjects1Col.append(','.join([str(elem) for elem in findSubject(input[i][0])]))\n",
        "    subjects2Col.append(','.join([str(elem) for elem in findSubject(input[i][1])]))"
      ],
      "metadata": {
        "id": "FLl5vb4D6iWA"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "subjects1Col = []\n",
        "subjects2Col = []\n",
        "with tf.device('/GPU:0'):\n",
        "  findSubjects(inputs)\n",
        "\n",
        "dftest = pd.DataFrame(scores, columns=['score'])\n",
        "dftest['Sentence1'] = inputs[:, 0]\n",
        "dftest['Subject1'] = subjects1Col\n",
        "dftest['Sentence2'] = inputs[:, 1]\n",
        "dftest['Subject2'] = subjects2Col\n",
        "dftest.style\n",
        "dftest.to_excel(\"dataset6.xlsx\", sheet_name='Sheet_name_1')"
      ],
      "metadata": {
        "id": "7ZflumiL6bcy"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Load Generated Dataset**"
      ],
      "metadata": {
        "id": "v3uUEAk9lsmY"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "dataset = pd.read_excel('/content/dataset.xlsx')"
      ],
      "metadata": {
        "id": "ZfFocT8UHF5a"
      },
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "dataset = dataset.replace(to_replace = np.nan, value = 'NA')"
      ],
      "metadata": {
        "id": "5f4ZfSGzNMr5"
      },
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# drop rows with unacceptable scores\n",
        "for j in range(len(dataset)):\n",
        "  try:\n",
        "      dataset['score'][j] = float(dataset['score'][j])\n",
        "  except (TypeError, ValueError):\n",
        "      dataset.drop(j, axis=0, inplace=True)\n"
      ],
      "metadata": {
        "id": "mttSEBvfZjRe"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "dataset.head()"
      ],
      "metadata": {
        "id": "CrnEBarGcRfa",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "outputId": "bcf3a475-7439-4413-daa8-7fe0418bb943"
      },
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   Unnamed: 0 score                                          Sentence1  \\\n",
              "0           0   4.5  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "1           1   3.2  گروهی از کودکان در خانه مشغول بازی هستند و هیچ...   \n",
              "2           2   4.7  پسران جوان در فضای باز بازی می‌کنند و مرد در ه...   \n",
              "3           3   3.4  بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...   \n",
              "4           4   3.7  پسران جوان در فضای باز بازی می‌کنند و مرد در ه...   \n",
              "\n",
              "        Subject1                                          Sentence2  \\\n",
              "0  گروهی,پیرمردی  گروهی از پسران در حیاط بازی می‌کنند و مردی در ...   \n",
              "1     گروهی,مردی  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "2      پسران,مرد  بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...   \n",
              "3         بچه‌ها  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "4      پسران,مرد  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "\n",
              "           Subject2  \n",
              "0  گروهی,مردی,زمینه  \n",
              "1     گروهی,پیرمردی  \n",
              "2            بچه‌ها  \n",
              "3     گروهی,پیرمردی  \n",
              "4     گروهی,پیرمردی  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-ce42d841-27c6-4d7f-8483-299329ce42d0\" class=\"colab-df-container\">\n",
              "    <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Unnamed: 0</th>\n",
              "      <th>score</th>\n",
              "      <th>Sentence1</th>\n",
              "      <th>Subject1</th>\n",
              "      <th>Sentence2</th>\n",
              "      <th>Subject2</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>4.5</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "      <td>گروهی از پسران در حیاط بازی می‌کنند و مردی در ...</td>\n",
              "      <td>گروهی,مردی,زمینه</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>3.2</td>\n",
              "      <td>گروهی از کودکان در خانه مشغول بازی هستند و هیچ...</td>\n",
              "      <td>گروهی,مردی</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2</td>\n",
              "      <td>4.7</td>\n",
              "      <td>پسران جوان در فضای باز بازی می‌کنند و مرد در ه...</td>\n",
              "      <td>پسران,مرد</td>\n",
              "      <td>بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...</td>\n",
              "      <td>بچه‌ها</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3</td>\n",
              "      <td>3.4</td>\n",
              "      <td>بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...</td>\n",
              "      <td>بچه‌ها</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4</td>\n",
              "      <td>3.7</td>\n",
              "      <td>پسران جوان در فضای باز بازی می‌کنند و مرد در ه...</td>\n",
              "      <td>پسران,مرد</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "    <div class=\"colab-df-buttons\">\n",
              "\n",
              "  <div class=\"colab-df-container\">\n",
              "    <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-ce42d841-27c6-4d7f-8483-299329ce42d0')\"\n",
              "            title=\"Convert this dataframe to an interactive table.\"\n",
              "            style=\"display:none;\">\n",
              "\n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\" viewBox=\"0 -960 960 960\">\n",
              "    <path d=\"M120-120v-720h720v720H120Zm60-500h600v-160H180v160Zm220 220h160v-160H400v160Zm0 220h160v-160H400v160ZM180-400h160v-160H180v160Zm440 0h160v-160H620v160ZM180-180h160v-160H180v160Zm440 0h160v-160H620v160Z\"/>\n",
              "  </svg>\n",
              "    </button>\n",
              "\n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    .colab-df-buttons div {\n",
              "      margin-bottom: 4px;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "    <script>\n",
              "      const buttonEl =\n",
              "        document.querySelector('#df-ce42d841-27c6-4d7f-8483-299329ce42d0 button.colab-df-convert');\n",
              "      buttonEl.style.display =\n",
              "        google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "      async function convertToInteractive(key) {\n",
              "        const element = document.querySelector('#df-ce42d841-27c6-4d7f-8483-299329ce42d0');\n",
              "        const dataTable =\n",
              "          await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                    [key], {});\n",
              "        if (!dataTable) return;\n",
              "\n",
              "        const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "          '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "          + ' to learn more about interactive tables.';\n",
              "        element.innerHTML = '';\n",
              "        dataTable['output_type'] = 'display_data';\n",
              "        await google.colab.output.renderOutput(dataTable, element);\n",
              "        const docLink = document.createElement('div');\n",
              "        docLink.innerHTML = docLinkHtml;\n",
              "        element.appendChild(docLink);\n",
              "      }\n",
              "    </script>\n",
              "  </div>\n",
              "\n",
              "\n",
              "<div id=\"df-a484de3c-48c2-43b8-aa4b-348b94dbb742\">\n",
              "  <button class=\"colab-df-quickchart\" onclick=\"quickchart('df-a484de3c-48c2-43b8-aa4b-348b94dbb742')\"\n",
              "            title=\"Suggest charts\"\n",
              "            style=\"display:none;\">\n",
              "\n",
              "<svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "     width=\"24px\">\n",
              "    <g>\n",
              "        <path d=\"M19 3H5c-1.1 0-2 .9-2 2v14c0 1.1.9 2 2 2h14c1.1 0 2-.9 2-2V5c0-1.1-.9-2-2-2zM9 17H7v-7h2v7zm4 0h-2V7h2v10zm4 0h-2v-4h2v4z\"/>\n",
              "    </g>\n",
              "</svg>\n",
              "  </button>\n",
              "\n",
              "<style>\n",
              "  .colab-df-quickchart {\n",
              "      --bg-color: #E8F0FE;\n",
              "      --fill-color: #1967D2;\n",
              "      --hover-bg-color: #E2EBFA;\n",
              "      --hover-fill-color: #174EA6;\n",
              "      --disabled-fill-color: #AAA;\n",
              "      --disabled-bg-color: #DDD;\n",
              "  }\n",
              "\n",
              "  [theme=dark] .colab-df-quickchart {\n",
              "      --bg-color: #3B4455;\n",
              "      --fill-color: #D2E3FC;\n",
              "      --hover-bg-color: #434B5C;\n",
              "      --hover-fill-color: #FFFFFF;\n",
              "      --disabled-bg-color: #3B4455;\n",
              "      --disabled-fill-color: #666;\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart {\n",
              "    background-color: var(--bg-color);\n",
              "    border: none;\n",
              "    border-radius: 50%;\n",
              "    cursor: pointer;\n",
              "    display: none;\n",
              "    fill: var(--fill-color);\n",
              "    height: 32px;\n",
              "    padding: 0;\n",
              "    width: 32px;\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart:hover {\n",
              "    background-color: var(--hover-bg-color);\n",
              "    box-shadow: 0 1px 2px rgba(60, 64, 67, 0.3), 0 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "    fill: var(--button-hover-fill-color);\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart-complete:disabled,\n",
              "  .colab-df-quickchart-complete:disabled:hover {\n",
              "    background-color: var(--disabled-bg-color);\n",
              "    fill: var(--disabled-fill-color);\n",
              "    box-shadow: none;\n",
              "  }\n",
              "\n",
              "  .colab-df-spinner {\n",
              "    border: 2px solid var(--fill-color);\n",
              "    border-color: transparent;\n",
              "    border-bottom-color: var(--fill-color);\n",
              "    animation:\n",
              "      spin 1s steps(1) infinite;\n",
              "  }\n",
              "\n",
              "  @keyframes spin {\n",
              "    0% {\n",
              "      border-color: transparent;\n",
              "      border-bottom-color: var(--fill-color);\n",
              "      border-left-color: var(--fill-color);\n",
              "    }\n",
              "    20% {\n",
              "      border-color: transparent;\n",
              "      border-left-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "    }\n",
              "    30% {\n",
              "      border-color: transparent;\n",
              "      border-left-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "      border-right-color: var(--fill-color);\n",
              "    }\n",
              "    40% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "    }\n",
              "    60% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "    }\n",
              "    80% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "      border-bottom-color: var(--fill-color);\n",
              "    }\n",
              "    90% {\n",
              "      border-color: transparent;\n",
              "      border-bottom-color: var(--fill-color);\n",
              "    }\n",
              "  }\n",
              "</style>\n",
              "\n",
              "  <script>\n",
              "    async function quickchart(key) {\n",
              "      const quickchartButtonEl =\n",
              "        document.querySelector('#' + key + ' button');\n",
              "      quickchartButtonEl.disabled = true;  // To prevent multiple clicks.\n",
              "      quickchartButtonEl.classList.add('colab-df-spinner');\n",
              "      try {\n",
              "        const charts = await google.colab.kernel.invokeFunction(\n",
              "            'suggestCharts', [key], {});\n",
              "      } catch (error) {\n",
              "        console.error('Error during call to suggestCharts:', error);\n",
              "      }\n",
              "      quickchartButtonEl.classList.remove('colab-df-spinner');\n",
              "      quickchartButtonEl.classList.add('colab-df-quickchart-complete');\n",
              "    }\n",
              "    (() => {\n",
              "      let quickchartButtonEl =\n",
              "        document.querySelector('#df-a484de3c-48c2-43b8-aa4b-348b94dbb742 button');\n",
              "      quickchartButtonEl.style.display =\n",
              "        google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "    })();\n",
              "  </script>\n",
              "</div>\n",
              "    </div>\n",
              "  </div>\n"
            ],
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "dataframe",
              "variable_name": "dataset",
              "summary": "{\n  \"name\": \"dataset\",\n  \"rows\": 3097,\n  \"fields\": [\n    {\n      \"column\": \"Unnamed: 0\",\n      \"properties\": {\n        \"dtype\": \"number\",\n        \"std\": 895,\n        \"min\": 0,\n        \"max\": 3100,\n        \"num_unique_values\": 3097,\n        \"samples\": [\n          1951,\n          1312,\n          1609\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"score\",\n      \"properties\": {\n        \"dtype\": \"date\",\n        \"min\": 1.0,\n        \"max\": 5.0,\n        \"num_unique_values\": 102,\n        \"samples\": [\n          3.3,\n          2.915,\n          3.65\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"Sentence1\",\n      \"properties\": {\n        \"dtype\": \"string\",\n        \"num_unique_values\": 2257,\n        \"samples\": [\n          \"\\u0632\\u0646 \\u0628\\u062f\\u0648\\u0646 \\u062a\\u0631\\u0633 \\u062f\\u0631 \\u0622\\u0628 \\u0627\\u0633\\u06a9\\u06cc \\u0645\\u06cc\\u200c\\u06a9\\u0646\\u062f\",\n          \"\\u0637\\u0648\\u0637\\u06cc \\u0645\\u0634\\u063a\\u0648\\u0644 \\u0635\\u062d\\u0628\\u062a \\u0628\\u0627 \\u0645\\u06cc\\u06a9\\u0631\\u0648\\u0641\\u0648\\u0646 \\u0627\\u0633\\u062a\",\n          \"\\u0627\\u06cc\\u0646 \\u0645\\u0631\\u062f \\u0644\\u0628\\u0627\\u0633 \\u067e\\u0634\\u0645\\u06cc \\u0628\\u0647 \\u062a\\u0646 \\u06a9\\u0631\\u062f\\u0647 \\u0648 \\u0631\\u0648\\u06cc \\u062a\\u062e\\u062a\\u0647 \\u0645\\u0648\\u062c \\u0633\\u0648\\u0627\\u0631\\u06cc \\u0633\\u0648\\u0627\\u0631 \\u0634\\u062f\\u0647 \\u0627\\u0633\\u062a\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"Subject1\",\n      \"properties\": {\n        \"dtype\": \"category\",\n        \"num_unique_values\": 427,\n        \"samples\": [\n          \"\\u062e\\u0627\\u0646\\u0645\\u06cc,\\u062a\\u062e\\u062a\",\n          \"\\u0645\\u0631\\u062f\\u06cc,\\u0686\\u0647\\u0631\\u0647\\u200c\\u0627\\u06cc\",\n          \"\\u0632\\u0646,\\u0622\\u0646\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"Sentence2\",\n      \"properties\": {\n        \"dtype\": \"string\",\n        \"num_unique_values\": 2232,\n        \"samples\": [\n          \"\\u06af\\u0631\\u0648\\u0647\\u06cc \\u0627\\u0632 \\u0628\\u0627\\u0632\\u06cc\\u06a9\\u0646\\u0627\\u0646 \\u0641\\u0648\\u062a\\u0628\\u0627\\u0644 \\u062f\\u0631 \\u0632\\u0645\\u06cc\\u0646 \\u062f\\u0631 \\u062d\\u0627\\u0644 \\u062f\\u0648\\u06cc\\u062f\\u0646 \\u0647\\u0633\\u062a\\u0646\\u062f\",\n          \"\\u06af\\u0631\\u0648\\u0647 \\u062f\\u062e\\u062a\\u0631\\u0627\\u0646 \\u0622\\u0633\\u06cc\\u0627\\u06cc\\u06cc \\u06a9\\u0646\\u0627\\u0631 \\u0647\\u0645 \\u0627\\u06cc\\u0633\\u062a\\u0627\\u062f\\u0647\\u200c\\u0627\\u0646\\u062f\",\n          \"\\u067e\\u0633\\u0631 \\u062f\\u0631 \\u06af\\u0644 \\u0628\\u0627\\u0632\\u06cc \\u0646\\u0645\\u06cc\\u200c\\u06a9\\u0646\\u062f\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"Subject2\",\n      \"properties\": {\n        \"dtype\": \"category\",\n        \"num_unique_values\": 423,\n        \"samples\": [\n          \"\\u062e\\u0648\\u0627\\u0646\\u0646\\u062f\\u0647\",\n          \"\\u0645\\u0648\\u062c\",\n          \"\\u0628\\u0628\\u0631\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    }\n  ]\n}"
            }
          },
          "metadata": {},
          "execution_count": 20
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "sub1 = np.array(dataset['Subject1'])\n",
        "sub2 = np.array(dataset['Subject2'])\n",
        "score = np.array(dataset['score'])"
      ],
      "metadata": {
        "id": "Z93xXf_WKpRn"
      },
      "execution_count": 21,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# NN for subject classification\n",
        "with tf.device('/GPU:0'):\n",
        "  scnn = SubjectClassificationNN(X_train_sc_OH, y_train_sc)\n",
        "  scnn.train()\n",
        "  scnn.getModelResult()"
      ],
      "metadata": {
        "id": "oDmEwGAK68UR"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Subject Classification for samples\n",
        "\n",
        "subClass = []\n",
        "\n",
        "with tf.device('/GPU:0'):\n",
        "  for i in range(len(sub1)):\n",
        "    sub1_classes = scnn.predictClass(sub1[i].split(','))\n",
        "    sub2_classes = scnn.predictClass(sub2[i].split(','))\n",
        "\n",
        "    commonClasses = np.intersect1d(sub1_classes, sub2_classes)\n",
        "    if commonClasses.size:\n",
        "      subClass.append(commonClasses[0]) # append common class number between subjects\n",
        "    else:\n",
        "      subClass.append(6)  # append NA class(6) for non-uniform subjects\n",
        "\n",
        "subClass = np.array(subClass, dtype='int32')"
      ],
      "metadata": {
        "id": "bByv5i7rtHlt"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Score Classification\n",
        "print('class 1: ', np.count_nonzero(score < 2.6))\n",
        "print('class 2: ', np.count_nonzero(score < 3.4) - np.count_nonzero(score < 2.6))\n",
        "print('class 3: ', np.count_nonzero(score < 4) - np.count_nonzero(score < 3.4))\n",
        "print('class 4: ', np.count_nonzero(score < 4.5) - np.count_nonzero(score < 4))\n",
        "print('class 5: ', np.count_nonzero(score <= 5) - np.count_nonzero(score < 4.5))\n",
        "\n",
        "scoreClass = []\n",
        "\n",
        "for x in score:\n",
        "  if x < 2.6:\n",
        "    scoreClass.append(1)\n",
        "  elif x < 3.4:\n",
        "    scoreClass.append(2)\n",
        "  elif x < 4:\n",
        "    scoreClass.append(3)\n",
        "  elif x < 4.5:\n",
        "    scoreClass.append(4)\n",
        "  elif x <= 5:\n",
        "    scoreClass.append(5)\n",
        "\n",
        "scoreClass = np.array(scoreClass)\n",
        "\n",
        "print('subject classes shape: ', subClass.shape)\n",
        "print('score classes shape: ', scoreClass.shape)"
      ],
      "metadata": {
        "id": "QhHjvARac5cm",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "794008fe-19b7-474c-c3de-21287c8fe357"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "class 1:  434\n",
            "class 2:  664\n",
            "class 3:  787\n",
            "class 4:  550\n",
            "class 5:  662\n",
            "score classes shape:  3097\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "dataset = pd.read_excel('/content/datasetwithclasses.xlsx')"
      ],
      "metadata": {
        "id": "7xtDO0rW7MIt"
      },
      "execution_count": 23,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# save subject classes into dataframe and save dataframe\n",
        "dataset['subject classes'] = subClass\n",
        "dataset['score classes'] = scoreClass\n",
        "\n",
        "dataset.to_excel(\"datasetwithclasses.xlsx\")\n",
        "dataset.head()"
      ],
      "metadata": {
        "id": "f1JHL0oz7n2S",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "outputId": "55884039-1028-46af-b2d0-849ef954b9cc"
      },
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   Unnamed: 0  score                                          Sentence1  \\\n",
              "0           0    4.5  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "1           1    3.2  گروهی از کودکان در خانه مشغول بازی هستند و هیچ...   \n",
              "2           2    4.7  پسران جوان در فضای باز بازی می‌کنند و مرد در ه...   \n",
              "3           3    3.4  بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...   \n",
              "4           4    3.7  پسران جوان در فضای باز بازی می‌کنند و مرد در ه...   \n",
              "\n",
              "        Subject1                                          Sentence2  \\\n",
              "0  گروهی,پیرمردی  گروهی از پسران در حیاط بازی می‌کنند و مردی در ...   \n",
              "1     گروهی,مردی  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "2      پسران,مرد  بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...   \n",
              "3         بچه‌ها  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "4      پسران,مرد  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "\n",
              "           Subject2  subject classes  score classes  \n",
              "0  گروهی,مردی,زمینه                1              5  \n",
              "1     گروهی,پیرمردی                1              2  \n",
              "2            بچه‌ها                6              5  \n",
              "3     گروهی,پیرمردی                6              3  \n",
              "4     گروهی,پیرمردی                1              3  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-a282a15d-c9bf-46c2-9f9d-ed9488da5be3\" class=\"colab-df-container\">\n",
              "    <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Unnamed: 0</th>\n",
              "      <th>score</th>\n",
              "      <th>Sentence1</th>\n",
              "      <th>Subject1</th>\n",
              "      <th>Sentence2</th>\n",
              "      <th>Subject2</th>\n",
              "      <th>subject classes</th>\n",
              "      <th>score classes</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>4.5</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "      <td>گروهی از پسران در حیاط بازی می‌کنند و مردی در ...</td>\n",
              "      <td>گروهی,مردی,زمینه</td>\n",
              "      <td>1</td>\n",
              "      <td>5</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>3.2</td>\n",
              "      <td>گروهی از کودکان در خانه مشغول بازی هستند و هیچ...</td>\n",
              "      <td>گروهی,مردی</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "      <td>1</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2</td>\n",
              "      <td>4.7</td>\n",
              "      <td>پسران جوان در فضای باز بازی می‌کنند و مرد در ه...</td>\n",
              "      <td>پسران,مرد</td>\n",
              "      <td>بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...</td>\n",
              "      <td>بچه‌ها</td>\n",
              "      <td>6</td>\n",
              "      <td>5</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3</td>\n",
              "      <td>3.4</td>\n",
              "      <td>بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...</td>\n",
              "      <td>بچه‌ها</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "      <td>6</td>\n",
              "      <td>3</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4</td>\n",
              "      <td>3.7</td>\n",
              "      <td>پسران جوان در فضای باز بازی می‌کنند و مرد در ه...</td>\n",
              "      <td>پسران,مرد</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "      <td>1</td>\n",
              "      <td>3</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "    <div class=\"colab-df-buttons\">\n",
              "\n",
              "  <div class=\"colab-df-container\">\n",
              "    <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-a282a15d-c9bf-46c2-9f9d-ed9488da5be3')\"\n",
              "            title=\"Convert this dataframe to an interactive table.\"\n",
              "            style=\"display:none;\">\n",
              "\n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\" viewBox=\"0 -960 960 960\">\n",
              "    <path d=\"M120-120v-720h720v720H120Zm60-500h600v-160H180v160Zm220 220h160v-160H400v160Zm0 220h160v-160H400v160ZM180-400h160v-160H180v160Zm440 0h160v-160H620v160ZM180-180h160v-160H180v160Zm440 0h160v-160H620v160Z\"/>\n",
              "  </svg>\n",
              "    </button>\n",
              "\n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    .colab-df-buttons div {\n",
              "      margin-bottom: 4px;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "    <script>\n",
              "      const buttonEl =\n",
              "        document.querySelector('#df-a282a15d-c9bf-46c2-9f9d-ed9488da5be3 button.colab-df-convert');\n",
              "      buttonEl.style.display =\n",
              "        google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "      async function convertToInteractive(key) {\n",
              "        const element = document.querySelector('#df-a282a15d-c9bf-46c2-9f9d-ed9488da5be3');\n",
              "        const dataTable =\n",
              "          await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                    [key], {});\n",
              "        if (!dataTable) return;\n",
              "\n",
              "        const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "          '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "          + ' to learn more about interactive tables.';\n",
              "        element.innerHTML = '';\n",
              "        dataTable['output_type'] = 'display_data';\n",
              "        await google.colab.output.renderOutput(dataTable, element);\n",
              "        const docLink = document.createElement('div');\n",
              "        docLink.innerHTML = docLinkHtml;\n",
              "        element.appendChild(docLink);\n",
              "      }\n",
              "    </script>\n",
              "  </div>\n",
              "\n",
              "\n",
              "<div id=\"df-492a003f-36ec-4d53-8384-27a8f362e025\">\n",
              "  <button class=\"colab-df-quickchart\" onclick=\"quickchart('df-492a003f-36ec-4d53-8384-27a8f362e025')\"\n",
              "            title=\"Suggest charts\"\n",
              "            style=\"display:none;\">\n",
              "\n",
              "<svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "     width=\"24px\">\n",
              "    <g>\n",
              "        <path d=\"M19 3H5c-1.1 0-2 .9-2 2v14c0 1.1.9 2 2 2h14c1.1 0 2-.9 2-2V5c0-1.1-.9-2-2-2zM9 17H7v-7h2v7zm4 0h-2V7h2v10zm4 0h-2v-4h2v4z\"/>\n",
              "    </g>\n",
              "</svg>\n",
              "  </button>\n",
              "\n",
              "<style>\n",
              "  .colab-df-quickchart {\n",
              "      --bg-color: #E8F0FE;\n",
              "      --fill-color: #1967D2;\n",
              "      --hover-bg-color: #E2EBFA;\n",
              "      --hover-fill-color: #174EA6;\n",
              "      --disabled-fill-color: #AAA;\n",
              "      --disabled-bg-color: #DDD;\n",
              "  }\n",
              "\n",
              "  [theme=dark] .colab-df-quickchart {\n",
              "      --bg-color: #3B4455;\n",
              "      --fill-color: #D2E3FC;\n",
              "      --hover-bg-color: #434B5C;\n",
              "      --hover-fill-color: #FFFFFF;\n",
              "      --disabled-bg-color: #3B4455;\n",
              "      --disabled-fill-color: #666;\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart {\n",
              "    background-color: var(--bg-color);\n",
              "    border: none;\n",
              "    border-radius: 50%;\n",
              "    cursor: pointer;\n",
              "    display: none;\n",
              "    fill: var(--fill-color);\n",
              "    height: 32px;\n",
              "    padding: 0;\n",
              "    width: 32px;\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart:hover {\n",
              "    background-color: var(--hover-bg-color);\n",
              "    box-shadow: 0 1px 2px rgba(60, 64, 67, 0.3), 0 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "    fill: var(--button-hover-fill-color);\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart-complete:disabled,\n",
              "  .colab-df-quickchart-complete:disabled:hover {\n",
              "    background-color: var(--disabled-bg-color);\n",
              "    fill: var(--disabled-fill-color);\n",
              "    box-shadow: none;\n",
              "  }\n",
              "\n",
              "  .colab-df-spinner {\n",
              "    border: 2px solid var(--fill-color);\n",
              "    border-color: transparent;\n",
              "    border-bottom-color: var(--fill-color);\n",
              "    animation:\n",
              "      spin 1s steps(1) infinite;\n",
              "  }\n",
              "\n",
              "  @keyframes spin {\n",
              "    0% {\n",
              "      border-color: transparent;\n",
              "      border-bottom-color: var(--fill-color);\n",
              "      border-left-color: var(--fill-color);\n",
              "    }\n",
              "    20% {\n",
              "      border-color: transparent;\n",
              "      border-left-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "    }\n",
              "    30% {\n",
              "      border-color: transparent;\n",
              "      border-left-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "      border-right-color: var(--fill-color);\n",
              "    }\n",
              "    40% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "    }\n",
              "    60% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "    }\n",
              "    80% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "      border-bottom-color: var(--fill-color);\n",
              "    }\n",
              "    90% {\n",
              "      border-color: transparent;\n",
              "      border-bottom-color: var(--fill-color);\n",
              "    }\n",
              "  }\n",
              "</style>\n",
              "\n",
              "  <script>\n",
              "    async function quickchart(key) {\n",
              "      const quickchartButtonEl =\n",
              "        document.querySelector('#' + key + ' button');\n",
              "      quickchartButtonEl.disabled = true;  // To prevent multiple clicks.\n",
              "      quickchartButtonEl.classList.add('colab-df-spinner');\n",
              "      try {\n",
              "        const charts = await google.colab.kernel.invokeFunction(\n",
              "            'suggestCharts', [key], {});\n",
              "      } catch (error) {\n",
              "        console.error('Error during call to suggestCharts:', error);\n",
              "      }\n",
              "      quickchartButtonEl.classList.remove('colab-df-spinner');\n",
              "      quickchartButtonEl.classList.add('colab-df-quickchart-complete');\n",
              "    }\n",
              "    (() => {\n",
              "      let quickchartButtonEl =\n",
              "        document.querySelector('#df-492a003f-36ec-4d53-8384-27a8f362e025 button');\n",
              "      quickchartButtonEl.style.display =\n",
              "        google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "    })();\n",
              "  </script>\n",
              "</div>\n",
              "    </div>\n",
              "  </div>\n"
            ],
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "dataframe",
              "variable_name": "dataset",
              "summary": "{\n  \"name\": \"dataset\",\n  \"rows\": 3097,\n  \"fields\": [\n    {\n      \"column\": \"Unnamed: 0\",\n      \"properties\": {\n        \"dtype\": \"number\",\n        \"std\": 894,\n        \"min\": 0,\n        \"max\": 3096,\n        \"num_unique_values\": 3097,\n        \"samples\": [\n          1947,\n          1309,\n          1606\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"score\",\n      \"properties\": {\n        \"dtype\": \"number\",\n        \"std\": 0.991643781134316,\n        \"min\": 1.0,\n        \"max\": 5.0,\n        \"num_unique_values\": 102,\n        \"samples\": [\n          3.3,\n          2.915,\n          3.65\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"Sentence1\",\n      \"properties\": {\n        \"dtype\": \"string\",\n        \"num_unique_values\": 2257,\n        \"samples\": [\n          \"\\u0632\\u0646 \\u0628\\u062f\\u0648\\u0646 \\u062a\\u0631\\u0633 \\u062f\\u0631 \\u0622\\u0628 \\u0627\\u0633\\u06a9\\u06cc \\u0645\\u06cc\\u200c\\u06a9\\u0646\\u062f\",\n          \"\\u0637\\u0648\\u0637\\u06cc \\u0645\\u0634\\u063a\\u0648\\u0644 \\u0635\\u062d\\u0628\\u062a \\u0628\\u0627 \\u0645\\u06cc\\u06a9\\u0631\\u0648\\u0641\\u0648\\u0646 \\u0627\\u0633\\u062a\",\n          \"\\u0627\\u06cc\\u0646 \\u0645\\u0631\\u062f \\u0644\\u0628\\u0627\\u0633 \\u067e\\u0634\\u0645\\u06cc \\u0628\\u0647 \\u062a\\u0646 \\u06a9\\u0631\\u062f\\u0647 \\u0648 \\u0631\\u0648\\u06cc \\u062a\\u062e\\u062a\\u0647 \\u0645\\u0648\\u062c \\u0633\\u0648\\u0627\\u0631\\u06cc \\u0633\\u0648\\u0627\\u0631 \\u0634\\u062f\\u0647 \\u0627\\u0633\\u062a\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"Subject1\",\n      \"properties\": {\n        \"dtype\": \"category\",\n        \"num_unique_values\": 426,\n        \"samples\": [\n          \"\\u062e\\u0627\\u0646\\u0645\\u06cc,\\u062a\\u062e\\u062a\",\n          \"\\u0628\\u0632\\u0631\\u06af\\u0633\\u0627\\u0644\",\n          \"\\u0632\\u0646,\\u0622\\u0646\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"Sentence2\",\n      \"properties\": {\n        \"dtype\": \"string\",\n        \"num_unique_values\": 2232,\n        \"samples\": [\n          \"\\u06af\\u0631\\u0648\\u0647\\u06cc \\u0627\\u0632 \\u0628\\u0627\\u0632\\u06cc\\u06a9\\u0646\\u0627\\u0646 \\u0641\\u0648\\u062a\\u0628\\u0627\\u0644 \\u062f\\u0631 \\u0632\\u0645\\u06cc\\u0646 \\u062f\\u0631 \\u062d\\u0627\\u0644 \\u062f\\u0648\\u06cc\\u062f\\u0646 \\u0647\\u0633\\u062a\\u0646\\u062f\",\n          \"\\u06af\\u0631\\u0648\\u0647 \\u062f\\u062e\\u062a\\u0631\\u0627\\u0646 \\u0622\\u0633\\u06cc\\u0627\\u06cc\\u06cc \\u06a9\\u0646\\u0627\\u0631 \\u0647\\u0645 \\u0627\\u06cc\\u0633\\u062a\\u0627\\u062f\\u0647\\u200c\\u0627\\u0646\\u062f\",\n          \"\\u067e\\u0633\\u0631 \\u062f\\u0631 \\u06af\\u0644 \\u0628\\u0627\\u0632\\u06cc \\u0646\\u0645\\u06cc\\u200c\\u06a9\\u0646\\u062f\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"Subject2\",\n      \"properties\": {\n        \"dtype\": \"category\",\n        \"num_unique_values\": 422,\n        \"samples\": [\n          \"\\u062d\\u06cc\\u0648\\u0627\\u0646\",\n          \"\\u0641\\u0648\\u062a\\u0628\\u0627\\u0644\\u06cc\\u0633\\u062a\",\n          \"\\u062a\\u062e\\u0645\"\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"subject classes\",\n      \"properties\": {\n        \"dtype\": \"number\",\n        \"std\": 1,\n        \"min\": 1,\n        \"max\": 6,\n        \"num_unique_values\": 6,\n        \"samples\": [\n          1,\n          6,\n          3\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    },\n    {\n      \"column\": \"score classes\",\n      \"properties\": {\n        \"dtype\": \"number\",\n        \"std\": 1,\n        \"min\": 1,\n        \"max\": 5,\n        \"num_unique_values\": 5,\n        \"samples\": [\n          2,\n          1,\n          3\n        ],\n        \"semantic_type\": \"\",\n        \"description\": \"\"\n      }\n    }\n  ]\n}"
            }
          },
          "metadata": {},
          "execution_count": 25
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Preprocessing Generated Classification Dataset**"
      ],
      "metadata": {
        "id": "JtId7nYzVJGC"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "### **Dataset**"
      ],
      "metadata": {
        "id": "nLMSBhbc8L63"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "dataset = pd.read_excel('/content/datasetwithclasses.xlsx')"
      ],
      "metadata": {
        "id": "nEYWLJLOVYcv"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "dataset.head()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "id": "cWBbwmRBta7E",
        "outputId": "6f67d4fd-4a44-40e0-ad05-1e82701ec0c5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   Unnamed: 0  score                                          Sentence1  \\\n",
              "0           0    4.5  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "1           1    3.2  گروهی از کودکان در خانه مشغول بازی هستند و هیچ...   \n",
              "2           2    4.7  پسران جوان در فضای باز بازی می‌کنند و مرد در ه...   \n",
              "3           3    3.4  بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...   \n",
              "4           4    3.7  پسران جوان در فضای باز بازی می‌کنند و مرد در ه...   \n",
              "\n",
              "        Subject1                                          Sentence2  \\\n",
              "0  گروهی,پیرمردی  گروهی از پسران در حیاط بازی می‌کنند و مردی در ...   \n",
              "1     گروهی,مردی  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "2      پسران,مرد  بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...   \n",
              "3         بچه‌ها  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "4      پسران,مرد  گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...   \n",
              "\n",
              "           Subject2  subject classes  score classes  \n",
              "0  گروهی,مردی,زمینه                1              5  \n",
              "1     گروهی,پیرمردی                1              2  \n",
              "2            بچه‌ها                6              5  \n",
              "3     گروهی,پیرمردی                6              3  \n",
              "4     گروهی,پیرمردی                1              3  "
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-47ebb41d-353d-4eca-88a2-8572129bcc8a\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Unnamed: 0</th>\n",
              "      <th>score</th>\n",
              "      <th>Sentence1</th>\n",
              "      <th>Subject1</th>\n",
              "      <th>Sentence2</th>\n",
              "      <th>Subject2</th>\n",
              "      <th>subject classes</th>\n",
              "      <th>score classes</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>4.5</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "      <td>گروهی از پسران در حیاط بازی می‌کنند و مردی در ...</td>\n",
              "      <td>گروهی,مردی,زمینه</td>\n",
              "      <td>1</td>\n",
              "      <td>5</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>3.2</td>\n",
              "      <td>گروهی از کودکان در خانه مشغول بازی هستند و هیچ...</td>\n",
              "      <td>گروهی,مردی</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "      <td>1</td>\n",
              "      <td>2</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>2</td>\n",
              "      <td>4.7</td>\n",
              "      <td>پسران جوان در فضای باز بازی می‌کنند و مرد در ه...</td>\n",
              "      <td>پسران,مرد</td>\n",
              "      <td>بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...</td>\n",
              "      <td>بچه‌ها</td>\n",
              "      <td>6</td>\n",
              "      <td>5</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3</td>\n",
              "      <td>3.4</td>\n",
              "      <td>بچه‌ها در بیرون از خانه و در کنار یک مرد با لب...</td>\n",
              "      <td>بچه‌ها</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "      <td>6</td>\n",
              "      <td>3</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4</td>\n",
              "      <td>3.7</td>\n",
              "      <td>پسران جوان در فضای باز بازی می‌کنند و مرد در ه...</td>\n",
              "      <td>پسران,مرد</td>\n",
              "      <td>گروهی از بچه‌ها در حیاط بازی می‌کنند و پیرمردی...</td>\n",
              "      <td>گروهی,پیرمردی</td>\n",
              "      <td>1</td>\n",
              "      <td>3</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-47ebb41d-353d-4eca-88a2-8572129bcc8a')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-47ebb41d-353d-4eca-88a2-8572129bcc8a button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-47ebb41d-353d-4eca-88a2-8572129bcc8a');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 21
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "subClasses = np.array(dataset['subject classes'])\n",
        "scoreClasses = np.array(dataset['score classes'])\n",
        "\n",
        "print(\"Subject Class Men/Boy: \", np.count_nonzero(subClasses == 1))\n",
        "print(\"Subject Class Women/Girl: \", np.count_nonzero(subClasses == 2))\n",
        "print(\"Subject Class Child: \", np.count_nonzero(subClasses == 3))\n",
        "print(\"Subject Class Animal: \", np.count_nonzero(subClasses == 4))\n",
        "print(\"Subject Class Other: \", np.count_nonzero(subClasses == 5))\n",
        "print(\"Subject Class NA: \", np.count_nonzero(subClasses == 6))\n",
        "scoreClasses.shape\n",
        "#dataset.style"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "RZ-HpPwYVdJo",
        "outputId": "4cc29f9d-19c8-4f84-9ebc-0b57b50b08ab"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Subject Class Men/Boy:  669\n",
            "Subject Class Women/Girl:  390\n",
            "Subject Class Child:  99\n",
            "Subject Class Animal:  424\n",
            "Subject Class Other:  568\n",
            "Subject Class NA:  947\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(3097,)"
            ]
          },
          "metadata": {},
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "### **Create Y_true**"
      ],
      "metadata": {
        "id": "43UxIhy88Xfm"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "subClasses_OH = to_categorical(subClasses, num_classes=7)[:, 1:]\n",
        "scoreClasses_OH = to_categorical(scoreClasses, num_classes=6)[:, 1:]\n",
        "\n",
        "print(scoreClasses_OH.shape)\n",
        "print(subClasses_OH.shape)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5Q9E1JhHXlgS",
        "outputId": "24a6808d-a900-4c27-c805-4a4069a05770"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(3097, 5)\n",
            "(3097, 6)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Peparing Y true\n",
        "y = np.concatenate((subClasses_OH, scoreClasses_OH), axis=1)\n",
        "y.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "eaxag_QujZcC",
        "outputId": "d4159b81-e84c-4159-c091-8c512865c660"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(3097, 11)"
            ]
          },
          "metadata": {},
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "### **Create X**"
      ],
      "metadata": {
        "id": "51aHU20h8kdg"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# Preparing X\n",
        "sentence1 = np.array(dataset['Sentence1'])\n",
        "sentence2 = np.array(dataset['Sentence2'])"
      ],
      "metadata": {
        "id": "XdwN-XVplHRw"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Keras Tokenizer\n",
        "filters = '!\"#$%&()*+,-./:;<=>?@[\\\\]^_`{|}~\\t\\n'\n",
        "num_words = 1000\n",
        "\n",
        "sentenceTokenizer = Tokenizer(num_words=num_words, filters=filters, split=' ', oov_token='UNK')\n",
        "\n",
        "# train on sentences\n",
        "sentenceTokenizer.fit_on_texts(sentence1)\n",
        "# sentenceTokenizer.fit_on_texts(sentence2)\n",
        "\n",
        "# encode sentences\n",
        "sentence1_encoded = sentenceTokenizer.texts_to_sequences(sentence1)\n",
        "sentence2_encoded = sentenceTokenizer.texts_to_sequences(sentence2)"
      ],
      "metadata": {
        "id": "SVxAPcRXuBr1"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# max number of words in a sentence\n",
        "max_words_num = max(max([len(x) for x in sentence1_encoded]), max([len(x) for x in sentence2_encoded]))\n",
        "max_words_num"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "uSh14B6w1q84",
        "outputId": "067e7e74-dbe5-42f5-fc60-212dfc4a301d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "26"
            ]
          },
          "metadata": {},
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# normalize sentence lenght with padding\n",
        "maxlen = 20\n",
        "sentence1_pad = pad_sequences(sentence1_encoded, maxlen=maxlen, padding='post')\n",
        "sentence2_pad = pad_sequences(sentence2_encoded, maxlen=maxlen, padding='post')\n",
        "sentence1_pad.shape, sentence2_pad.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Ig1ScKu65PCU",
        "outputId": "bebd8d33-0fda-4485-c9fa-e7006e27ba24"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "((3097, 20), (3097, 20))"
            ]
          },
          "metadata": {},
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# change input to One Hot\n",
        "sentence1_OH = to_categorical(sentence1_pad, num_classes=num_words)\n",
        "sentence2_OH = to_categorical(sentence2_pad, num_classes=num_words)\n",
        "sentence1_OH.shape, sentence2_OH.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "YSBxYNck6DBx",
        "outputId": "d8a8d89a-d6f8-49eb-fe09-0a7ffcd203fa"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "((3097, 20, 1000), (3097, 20, 1000))"
            ]
          },
          "metadata": {},
          "execution_count": 17
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# concatenate two sentences\n",
        "sentences_OH = np.concatenate((sentence1_OH, sentence2_OH), axis=1)\n",
        "sentences_OH.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "cXb1DPbM6rHJ",
        "outputId": "130914d0-c919-46d6-eee9-f76c666dfcad"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(3097, 40, 1000)"
            ]
          },
          "metadata": {},
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "### **Split data**"
      ],
      "metadata": {
        "id": "PBQmryY-9QMc"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "X_train_1, X_test_1, X_train_2, X_test_2, y_train, y_test = train_test_split(sentence1_OH, sentence2_OH, y, test_size=0.2)\n",
        "X_train_1.shape, X_train_2.shape, y_train.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "PdwfHR2i86Vq",
        "outputId": "4a4fadc2-8d8d-4303-9544-0075fe9a62b9"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "((2477, 20, 1000), (2477, 20, 1000), (2477, 11))"
            ]
          },
          "metadata": {},
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "X_train_list = [X_train_1, X_train_2]\n",
        "X_test_list = [X_test_1, X_test_2]"
      ],
      "metadata": {
        "id": "-oqVjgrpTR9G"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XCpPYCUde4AK"
      },
      "source": [
        "# **Implementation**\n",
        "\n",
        "---\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "bpqKwAvEKEVc"
      },
      "outputs": [],
      "source": [
        "class RNN:\n",
        "  def __init__(self, X_train, y_train, model_num):\n",
        "    self.X_train = X_train\n",
        "    self.y_train = y_train\n",
        "\n",
        "    input1 = Input(shape=np.array(X_train[0]).shape[1:])\n",
        "    input2 = Input(shape=np.array(X_train[1]).shape[1:])\n",
        "\n",
        "    if model_num == 1:\n",
        "      m = self.LSTM_model_1(input1, input2)\n",
        "    elif model_num == 2:\n",
        "      m = self.LSTM_model_2(input1, input2)\n",
        "    elif model_num == 3:\n",
        "      m = self.LSTM_model_3(input1, input2)\n",
        "    elif model_num == 4:\n",
        "      m = self.GRU_model_1(input1, input2)\n",
        "    elif model_num == 5:\n",
        "      m = self.GRU_model_2(input1, input2)\n",
        "    elif model_num == 6:\n",
        "      m = self.SimpleRNN_model_1(input1, input2)\n",
        "    elif model_num == 7:\n",
        "      m = self.SimpleRNN_model_2(input1, input2)\n",
        "    elif model_num == 8:\n",
        "      m = self.LSTM_model_4(input1, input2)\n",
        "\n",
        "    output = Dense(11, activation=\"sigmoid\")(m)\n",
        "\n",
        "    # Compile the Model\n",
        "    self.model = Model(inputs=[input1, input2], outputs=output)\n",
        "    self.model.compile(loss=\"binary_crossentropy\", optimizer=Adam(), metrics=['accuracy'])\n",
        "\n",
        "    # self.model.summary()\n",
        "\n",
        "  def LSTM_model_1(self, input1, input2):\n",
        "    x1 = LSTM(128, return_sequences=True)(input1)\n",
        "    x1 = LSTM(128)(x1)\n",
        "\n",
        "    x2 = LSTM(128, return_sequences=True)(input2)\n",
        "    x2 = LSTM(128)(x2)\n",
        "\n",
        "    c = Concatenate(axis=1)([x1, x2])\n",
        "\n",
        "    d = Dense(64, activation='relu')(c)\n",
        "    d = Dense(32, activation='relu')(d)\n",
        "    d = Dense(16, activation='relu')(d)\n",
        "\n",
        "    return d\n",
        "\n",
        "  def LSTM_model_2(self, input1, input2):\n",
        "    x1 = LSTM(128, return_sequences=True)(input1)\n",
        "    x1 = LSTM(128, return_sequences=True)(x1)\n",
        "    c1 = GlobalMaxPooling1D()(x1)\n",
        "\n",
        "    x2 = LSTM(128, return_sequences=True)(input2)\n",
        "    x2 = LSTM(128, return_sequences=True)(x2)\n",
        "    c2 = GlobalMaxPooling1D()(x2)\n",
        "\n",
        "    c = Concatenate(axis=1)([c1, c2])\n",
        "\n",
        "    d = Dense(64, activation='relu')(c)\n",
        "    d = Dense(32, activation='relu')(d)\n",
        "    d = Dense(16, activation='relu')(d)\n",
        "\n",
        "    return d\n",
        "\n",
        "\n",
        "  def LSTM_model_3(self, input1, input2):\n",
        "    x1 = LSTM(128, return_sequences=True)(input1)\n",
        "    x1 = GRU(128)(x1)\n",
        "\n",
        "    x2 = LSTM(128, return_sequences=True)(input2)\n",
        "    x2 = GRU(128)(x2)\n",
        "\n",
        "    c = Concatenate(axis=1)([x1, x2])\n",
        "\n",
        "    d = Dense(64, activation='relu')(c)\n",
        "    d = Dense(32, activation='relu')(d)\n",
        "    d = Dense(16, activation='relu')(d)\n",
        "\n",
        "    return d\n",
        "\n",
        "  def LSTM_model_4(self, input1, input2):\n",
        "    x1 = LSTM(128, return_sequences=True)(input1)\n",
        "    x1 = SimpleRNN(128, return_sequences=True)(x1)\n",
        "    x1 = SimpleRNN(128, return_sequences=True)(x1)\n",
        "    c1 = GlobalMaxPooling1D()(x1)\n",
        "\n",
        "    x2 = LSTM(128, return_sequences=True)(input2)\n",
        "    x2 = SimpleRNN(128, return_sequences=True)(x2)\n",
        "    x2 = SimpleRNN(128, return_sequences=True)(x2)\n",
        "    c2 = GlobalMaxPooling1D()(x2)\n",
        "\n",
        "    c = Concatenate(axis=1)([c1, c2])\n",
        "\n",
        "    d = Dense(64, activation='relu')(c)\n",
        "    d = Dropout(0.2)(d)\n",
        "    d = Dense(32, activation='relu')(d)\n",
        "    d = Dropout(0.1)(d)\n",
        "    d = Dense(16, activation='relu')(d)\n",
        "\n",
        "    return d\n",
        "\n",
        "  def GRU_model_1(self, input1, input2):\n",
        "    x1 = GRU(128, return_sequences=True)(input1)\n",
        "    x1 = GRU(128)(x1)\n",
        "\n",
        "    x2 = GRU(128, return_sequences=True)(input2)\n",
        "    x2 = GRU(128)(x2)\n",
        "\n",
        "    c = Concatenate(axis=1)([x1, x2])\n",
        "\n",
        "    d = Dense(64, activation='relu')(c)\n",
        "    d = Dense(32, activation='relu')(d)\n",
        "    d = Dense(16, activation='relu')(d)\n",
        "\n",
        "    return d\n",
        "\n",
        "  def GRU_model_2(self, input1, input2):\n",
        "    x1 = GRU(256, return_sequences=True)(input1)\n",
        "    c1 = GlobalMaxPooling1D()(x1)\n",
        "\n",
        "    x2 = GRU(256, return_sequences=True)(input2)\n",
        "    c2 = GlobalMaxPooling1D()(x2)\n",
        "\n",
        "    c = Concatenate(axis=1)([c1, c2])\n",
        "\n",
        "    d = Dense(64, activation='relu')(c)\n",
        "    d = Dense(32, activation='relu')(d)\n",
        "    d = Dense(16, activation='relu')(d)\n",
        "\n",
        "    return d\n",
        "\n",
        "  def SimpleRNN_model_1(self, input1, input2):\n",
        "    x1 = SimpleRNN(128, return_sequences=True)(input1)\n",
        "    x1 = SimpleRNN(128)(x1)\n",
        "\n",
        "    x2 = SimpleRNN(128, return_sequences=True)(input2)\n",
        "    x2 = SimpleRNN(128)(x2)\n",
        "\n",
        "    c = Concatenate(axis=1)([x1, x2])\n",
        "\n",
        "    d = Dense(64, activation='relu')(c)\n",
        "    # d = Dense(64, activation='relu')(d)\n",
        "    d = Dense(16, activation='relu')(d)\n",
        "\n",
        "    return d\n",
        "\n",
        "  def SimpleRNN_model_2(self, input1, input2):\n",
        "    x1 = SimpleRNN(128, return_sequences=True)(input1)\n",
        "    x1 = SimpleRNN(128, return_sequences=True)(x1)\n",
        "    c1 = GlobalMaxPooling1D()(x1)\n",
        "\n",
        "    x2 = SimpleRNN(128, return_sequences=True)(input2)\n",
        "    x2 = SimpleRNN(128, return_sequences=True)(x2)\n",
        "    c2 = GlobalMaxPooling1D()(x2)\n",
        "\n",
        "    c = Concatenate(axis=1)([c1, c2])\n",
        "\n",
        "    d = Dense(64, activation='relu')(c)\n",
        "    d = Dense(32, activation='relu')(d)\n",
        "    d = Dense(16, activation='relu')(d)\n",
        "\n",
        "    return d\n",
        "\n",
        "\n",
        "  def train(self, validation_data):\n",
        "    filepath = 'my_best_model.hdf5'\n",
        "    checkpoint = ModelCheckpoint(filepath=filepath, monitor='val_accuracy', verbose=0, save_best_only=True, mode='max')\n",
        "    self.history = self.model.fit(self.X_train, self.y_train, validation_data=validation_data, epochs=50, batch_size=32, verbose=2, callbacks=[checkpoint]).history\n",
        "    self.model = load_model(filepath)\n",
        "\n",
        "  def predict(self):\n",
        "    self.y_train_pre = self.model.predict(self.X_train)\n",
        "    self.y_test_pre = self.model.predict(self.X_test)\n",
        "\n",
        "  def evaluate(self, X_test, y_test):\n",
        "    self.X_test = X_test\n",
        "    self.y_test = y_test\n",
        "    self.test_result = self.model.evaluate(X_test, y_test, verbose=0)\n",
        "\n",
        "  def outputResult(self):\n",
        "\n",
        "    # best models based on acc or loss in tarin set or test set\n",
        "    trainHistory = list(\n",
        "        map(lambda x, y: [x, y], self.history['accuracy'], self.history['loss']))\n",
        "    testHistory = list(\n",
        "        map(lambda x, y: [x, y], self.history['val_accuracy'], self.history['val_loss']))\n",
        "    print(\n",
        "        f\"\\nbest model based on min training set loss:  acc= {min(trainHistory, key = lambda k: k[1])[0]}  loss= {min(trainHistory, key = lambda k: k[1])[1]}\")\n",
        "    print(\n",
        "        f\"best model based on min test set loss:  acc= {min(testHistory, key = lambda k: k[1])[0]}  loss= {min(testHistory, key = lambda k: k[1])[1]}\")\n",
        "    print(\n",
        "        f\"best model based on max training set accuracy:  acc= {max(trainHistory, key = lambda k: k[0])[0]}  loss= {max(trainHistory, key = lambda k: k[0])[1]}\")\n",
        "    print(\n",
        "        f\"best model based on max test set accuracy:  acc= {max(testHistory, key = lambda k: k[0])[0]}  loss= {max(testHistory, key = lambda k: k[0])[1]}\")\n",
        "\n",
        "    print(\"\\nevaluate dataset with best model based on maximum test set accuracy\")\n",
        "    print(\"evaluate train set= \", self.model.evaluate(\n",
        "        self.X_train, self.y_train, verbose=0))\n",
        "    print(\"evaluate test set= \", self.test_result)\n",
        "\n",
        "    y_train_pred = self.model.predict(self.X_train)\n",
        "    # convert probablities to 0s and 1s\n",
        "    y_train_pred_score = argmaxKeepDimensions(y_train_pred[:, :6])\n",
        "    y_train_pred_subject = argmaxKeepDimensions(y_train_pred[:, 6:])\n",
        "    y_train_pred = np.concatenate((y_train_pred_score, y_train_pred_subject), axis=1)\n",
        "\n",
        "    y_test_pred = self.model.predict(self.X_test)\n",
        "    # convert probablities to 0s and 1s\n",
        "    y_test_pred_score = argmaxKeepDimensions(y_test_pred[:, :6])\n",
        "    y_test_pred_subject = argmaxKeepDimensions(y_test_pred[:, 6:])\n",
        "    y_test_pred = np.concatenate((y_test_pred_score, y_test_pred_subject), axis=1)\n",
        "\n",
        "    # confusion matrix and precision, recall and f1 report\n",
        "    print('\\n', '-'*30, 'metrics for traning set', '-'*30)\n",
        "    print(\"confusion matrix: \\n\", metrics.multilabel_confusion_matrix(self.y_train, y_train_pred))\n",
        "    print(metrics.classification_report(self.y_train,\n",
        "          y_train_pred, digits=3, target_names=['Men/Boy', 'Women/Girl', 'Child', 'Animal', 'Other', 'NA', 'Score1', 'Score2', 'Score3', 'Score4', 'Score5']))\n",
        "\n",
        "    # confusion matrix and precision, recall and f1 report\n",
        "    print('-'*30, 'metrics for test set', '-'*30)\n",
        "    print(\"confusion matrix: \\n\", metrics.multilabel_confusion_matrix(self.y_test, y_test_pred))\n",
        "    print(metrics.classification_report(self.y_test,\n",
        "          y_test_pred, digits=3, target_names=['Men/Boy', 'Women/Girl', 'Child', 'Animal', 'Other', 'NA', 'Score1', 'Score2', 'Score3', 'Score4', 'Score5']))\n",
        "\n",
        "    # self.model.summary()\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **Simple RNN**"
      ],
      "metadata": {
        "id": "uRUJbQgFOtuT"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Model 1**"
      ],
      "metadata": {
        "id": "buodTk7DPAv6"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "with tf.device('/GPU:0'):\n",
        "  model6 = RNN(X_train_list, y_train, model_num=6)\n",
        "  model6.train((X_test_list, y_test))\n",
        "  model6.evaluate(X_test_list, y_test)\n",
        "  model6.outputResult()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "V6JwZUHzPAAo",
        "outputId": "c8a8bba0-cda0-4ae3-e866-b597f4fde5f2"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "78/78 - 8s - loss: 0.5841 - accuracy: 0.2406 - val_loss: 0.4446 - val_accuracy: 0.4210 - 8s/epoch - 97ms/step\n",
            "Epoch 2/50\n",
            "78/78 - 5s - loss: 0.3835 - accuracy: 0.6258 - val_loss: 0.3554 - val_accuracy: 0.7516 - 5s/epoch - 61ms/step\n",
            "Epoch 3/50\n",
            "78/78 - 5s - loss: 0.3189 - accuracy: 0.8119 - val_loss: 0.3318 - val_accuracy: 0.7935 - 5s/epoch - 63ms/step\n",
            "Epoch 4/50\n",
            "78/78 - 5s - loss: 0.2793 - accuracy: 0.8708 - val_loss: 0.3359 - val_accuracy: 0.7871 - 5s/epoch - 61ms/step\n",
            "Epoch 5/50\n",
            "78/78 - 5s - loss: 0.2572 - accuracy: 0.9027 - val_loss: 0.3296 - val_accuracy: 0.8113 - 5s/epoch - 63ms/step\n",
            "Epoch 6/50\n",
            "78/78 - 5s - loss: 0.2362 - accuracy: 0.9241 - val_loss: 0.3319 - val_accuracy: 0.8145 - 5s/epoch - 64ms/step\n",
            "Epoch 7/50\n",
            "78/78 - 5s - loss: 0.2225 - accuracy: 0.9273 - val_loss: 0.3393 - val_accuracy: 0.8113 - 5s/epoch - 63ms/step\n",
            "Epoch 8/50\n",
            "78/78 - 5s - loss: 0.2011 - accuracy: 0.9475 - val_loss: 0.3637 - val_accuracy: 0.7694 - 5s/epoch - 62ms/step\n",
            "Epoch 9/50\n",
            "78/78 - 5s - loss: 0.1836 - accuracy: 0.9491 - val_loss: 0.3819 - val_accuracy: 0.7581 - 5s/epoch - 61ms/step\n",
            "Epoch 10/50\n",
            "78/78 - 5s - loss: 0.1700 - accuracy: 0.9459 - val_loss: 0.3988 - val_accuracy: 0.7919 - 5s/epoch - 60ms/step\n",
            "Epoch 11/50\n",
            "78/78 - 5s - loss: 0.1587 - accuracy: 0.9423 - val_loss: 0.4325 - val_accuracy: 0.7484 - 5s/epoch - 70ms/step\n",
            "Epoch 12/50\n",
            "78/78 - 5s - loss: 0.1525 - accuracy: 0.9241 - val_loss: 0.4445 - val_accuracy: 0.7726 - 5s/epoch - 61ms/step\n",
            "Epoch 13/50\n",
            "78/78 - 5s - loss: 0.1430 - accuracy: 0.9269 - val_loss: 0.4597 - val_accuracy: 0.7419 - 5s/epoch - 61ms/step\n",
            "Epoch 14/50\n",
            "78/78 - 5s - loss: 0.1218 - accuracy: 0.9289 - val_loss: 0.5074 - val_accuracy: 0.6952 - 5s/epoch - 60ms/step\n",
            "Epoch 15/50\n",
            "78/78 - 5s - loss: 0.1202 - accuracy: 0.9229 - val_loss: 0.4994 - val_accuracy: 0.7387 - 5s/epoch - 61ms/step\n",
            "Epoch 16/50\n",
            "78/78 - 5s - loss: 0.1067 - accuracy: 0.9193 - val_loss: 0.5221 - val_accuracy: 0.7452 - 5s/epoch - 61ms/step\n",
            "Epoch 17/50\n",
            "78/78 - 5s - loss: 0.1026 - accuracy: 0.9047 - val_loss: 0.5461 - val_accuracy: 0.7242 - 5s/epoch - 60ms/step\n",
            "Epoch 18/50\n",
            "78/78 - 5s - loss: 0.0887 - accuracy: 0.9071 - val_loss: 0.5854 - val_accuracy: 0.7048 - 5s/epoch - 61ms/step\n",
            "Epoch 19/50\n",
            "78/78 - 5s - loss: 0.0755 - accuracy: 0.9007 - val_loss: 0.5835 - val_accuracy: 0.7403 - 5s/epoch - 61ms/step\n",
            "Epoch 20/50\n",
            "78/78 - 5s - loss: 0.0687 - accuracy: 0.8922 - val_loss: 0.6404 - val_accuracy: 0.6919 - 5s/epoch - 61ms/step\n",
            "Epoch 21/50\n",
            "78/78 - 5s - loss: 0.0654 - accuracy: 0.8898 - val_loss: 0.6495 - val_accuracy: 0.7484 - 5s/epoch - 61ms/step\n",
            "Epoch 22/50\n",
            "78/78 - 5s - loss: 0.0748 - accuracy: 0.8793 - val_loss: 0.6615 - val_accuracy: 0.6774 - 5s/epoch - 61ms/step\n",
            "Epoch 23/50\n",
            "78/78 - 5s - loss: 0.0569 - accuracy: 0.8676 - val_loss: 0.6977 - val_accuracy: 0.7419 - 5s/epoch - 60ms/step\n",
            "Epoch 24/50\n",
            "78/78 - 5s - loss: 0.0520 - accuracy: 0.8648 - val_loss: 0.7331 - val_accuracy: 0.6710 - 5s/epoch - 62ms/step\n",
            "Epoch 25/50\n",
            "78/78 - 5s - loss: 0.0438 - accuracy: 0.8668 - val_loss: 0.7897 - val_accuracy: 0.6516 - 5s/epoch - 61ms/step\n",
            "Epoch 26/50\n",
            "78/78 - 5s - loss: 0.0389 - accuracy: 0.8446 - val_loss: 0.8015 - val_accuracy: 0.6548 - 5s/epoch - 59ms/step\n",
            "Epoch 27/50\n",
            "78/78 - 5s - loss: 0.0332 - accuracy: 0.8426 - val_loss: 0.8379 - val_accuracy: 0.6613 - 5s/epoch - 61ms/step\n",
            "Epoch 28/50\n",
            "78/78 - 5s - loss: 0.0332 - accuracy: 0.8595 - val_loss: 0.8441 - val_accuracy: 0.6903 - 5s/epoch - 60ms/step\n",
            "Epoch 29/50\n",
            "78/78 - 5s - loss: 0.0289 - accuracy: 0.8498 - val_loss: 0.8355 - val_accuracy: 0.7097 - 5s/epoch - 61ms/step\n",
            "Epoch 30/50\n",
            "78/78 - 5s - loss: 0.0311 - accuracy: 0.8272 - val_loss: 0.8507 - val_accuracy: 0.6645 - 5s/epoch - 60ms/step\n",
            "Epoch 31/50\n",
            "78/78 - 5s - loss: 0.0416 - accuracy: 0.8187 - val_loss: 0.8543 - val_accuracy: 0.7081 - 5s/epoch - 60ms/step\n",
            "Epoch 32/50\n",
            "78/78 - 5s - loss: 0.0616 - accuracy: 0.8155 - val_loss: 0.8156 - val_accuracy: 0.6790 - 5s/epoch - 60ms/step\n",
            "Epoch 33/50\n",
            "78/78 - 5s - loss: 0.0623 - accuracy: 0.8304 - val_loss: 0.8020 - val_accuracy: 0.6839 - 5s/epoch - 61ms/step\n",
            "Epoch 34/50\n",
            "78/78 - 5s - loss: 0.0582 - accuracy: 0.8199 - val_loss: 0.8613 - val_accuracy: 0.6839 - 5s/epoch - 61ms/step\n",
            "Epoch 35/50\n",
            "78/78 - 5s - loss: 0.0494 - accuracy: 0.8264 - val_loss: 0.8770 - val_accuracy: 0.6710 - 5s/epoch - 64ms/step\n",
            "Epoch 36/50\n",
            "78/78 - 5s - loss: 0.0295 - accuracy: 0.8385 - val_loss: 0.8407 - val_accuracy: 0.7113 - 5s/epoch - 61ms/step\n",
            "Epoch 37/50\n",
            "78/78 - 5s - loss: 0.0205 - accuracy: 0.8304 - val_loss: 0.8900 - val_accuracy: 0.7097 - 5s/epoch - 60ms/step\n",
            "Epoch 38/50\n",
            "78/78 - 5s - loss: 0.0147 - accuracy: 0.8208 - val_loss: 0.9484 - val_accuracy: 0.6516 - 5s/epoch - 63ms/step\n",
            "Epoch 39/50\n",
            "78/78 - 5s - loss: 0.0116 - accuracy: 0.8296 - val_loss: 0.9409 - val_accuracy: 0.6774 - 5s/epoch - 62ms/step\n",
            "Epoch 40/50\n",
            "78/78 - 5s - loss: 0.0114 - accuracy: 0.8050 - val_loss: 0.9677 - val_accuracy: 0.6661 - 5s/epoch - 60ms/step\n",
            "Epoch 41/50\n",
            "78/78 - 5s - loss: 0.0114 - accuracy: 0.8034 - val_loss: 0.9967 - val_accuracy: 0.6548 - 5s/epoch - 61ms/step\n",
            "Epoch 42/50\n",
            "78/78 - 5s - loss: 0.0105 - accuracy: 0.7800 - val_loss: 1.0061 - val_accuracy: 0.6887 - 5s/epoch - 63ms/step\n",
            "Epoch 43/50\n",
            "78/78 - 5s - loss: 0.0079 - accuracy: 0.8026 - val_loss: 1.0188 - val_accuracy: 0.6726 - 5s/epoch - 62ms/step\n",
            "Epoch 44/50\n",
            "78/78 - 5s - loss: 0.0085 - accuracy: 0.7852 - val_loss: 1.0513 - val_accuracy: 0.6823 - 5s/epoch - 62ms/step\n",
            "Epoch 45/50\n",
            "78/78 - 5s - loss: 0.0077 - accuracy: 0.7671 - val_loss: 1.0550 - val_accuracy: 0.6677 - 5s/epoch - 60ms/step\n",
            "Epoch 46/50\n",
            "78/78 - 5s - loss: 0.0087 - accuracy: 0.7723 - val_loss: 1.1045 - val_accuracy: 0.6516 - 5s/epoch - 62ms/step\n",
            "Epoch 47/50\n",
            "78/78 - 5s - loss: 0.0366 - accuracy: 0.7558 - val_loss: 1.0211 - val_accuracy: 0.6677 - 5s/epoch - 62ms/step\n",
            "Epoch 48/50\n",
            "78/78 - 5s - loss: 0.1609 - accuracy: 0.7650 - val_loss: 0.7469 - val_accuracy: 0.6323 - 5s/epoch - 61ms/step\n",
            "Epoch 49/50\n",
            "78/78 - 5s - loss: 0.1242 - accuracy: 0.7969 - val_loss: 0.7438 - val_accuracy: 0.6532 - 5s/epoch - 62ms/step\n",
            "Epoch 50/50\n",
            "78/78 - 5s - loss: 0.0435 - accuracy: 0.8147 - val_loss: 0.7786 - val_accuracy: 0.6323 - 5s/epoch - 62ms/step\n",
            "\n",
            "best model based on min training set loss:  acc= 0.767056941986084  loss= 0.007680141367018223\n",
            "best model based on min test set loss:  acc= 0.8112903237342834  loss= 0.3295774459838867\n",
            "best model based on max training set accuracy:  acc= 0.9491320252418518  loss= 0.1835612952709198\n",
            "best model based on max test set accuracy:  acc= 0.8145161271095276  loss= 0.3318910300731659\n",
            "\n",
            "evaluate dataset with best model based on maximum test set accuracy\n",
            "evaluate train set=  [0.21393226087093353, 0.9422688484191895]\n",
            "evaluate test set=  [0.3318910300731659, 0.8145161271095276]\n",
            "\n",
            " ------------------------------ metrics for traning set ------------------------------\n",
            "confusion matrix: \n",
            " [[[1934   18]\n",
            "  [   2  523]]\n",
            "\n",
            " [[2151   11]\n",
            "  [   2  313]]\n",
            "\n",
            " [[2396    0]\n",
            "  [   9   72]]\n",
            "\n",
            " [[2134    9]\n",
            "  [   8  326]]\n",
            "\n",
            " [[2003   14]\n",
            "  [  20  440]]\n",
            "\n",
            " [[1678   37]\n",
            "  [  48  714]]\n",
            "\n",
            " [[2045   91]\n",
            "  [ 193  148]]\n",
            "\n",
            " [[1700  259]\n",
            "  [ 348  170]]\n",
            "\n",
            " [[1226  616]\n",
            "  [ 212  423]]\n",
            "\n",
            " [[1927   98]\n",
            "  [ 290  162]]\n",
            "\n",
            " [[1717  229]\n",
            "  [ 250  281]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.967     0.996     0.981       525\n",
            "  Women/Girl      0.966     0.994     0.980       315\n",
            "       Child      1.000     0.889     0.941        81\n",
            "      Animal      0.973     0.976     0.975       334\n",
            "       Other      0.969     0.957     0.963       460\n",
            "          NA      0.951     0.937     0.944       762\n",
            "      Score1      0.619     0.434     0.510       341\n",
            "      Score2      0.396     0.328     0.359       518\n",
            "      Score3      0.407     0.666     0.505       635\n",
            "      Score4      0.623     0.358     0.455       452\n",
            "      Score5      0.551     0.529     0.540       531\n",
            "\n",
            "   micro avg      0.721     0.721     0.721      4954\n",
            "   macro avg      0.766     0.733     0.741      4954\n",
            "weighted avg      0.734     0.721     0.719      4954\n",
            " samples avg      0.721     0.721     0.721      4954\n",
            "\n",
            "------------------------------ metrics for test set ------------------------------\n",
            "confusion matrix: \n",
            " [[[450  26]\n",
            "  [ 11 133]]\n",
            "\n",
            " [[532  13]\n",
            "  [  5  70]]\n",
            "\n",
            " [[598   4]\n",
            "  [  7  11]]\n",
            "\n",
            " [[522   8]\n",
            "  [ 13  77]]\n",
            "\n",
            " [[501  11]\n",
            "  [ 15  93]]\n",
            "\n",
            " [[390  45]\n",
            "  [ 56 129]]\n",
            "\n",
            " [[489  38]\n",
            "  [ 71  22]]\n",
            "\n",
            " [[396  78]\n",
            "  [123  23]]\n",
            "\n",
            " [[289 179]\n",
            "  [ 87  65]]\n",
            "\n",
            " [[458  64]\n",
            "  [ 87  11]]\n",
            "\n",
            " [[382 107]\n",
            "  [ 98  33]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.836     0.924     0.878       144\n",
            "  Women/Girl      0.843     0.933     0.886        75\n",
            "       Child      0.733     0.611     0.667        18\n",
            "      Animal      0.906     0.856     0.880        90\n",
            "       Other      0.894     0.861     0.877       108\n",
            "          NA      0.741     0.697     0.719       185\n",
            "      Score1      0.367     0.237     0.288        93\n",
            "      Score2      0.228     0.158     0.186       146\n",
            "      Score3      0.266     0.428     0.328       152\n",
            "      Score4      0.147     0.112     0.127        98\n",
            "      Score5      0.236     0.252     0.244       131\n",
            "\n",
            "   micro avg      0.538     0.538     0.538      1240\n",
            "   macro avg      0.563     0.552     0.553      1240\n",
            "weighted avg      0.536     0.538     0.532      1240\n",
            " samples avg      0.538     0.538     0.538      1240\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Model 2**"
      ],
      "metadata": {
        "id": "vMWnMMgLRlw7"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "with tf.device('/GPU:0'):\n",
        "  model7 = RNN(X_train_list, y_train, model_num=7)\n",
        "  model7.train((X_test_list, y_test))\n",
        "  model7.evaluate(X_test_list, y_test)\n",
        "  model7.outputResult()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "NWUFCmsbRsul",
        "outputId": "505aeee4-9f00-4302-91f8-5698a15bd001"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "78/78 - 8s - loss: 0.5994 - accuracy: 0.0775 - val_loss: 0.5112 - val_accuracy: 0.0468 - 8s/epoch - 101ms/step\n",
            "Epoch 2/50\n",
            "78/78 - 5s - loss: 0.4364 - accuracy: 0.3746 - val_loss: 0.4022 - val_accuracy: 0.6081 - 5s/epoch - 67ms/step\n",
            "Epoch 3/50\n",
            "78/78 - 5s - loss: 0.3794 - accuracy: 0.6609 - val_loss: 0.3710 - val_accuracy: 0.6597 - 5s/epoch - 66ms/step\n",
            "Epoch 4/50\n",
            "78/78 - 5s - loss: 0.3482 - accuracy: 0.7323 - val_loss: 0.3542 - val_accuracy: 0.6758 - 5s/epoch - 70ms/step\n",
            "Epoch 5/50\n",
            "78/78 - 5s - loss: 0.3258 - accuracy: 0.7957 - val_loss: 0.3420 - val_accuracy: 0.7806 - 5s/epoch - 66ms/step\n",
            "Epoch 6/50\n",
            "78/78 - 5s - loss: 0.3068 - accuracy: 0.8421 - val_loss: 0.3276 - val_accuracy: 0.8065 - 5s/epoch - 66ms/step\n",
            "Epoch 7/50\n",
            "78/78 - 5s - loss: 0.2960 - accuracy: 0.8797 - val_loss: 0.3135 - val_accuracy: 0.8516 - 5s/epoch - 66ms/step\n",
            "Epoch 8/50\n",
            "78/78 - 5s - loss: 0.2863 - accuracy: 0.8958 - val_loss: 0.3345 - val_accuracy: 0.7952 - 5s/epoch - 67ms/step\n",
            "Epoch 9/50\n",
            "78/78 - 5s - loss: 0.2817 - accuracy: 0.9084 - val_loss: 0.3161 - val_accuracy: 0.8387 - 5s/epoch - 66ms/step\n",
            "Epoch 10/50\n",
            "78/78 - 5s - loss: 0.2759 - accuracy: 0.9071 - val_loss: 0.3166 - val_accuracy: 0.8387 - 5s/epoch - 65ms/step\n",
            "Epoch 11/50\n",
            "78/78 - 5s - loss: 0.2749 - accuracy: 0.9031 - val_loss: 0.3092 - val_accuracy: 0.8532 - 5s/epoch - 67ms/step\n",
            "Epoch 12/50\n",
            "78/78 - 5s - loss: 0.2731 - accuracy: 0.9136 - val_loss: 0.3096 - val_accuracy: 0.8597 - 5s/epoch - 66ms/step\n",
            "Epoch 13/50\n",
            "78/78 - 5s - loss: 0.2756 - accuracy: 0.8995 - val_loss: 0.3106 - val_accuracy: 0.8516 - 5s/epoch - 67ms/step\n",
            "Epoch 14/50\n",
            "78/78 - 5s - loss: 0.2638 - accuracy: 0.9302 - val_loss: 0.3064 - val_accuracy: 0.8452 - 5s/epoch - 65ms/step\n",
            "Epoch 15/50\n",
            "78/78 - 5s - loss: 0.2652 - accuracy: 0.9176 - val_loss: 0.3034 - val_accuracy: 0.8613 - 5s/epoch - 65ms/step\n",
            "Epoch 16/50\n",
            "78/78 - 5s - loss: 0.2595 - accuracy: 0.9342 - val_loss: 0.3096 - val_accuracy: 0.8645 - 5s/epoch - 70ms/step\n",
            "Epoch 17/50\n",
            "78/78 - 5s - loss: 0.2524 - accuracy: 0.9447 - val_loss: 0.3107 - val_accuracy: 0.8661 - 5s/epoch - 68ms/step\n",
            "Epoch 18/50\n",
            "78/78 - 5s - loss: 0.2443 - accuracy: 0.9556 - val_loss: 0.3217 - val_accuracy: 0.8484 - 5s/epoch - 66ms/step\n",
            "Epoch 19/50\n",
            "78/78 - 5s - loss: 0.2390 - accuracy: 0.9600 - val_loss: 0.3209 - val_accuracy: 0.8597 - 5s/epoch - 65ms/step\n",
            "Epoch 20/50\n",
            "78/78 - 5s - loss: 0.2344 - accuracy: 0.9637 - val_loss: 0.3209 - val_accuracy: 0.8581 - 5s/epoch - 64ms/step\n",
            "Epoch 21/50\n",
            "78/78 - 5s - loss: 0.2286 - accuracy: 0.9649 - val_loss: 0.3291 - val_accuracy: 0.8516 - 5s/epoch - 63ms/step\n",
            "Epoch 22/50\n",
            "78/78 - 5s - loss: 0.2242 - accuracy: 0.9705 - val_loss: 0.3323 - val_accuracy: 0.8339 - 5s/epoch - 66ms/step\n",
            "Epoch 23/50\n",
            "78/78 - 5s - loss: 0.2184 - accuracy: 0.9697 - val_loss: 0.3406 - val_accuracy: 0.8387 - 5s/epoch - 64ms/step\n",
            "Epoch 24/50\n",
            "78/78 - 5s - loss: 0.2144 - accuracy: 0.9705 - val_loss: 0.3369 - val_accuracy: 0.8516 - 5s/epoch - 66ms/step\n",
            "Epoch 25/50\n",
            "78/78 - 5s - loss: 0.2067 - accuracy: 0.9762 - val_loss: 0.3482 - val_accuracy: 0.8323 - 5s/epoch - 65ms/step\n",
            "Epoch 26/50\n",
            "78/78 - 5s - loss: 0.2135 - accuracy: 0.9544 - val_loss: 0.3590 - val_accuracy: 0.8371 - 5s/epoch - 65ms/step\n",
            "Epoch 27/50\n",
            "78/78 - 5s - loss: 0.2078 - accuracy: 0.9629 - val_loss: 0.3744 - val_accuracy: 0.8371 - 5s/epoch - 66ms/step\n",
            "Epoch 28/50\n",
            "78/78 - 6s - loss: 0.2083 - accuracy: 0.9604 - val_loss: 0.3619 - val_accuracy: 0.8371 - 6s/epoch - 75ms/step\n",
            "Epoch 29/50\n",
            "78/78 - 5s - loss: 0.1975 - accuracy: 0.9734 - val_loss: 0.3725 - val_accuracy: 0.8290 - 5s/epoch - 64ms/step\n",
            "Epoch 30/50\n",
            "78/78 - 5s - loss: 0.1960 - accuracy: 0.9693 - val_loss: 0.3778 - val_accuracy: 0.8194 - 5s/epoch - 65ms/step\n",
            "Epoch 31/50\n",
            "78/78 - 5s - loss: 0.1897 - accuracy: 0.9713 - val_loss: 0.4063 - val_accuracy: 0.8065 - 5s/epoch - 65ms/step\n",
            "Epoch 32/50\n",
            "78/78 - 5s - loss: 0.1943 - accuracy: 0.9612 - val_loss: 0.3941 - val_accuracy: 0.8081 - 5s/epoch - 65ms/step\n",
            "Epoch 33/50\n",
            "78/78 - 5s - loss: 0.1921 - accuracy: 0.9596 - val_loss: 0.3889 - val_accuracy: 0.8306 - 5s/epoch - 67ms/step\n",
            "Epoch 34/50\n",
            "78/78 - 5s - loss: 0.1800 - accuracy: 0.9681 - val_loss: 0.4220 - val_accuracy: 0.7952 - 5s/epoch - 65ms/step\n",
            "Epoch 35/50\n",
            "78/78 - 5s - loss: 0.1760 - accuracy: 0.9701 - val_loss: 0.4183 - val_accuracy: 0.8016 - 5s/epoch - 65ms/step\n",
            "Epoch 36/50\n",
            "78/78 - 5s - loss: 0.1668 - accuracy: 0.9734 - val_loss: 0.4300 - val_accuracy: 0.8210 - 5s/epoch - 64ms/step\n",
            "Epoch 37/50\n",
            "78/78 - 5s - loss: 0.1613 - accuracy: 0.9754 - val_loss: 0.4366 - val_accuracy: 0.8290 - 5s/epoch - 63ms/step\n",
            "Epoch 38/50\n",
            "78/78 - 5s - loss: 0.1598 - accuracy: 0.9697 - val_loss: 0.4692 - val_accuracy: 0.7871 - 5s/epoch - 63ms/step\n",
            "Epoch 39/50\n",
            "78/78 - 5s - loss: 0.1551 - accuracy: 0.9717 - val_loss: 0.4564 - val_accuracy: 0.7919 - 5s/epoch - 65ms/step\n",
            "Epoch 40/50\n",
            "78/78 - 5s - loss: 0.1524 - accuracy: 0.9661 - val_loss: 0.4771 - val_accuracy: 0.8097 - 5s/epoch - 64ms/step\n",
            "Epoch 41/50\n",
            "78/78 - 5s - loss: 0.1476 - accuracy: 0.9709 - val_loss: 0.4989 - val_accuracy: 0.7774 - 5s/epoch - 63ms/step\n",
            "Epoch 42/50\n",
            "78/78 - 5s - loss: 0.1418 - accuracy: 0.9645 - val_loss: 0.5188 - val_accuracy: 0.8065 - 5s/epoch - 67ms/step\n",
            "Epoch 43/50\n",
            "78/78 - 5s - loss: 0.1423 - accuracy: 0.9649 - val_loss: 0.5260 - val_accuracy: 0.7871 - 5s/epoch - 65ms/step\n",
            "Epoch 44/50\n",
            "78/78 - 5s - loss: 0.1493 - accuracy: 0.9499 - val_loss: 0.5099 - val_accuracy: 0.7839 - 5s/epoch - 66ms/step\n",
            "Epoch 45/50\n",
            "78/78 - 5s - loss: 0.1360 - accuracy: 0.9669 - val_loss: 0.5272 - val_accuracy: 0.8065 - 5s/epoch - 66ms/step\n",
            "Epoch 46/50\n",
            "78/78 - 5s - loss: 0.1310 - accuracy: 0.9604 - val_loss: 0.5678 - val_accuracy: 0.7903 - 5s/epoch - 64ms/step\n",
            "Epoch 47/50\n",
            "78/78 - 5s - loss: 0.1355 - accuracy: 0.9491 - val_loss: 0.5425 - val_accuracy: 0.7774 - 5s/epoch - 67ms/step\n",
            "Epoch 48/50\n",
            "78/78 - 5s - loss: 0.1214 - accuracy: 0.9580 - val_loss: 0.5676 - val_accuracy: 0.8048 - 5s/epoch - 67ms/step\n",
            "Epoch 49/50\n",
            "78/78 - 5s - loss: 0.1114 - accuracy: 0.9604 - val_loss: 0.6028 - val_accuracy: 0.7742 - 5s/epoch - 67ms/step\n",
            "Epoch 50/50\n",
            "78/78 - 5s - loss: 0.1070 - accuracy: 0.9604 - val_loss: 0.6207 - val_accuracy: 0.7629 - 5s/epoch - 66ms/step\n",
            "\n",
            "best model based on min training set loss:  acc= 0.9604359865188599  loss= 0.10700482875108719\n",
            "best model based on min test set loss:  acc= 0.8612903356552124  loss= 0.30343344807624817\n",
            "best model based on max training set accuracy:  acc= 0.9761808514595032  loss= 0.2067234069108963\n",
            "best model based on max test set accuracy:  acc= 0.8661290407180786  loss= 0.31066611409187317\n",
            "\n",
            "evaluate dataset with best model based on maximum test set accuracy\n",
            "evaluate train set=  [0.24371099472045898, 0.9495357275009155]\n",
            "evaluate test set=  [0.31066611409187317, 0.8661290407180786]\n",
            "\n",
            " ------------------------------ metrics for traning set ------------------------------\n",
            "confusion matrix: \n",
            " [[[1935    5]\n",
            "  [  17  520]]\n",
            "\n",
            " [[2140   25]\n",
            "  [   0  312]]\n",
            "\n",
            " [[2395    1]\n",
            "  [  11   70]]\n",
            "\n",
            " [[2131   10]\n",
            "  [   0  336]]\n",
            "\n",
            " [[2004   13]\n",
            "  [  37  423]]\n",
            "\n",
            " [[1662   64]\n",
            "  [  53  698]]\n",
            "\n",
            " [[1925  202]\n",
            "  [ 244  106]]\n",
            "\n",
            " [[1731  218]\n",
            "  [ 422  106]]\n",
            "\n",
            " [[1463  388]\n",
            "  [ 403  223]]\n",
            "\n",
            " [[1963   66]\n",
            "  [ 401   47]]\n",
            "\n",
            " [[1190  762]\n",
            "  [ 166  359]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.990     0.968     0.979       537\n",
            "  Women/Girl      0.926     1.000     0.961       312\n",
            "       Child      0.986     0.864     0.921        81\n",
            "      Animal      0.971     1.000     0.985       336\n",
            "       Other      0.970     0.920     0.944       460\n",
            "          NA      0.916     0.929     0.923       751\n",
            "      Score1      0.344     0.303     0.322       350\n",
            "      Score2      0.327     0.201     0.249       528\n",
            "      Score3      0.365     0.356     0.361       626\n",
            "      Score4      0.416     0.105     0.168       448\n",
            "      Score5      0.320     0.684     0.436       525\n",
            "\n",
            "   micro avg      0.646     0.646     0.646      4954\n",
            "   macro avg      0.685     0.666     0.659      4954\n",
            "weighted avg      0.653     0.646     0.632      4954\n",
            " samples avg      0.646     0.646     0.646      4954\n",
            "\n",
            "------------------------------ metrics for test set ------------------------------\n",
            "confusion matrix: \n",
            " [[[479   9]\n",
            "  [  8 124]]\n",
            "\n",
            " [[527  15]\n",
            "  [  6  72]]\n",
            "\n",
            " [[601   1]\n",
            "  [  2  16]]\n",
            "\n",
            " [[524   8]\n",
            "  [  5  83]]\n",
            "\n",
            " [[505   7]\n",
            "  [ 23  85]]\n",
            "\n",
            " [[385  39]\n",
            "  [ 35 161]]\n",
            "\n",
            " [[481  55]\n",
            "  [ 70  14]]\n",
            "\n",
            " [[414  70]\n",
            "  [118  18]]\n",
            "\n",
            " [[351 108]\n",
            "  [119  42]]\n",
            "\n",
            " [[497  21]\n",
            "  [ 99   3]]\n",
            "\n",
            " [[256 227]\n",
            "  [ 75  62]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.932     0.939     0.936       132\n",
            "  Women/Girl      0.828     0.923     0.873        78\n",
            "       Child      0.941     0.889     0.914        18\n",
            "      Animal      0.912     0.943     0.927        88\n",
            "       Other      0.924     0.787     0.850       108\n",
            "          NA      0.805     0.821     0.813       196\n",
            "      Score1      0.203     0.167     0.183        84\n",
            "      Score2      0.205     0.132     0.161       136\n",
            "      Score3      0.280     0.261     0.270       161\n",
            "      Score4      0.125     0.029     0.048       102\n",
            "      Score5      0.215     0.453     0.291       137\n",
            "\n",
            "   micro avg      0.548     0.548     0.548      1240\n",
            "   macro avg      0.579     0.577     0.570      1240\n",
            "weighted avg      0.544     0.548     0.537      1240\n",
            " samples avg      0.548     0.548     0.548      1240\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "showPlots(model7)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 545
        },
        "id": "P1lt1WBpowO7",
        "outputId": "7d033c15-a2a4-447a-dc6f-7f50c1070149"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plot_model(model7.model, to_file='/SimpleRNN_plot.png')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 856
        },
        "id": "s_6lNCEeIC5z",
        "outputId": "358be066-7f81-4bf8-c731-cc6f31a4c9b3"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<IPython.core.display.Image object>"
            ],
            "image/png": "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\n"
          },
          "metadata": {},
          "execution_count": 48
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **GRU Model**"
      ],
      "metadata": {
        "id": "CseiX3GND2Wn"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Model 1**"
      ],
      "metadata": {
        "id": "xPG4nWnsF_dy"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "with tf.device('/GPU:0'):\n",
        "  model4 = RNN(X_train_list, y_train, model_num=4)\n",
        "  model4.train((X_test_list, y_test))\n",
        "  model4.evaluate(X_test_list, y_test)\n",
        "  model4.outputResult()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "YOhT_4ZID8x3",
        "outputId": "0430f475-5125-4d51-abef-2c68e37b4a92"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "78/78 - 8s - loss: 0.5300 - accuracy: 0.0287 - val_loss: 0.4705 - val_accuracy: 0.2984 - 8s/epoch - 98ms/step\n",
            "Epoch 2/50\n",
            "78/78 - 1s - loss: 0.4590 - accuracy: 0.3076 - val_loss: 0.4565 - val_accuracy: 0.2984 - 1s/epoch - 13ms/step\n",
            "Epoch 3/50\n",
            "78/78 - 1s - loss: 0.4567 - accuracy: 0.3076 - val_loss: 0.4561 - val_accuracy: 0.2984 - 1s/epoch - 13ms/step\n",
            "Epoch 4/50\n",
            "78/78 - 1s - loss: 0.4555 - accuracy: 0.2604 - val_loss: 0.4548 - val_accuracy: 0.2984 - 1s/epoch - 13ms/step\n",
            "Epoch 5/50\n",
            "78/78 - 1s - loss: 0.4398 - accuracy: 0.3096 - val_loss: 0.4254 - val_accuracy: 0.4935 - 1s/epoch - 14ms/step\n",
            "Epoch 6/50\n",
            "78/78 - 1s - loss: 0.4071 - accuracy: 0.4546 - val_loss: 0.3909 - val_accuracy: 0.5710 - 1s/epoch - 14ms/step\n",
            "Epoch 7/50\n",
            "78/78 - 1s - loss: 0.3730 - accuracy: 0.6193 - val_loss: 0.3564 - val_accuracy: 0.7387 - 1s/epoch - 14ms/step\n",
            "Epoch 8/50\n",
            "78/78 - 1s - loss: 0.3386 - accuracy: 0.7626 - val_loss: 0.3445 - val_accuracy: 0.7419 - 1s/epoch - 14ms/step\n",
            "Epoch 9/50\n",
            "78/78 - 1s - loss: 0.3202 - accuracy: 0.8034 - val_loss: 0.3265 - val_accuracy: 0.7984 - 1s/epoch - 14ms/step\n",
            "Epoch 10/50\n",
            "78/78 - 1s - loss: 0.3039 - accuracy: 0.8462 - val_loss: 0.3171 - val_accuracy: 0.8371 - 1s/epoch - 14ms/step\n",
            "Epoch 11/50\n",
            "78/78 - 1s - loss: 0.2931 - accuracy: 0.8813 - val_loss: 0.3133 - val_accuracy: 0.8516 - 1s/epoch - 14ms/step\n",
            "Epoch 12/50\n",
            "78/78 - 1s - loss: 0.2830 - accuracy: 0.9047 - val_loss: 0.3058 - val_accuracy: 0.8774 - 1s/epoch - 14ms/step\n",
            "Epoch 13/50\n",
            "78/78 - 1s - loss: 0.2764 - accuracy: 0.9140 - val_loss: 0.3050 - val_accuracy: 0.8548 - 987ms/epoch - 13ms/step\n",
            "Epoch 14/50\n",
            "78/78 - 1s - loss: 0.2709 - accuracy: 0.9217 - val_loss: 0.3006 - val_accuracy: 0.8806 - 1s/epoch - 14ms/step\n",
            "Epoch 15/50\n",
            "78/78 - 1s - loss: 0.2666 - accuracy: 0.9285 - val_loss: 0.2995 - val_accuracy: 0.8742 - 1s/epoch - 13ms/step\n",
            "Epoch 16/50\n",
            "78/78 - 1s - loss: 0.2591 - accuracy: 0.9415 - val_loss: 0.3053 - val_accuracy: 0.8565 - 1s/epoch - 13ms/step\n",
            "Epoch 17/50\n",
            "78/78 - 1s - loss: 0.2592 - accuracy: 0.9366 - val_loss: 0.3048 - val_accuracy: 0.8532 - 991ms/epoch - 13ms/step\n",
            "Epoch 18/50\n",
            "78/78 - 1s - loss: 0.2565 - accuracy: 0.9431 - val_loss: 0.3171 - val_accuracy: 0.8661 - 1s/epoch - 13ms/step\n",
            "Epoch 19/50\n",
            "78/78 - 1s - loss: 0.2521 - accuracy: 0.9447 - val_loss: 0.3042 - val_accuracy: 0.8645 - 1s/epoch - 13ms/step\n",
            "Epoch 20/50\n",
            "78/78 - 1s - loss: 0.2444 - accuracy: 0.9584 - val_loss: 0.3247 - val_accuracy: 0.8661 - 1s/epoch - 13ms/step\n",
            "Epoch 21/50\n",
            "78/78 - 1s - loss: 0.2448 - accuracy: 0.9576 - val_loss: 0.3213 - val_accuracy: 0.8532 - 1s/epoch - 13ms/step\n",
            "Epoch 22/50\n",
            "78/78 - 1s - loss: 0.2449 - accuracy: 0.9576 - val_loss: 0.3174 - val_accuracy: 0.8694 - 1s/epoch - 13ms/step\n",
            "Epoch 23/50\n",
            "78/78 - 1s - loss: 0.2393 - accuracy: 0.9588 - val_loss: 0.3260 - val_accuracy: 0.8710 - 1s/epoch - 13ms/step\n",
            "Epoch 24/50\n",
            "78/78 - 1s - loss: 0.2388 - accuracy: 0.9621 - val_loss: 0.3239 - val_accuracy: 0.8597 - 1s/epoch - 13ms/step\n",
            "Epoch 25/50\n",
            "78/78 - 1s - loss: 0.2368 - accuracy: 0.9604 - val_loss: 0.3299 - val_accuracy: 0.8613 - 1s/epoch - 13ms/step\n",
            "Epoch 26/50\n",
            "78/78 - 1s - loss: 0.2367 - accuracy: 0.9665 - val_loss: 0.3174 - val_accuracy: 0.8677 - 1s/epoch - 13ms/step\n",
            "Epoch 27/50\n",
            "78/78 - 1s - loss: 0.2397 - accuracy: 0.9499 - val_loss: 0.3267 - val_accuracy: 0.8516 - 1s/epoch - 13ms/step\n",
            "Epoch 28/50\n",
            "78/78 - 1s - loss: 0.2295 - accuracy: 0.9701 - val_loss: 0.3328 - val_accuracy: 0.8581 - 1s/epoch - 13ms/step\n",
            "Epoch 29/50\n",
            "78/78 - 1s - loss: 0.2259 - accuracy: 0.9681 - val_loss: 0.3372 - val_accuracy: 0.8597 - 1s/epoch - 13ms/step\n",
            "Epoch 30/50\n",
            "78/78 - 1s - loss: 0.2231 - accuracy: 0.9717 - val_loss: 0.3484 - val_accuracy: 0.8613 - 1s/epoch - 13ms/step\n",
            "Epoch 31/50\n",
            "78/78 - 1s - loss: 0.2291 - accuracy: 0.9600 - val_loss: 0.3396 - val_accuracy: 0.8565 - 994ms/epoch - 13ms/step\n",
            "Epoch 32/50\n",
            "78/78 - 1s - loss: 0.2244 - accuracy: 0.9653 - val_loss: 0.3477 - val_accuracy: 0.8597 - 991ms/epoch - 13ms/step\n",
            "Epoch 33/50\n",
            "78/78 - 1s - loss: 0.2162 - accuracy: 0.9786 - val_loss: 0.3518 - val_accuracy: 0.8565 - 999ms/epoch - 13ms/step\n",
            "Epoch 34/50\n",
            "78/78 - 1s - loss: 0.2114 - accuracy: 0.9822 - val_loss: 0.3678 - val_accuracy: 0.8629 - 1s/epoch - 13ms/step\n",
            "Epoch 35/50\n",
            "78/78 - 1s - loss: 0.2121 - accuracy: 0.9730 - val_loss: 0.3756 - val_accuracy: 0.8629 - 1s/epoch - 13ms/step\n",
            "Epoch 36/50\n",
            "78/78 - 1s - loss: 0.2130 - accuracy: 0.9713 - val_loss: 0.3842 - val_accuracy: 0.8516 - 1s/epoch - 13ms/step\n",
            "Epoch 37/50\n",
            "78/78 - 1s - loss: 0.2065 - accuracy: 0.9754 - val_loss: 0.3808 - val_accuracy: 0.8516 - 989ms/epoch - 13ms/step\n",
            "Epoch 38/50\n",
            "78/78 - 1s - loss: 0.1999 - accuracy: 0.9834 - val_loss: 0.4068 - val_accuracy: 0.8468 - 998ms/epoch - 13ms/step\n",
            "Epoch 39/50\n",
            "78/78 - 1s - loss: 0.1976 - accuracy: 0.9859 - val_loss: 0.4019 - val_accuracy: 0.8468 - 1s/epoch - 13ms/step\n",
            "Epoch 40/50\n",
            "78/78 - 1s - loss: 0.1989 - accuracy: 0.9762 - val_loss: 0.4073 - val_accuracy: 0.8516 - 1s/epoch - 13ms/step\n",
            "Epoch 41/50\n",
            "78/78 - 1s - loss: 0.1979 - accuracy: 0.9750 - val_loss: 0.4131 - val_accuracy: 0.8435 - 985ms/epoch - 13ms/step\n",
            "Epoch 42/50\n",
            "78/78 - 1s - loss: 0.1908 - accuracy: 0.9834 - val_loss: 0.4304 - val_accuracy: 0.8484 - 1s/epoch - 13ms/step\n",
            "Epoch 43/50\n",
            "78/78 - 1s - loss: 0.1951 - accuracy: 0.9750 - val_loss: 0.4236 - val_accuracy: 0.8403 - 1s/epoch - 13ms/step\n",
            "Epoch 44/50\n",
            "78/78 - 1s - loss: 0.1877 - accuracy: 0.9774 - val_loss: 0.4330 - val_accuracy: 0.8419 - 1s/epoch - 13ms/step\n",
            "Epoch 45/50\n",
            "78/78 - 1s - loss: 0.1893 - accuracy: 0.9721 - val_loss: 0.4380 - val_accuracy: 0.8500 - 1s/epoch - 13ms/step\n",
            "Epoch 46/50\n",
            "78/78 - 1s - loss: 0.1819 - accuracy: 0.9774 - val_loss: 0.4289 - val_accuracy: 0.8419 - 1s/epoch - 13ms/step\n",
            "Epoch 47/50\n",
            "78/78 - 1s - loss: 0.1996 - accuracy: 0.9584 - val_loss: 0.4406 - val_accuracy: 0.8194 - 1s/epoch - 13ms/step\n",
            "Epoch 48/50\n",
            "78/78 - 1s - loss: 0.1922 - accuracy: 0.9641 - val_loss: 0.4219 - val_accuracy: 0.8435 - 1s/epoch - 14ms/step\n",
            "Epoch 49/50\n",
            "78/78 - 1s - loss: 0.2193 - accuracy: 0.9374 - val_loss: 0.4007 - val_accuracy: 0.7435 - 1s/epoch - 14ms/step\n",
            "Epoch 50/50\n",
            "78/78 - 1s - loss: 0.2526 - accuracy: 0.8894 - val_loss: 0.3584 - val_accuracy: 0.8242 - 1s/epoch - 13ms/step\n",
            "\n",
            "best model based on min training set loss:  acc= 0.9773920178413391  loss= 0.18186011910438538\n",
            "best model based on min test set loss:  acc= 0.874193549156189  loss= 0.2995363771915436\n",
            "best model based on max training set accuracy:  acc= 0.9858700037002563  loss= 0.19760654866695404\n",
            "best model based on max test set accuracy:  acc= 0.8806451559066772  loss= 0.30063745379447937\n",
            "\n",
            "evaluate dataset with best model based on maximum test set accuracy\n",
            "evaluate train set=  [0.2656533122062683, 0.9345983266830444]\n",
            "evaluate test set=  [0.30063745379447937, 0.8806451559066772]\n",
            "\n",
            " ------------------------------ metrics for traning set ------------------------------\n",
            "confusion matrix: \n",
            " [[[1924   28]\n",
            "  [  15  510]]\n",
            "\n",
            " [[2146   16]\n",
            "  [   5  310]]\n",
            "\n",
            " [[2393    3]\n",
            "  [   0   81]]\n",
            "\n",
            " [[2138    5]\n",
            "  [  22  312]]\n",
            "\n",
            " [[1967   50]\n",
            "  [  39  421]]\n",
            "\n",
            " [[1656   59]\n",
            "  [  80  682]]\n",
            "\n",
            " [[1769  367]\n",
            "  [ 218  123]]\n",
            "\n",
            " [[1905   54]\n",
            "  [ 490   28]]\n",
            "\n",
            " [[ 889  953]\n",
            "  [ 247  388]]\n",
            "\n",
            " [[1983   42]\n",
            "  [ 426   26]]\n",
            "\n",
            " [[1571  375]\n",
            "  [ 410  121]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.948     0.971     0.960       525\n",
            "  Women/Girl      0.951     0.984     0.967       315\n",
            "       Child      0.964     1.000     0.982        81\n",
            "      Animal      0.984     0.934     0.959       334\n",
            "       Other      0.894     0.915     0.904       460\n",
            "          NA      0.920     0.895     0.908       762\n",
            "      Score1      0.251     0.361     0.296       341\n",
            "      Score2      0.341     0.054     0.093       518\n",
            "      Score3      0.289     0.611     0.393       635\n",
            "      Score4      0.382     0.058     0.100       452\n",
            "      Score5      0.244     0.228     0.236       531\n",
            "\n",
            "   micro avg      0.606     0.606     0.606      4954\n",
            "   macro avg      0.652     0.637     0.618      4954\n",
            "weighted avg      0.619     0.606     0.582      4954\n",
            " samples avg      0.606     0.606     0.606      4954\n",
            "\n",
            "------------------------------ metrics for test set ------------------------------\n",
            "confusion matrix: \n",
            " [[[459  17]\n",
            "  [  8 136]]\n",
            "\n",
            " [[539   6]\n",
            "  [  2  73]]\n",
            "\n",
            " [[601   1]\n",
            "  [  2  16]]\n",
            "\n",
            " [[521   9]\n",
            "  [ 10  80]]\n",
            "\n",
            " [[500  12]\n",
            "  [ 10  98]]\n",
            "\n",
            " [[407  28]\n",
            "  [ 41 144]]\n",
            "\n",
            " [[438  89]\n",
            "  [ 59  34]]\n",
            "\n",
            " [[459  15]\n",
            "  [141   5]]\n",
            "\n",
            " [[215 253]\n",
            "  [ 62  90]]\n",
            "\n",
            " [[505  17]\n",
            "  [ 94   4]]\n",
            "\n",
            " [[399  90]\n",
            "  [108  23]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.889     0.944     0.916       144\n",
            "  Women/Girl      0.924     0.973     0.948        75\n",
            "       Child      0.941     0.889     0.914        18\n",
            "      Animal      0.899     0.889     0.894        90\n",
            "       Other      0.891     0.907     0.899       108\n",
            "          NA      0.837     0.778     0.807       185\n",
            "      Score1      0.276     0.366     0.315        93\n",
            "      Score2      0.250     0.034     0.060       146\n",
            "      Score3      0.262     0.592     0.364       152\n",
            "      Score4      0.190     0.041     0.067        98\n",
            "      Score5      0.204     0.176     0.189       131\n",
            "\n",
            "   micro avg      0.567     0.567     0.567      1240\n",
            "   macro avg      0.597     0.599     0.579      1240\n",
            "weighted avg      0.559     0.567     0.541      1240\n",
            " samples avg      0.567     0.567     0.567      1240\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "showPlots(model4)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 541
        },
        "id": "F7_-xrr3of8E",
        "outputId": "e618377b-13c5-456a-84cf-b8d318c8d7bb"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEGCAYAAABo25JHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3dd3xUVfr48c8zkwZIDyBSDCqKK1UCGll3QxNcBexiQ10VhZWvruUnrL2jsuq6ll0VUREbFoogIiULSpCEFroUA0lokRJq2szz++NOGqRCJpNknvfrldfMvXPn3ucmk/vMOeeec0RVMcYYE7xcgQ7AGGNMYFkiMMaYIGeJwBhjgpwlAmOMCXKWCIwxJsiFBDqAioqMjNSoqKhAh2GMMTXK0qVLf1fVZsW9VuMSQVRUFImJiYEOwxhjahQR2VrSa1Y1ZIwxQc4SgTHGBDm/JQIR+UBEdovI6hJeFxF5Q0Q2iUiSiJzvr1iMMcaUzJ8lgg+BgaW8finQ3vczHHjHj7EYY4wpgd8SgaouAPaWsskQ4GN1LAYaiUhLf8VjjDGmeIFsI2gFpBRaTvWtO46IDBeRRBFJTE9Pr5LgjDEmWNSIxmJVfVdVo1U1ulmzYm+DNcbUIvEp8by48EXiU+IDHUpQCGQ/gjSgTaHl1r51xphqID4lnrjkOGKjYolpE+O3Y8zePJtOLTrRtmFb0g+nE58az9ifxpLrzSU8JJy5t8zlorYXndQx/H0eNV0gE8E04F4R+Ry4AMhQ1R0BjMeYoLRo2yJmbZ7FmY3PpFFEI7ZmbCU+JZ7JayfjUQ+hrlAmXzuZIR2GnPAxCl+Mz256NvN+m8cnSZ8w/dfpKCXPiZKZm8mQL4Zw1/l3cdW5V9G9ZXcWpy4u9sJe+BjnRJ7Dj5t/5KOVH/H9pu8BcIubZ3s/y8geI2kY0bDEBHEiiaOmJxvx18Q0IvIZEAtEAruAJ4FQAFX9j4gI8CbOnUVHgNtVtcwuw9HR0Wo9i01tU9ELSWVceDJzM3k67mle+vml4y7GIa4Qcr25RdZ1b9mdIecMIapRFCkHUugd1fu4Yx8bV2ZuJt+u+5bbpt5GjicHIP9Y4e5wsjxZALhwcVPnmxgRPYLUA6ncOuVWsj3ZuMRFl1O7sHzHcjzqoUW9Fuw5sgePeghxhfBQzEN0atGJbRnbeCLuCXI8OYgIqoqi1Ampw9Hco0VidIubjs07sjZ9LR6vs5/h3YdTL6wea9LXMHPjTLzqJURCeCDmAYZ0GMJ5zc4rkjxiWsfQIKIBq3ev5odNP/D5ms9RVSJCIpg7bG61TAYislRVo4t9rabNUGaJwNQEFblQx6fE0+fjPmR7sglzhzH52sn0P6M/4SHh+fv54+l/pG2DtmzN2MqcLXN48acX8Xg9hLnDmDdsXrmqTvL21blFZ5akLeGdxHdIP1Jw84ULF/dE38OTsU+yac8m+k3sR7Ynm1B3KLd1vY0VO1ewOHVx/vaC0L1ld9o0bEOD8AYcyj7E1A1TyfXm4hIXTeo04fcjvx8XR992fXm297Pkai4DJg7IP+/CF9Bjf397j+5l+obpjP15LOt/X1/mufZt15fn+zxPjjeHSyZekn+MVwe8SuqBVCasmMD2g9uLvCfUFUpESAQHsw8Wu88W9VqQfiQdr3qLrHeLG4968pdHRo/krcveKjPGqmaJwJiTUJEqhO0Ht/NJ0ic8Nu8xPOoh3B1e6jdEVeWqL65iyoYpx70W7g4n25NdatUJQLO6zXik1yPc0uUWmtdrXuw287bM4y+f/qXI/i4/+3IGnDmA//fj/yvXxRjgH3P/wdifxubvo23DtjQMb8iBrAPsOryLzNzM/GOef+r5DOkwBK/Xy9ifnTr/8hyjNPEp8fT9uG9+vJ9d/RkdIjsQnxrPPd/dU+5jHJt8pw2dRr8z+rE4dXGR/X9y1SeEucNYtWsVn6/5nKRdSYCTBId2HMoTf36C9CPpDJg4gCxPVn6S+L+e/8dzfZ6jfnj9Es+jqquSLBGYWqu0f6iK1gHHp8QzP3k+nVt05tRTTiXtQBoLty3kX7/8C4/Xg9vl5q7z76JT807sPLTTubh5cnG5XFzQ6gI279vMzkM7i8QgCI9e/CjP9nn2uNi37t/KndPvZM6WObjEuYEvxBXCqJ6jaBzRmFmbZ/Hztp9RFEG4osMVjIgewd6je7l96u35VSfnND2H1emrCXWFMvicwfRs1ZPlO5cT4Y5gf9Z+Vu1axeZ9m4vE9PcL/84/B/yzzN9hcb/vwhfKYy+4pb1WWRe+yqrbP5HPR1nn16NVD6asn8LbCW/TqkErRvUcRa43l7Obno1LXKzYuYJ5v81jUcoiFCXcHc78W+dXSTKwRGBqpanrp3LN5GvI9eYiCOdGnktkvUjC3GEcyT7CL2m/4FEPbnFzcduLaVq3KXuO7GHhtoV41JN/EXW73Ow9uve4qoKKaF6vOQPPGkj3lt2JCIngvln3kZWbhaKcEnYKT/zpCUZdMIqIkAhUlfHLx/PADw/gVS/jLhlH5+ad+d/W/x33rbW8F9a16WsZv2w845ePJyMrIz+u0xueTs9WPWkY3pCPkz7Or046mXrsE0m+tUV5z29x6mJu+uYmtuzbUmS9S1xE1o1k9+Hd+etu73o7Hwz5wG8x57FEYGqVpF1JjFs0jkmrJhWprz2ryVm0btCarNwstmZsLXJhb16vOc3qNiP9SHqRf8L2TdrTqUUntuzbwsqdK1EUFy5u7HQj9194PzsO7eC6ydflX4xn3DiDc5udy4LkBQybMqzYqggouGC0adiGT1d9yvebvuf0hqcztONQvln3DRv3bqR3VG/GDx5Pu8btSjzXil5Yn1vwHE/OfxIv3vy7ZMZcPOaE9mVOznMLnuOJ+U/kf6buib6HcZeMY8XOFfkJ3qteXOLigyEfMKzLML/GY4nABERl3oa3aNsixi8fz5r0NfyS9gv1QutxWfvLmPbrNHI8OeWupqjo+tJiqsj5zd0yl5EzRvLr3l8Bp2Fy/q3z6dW2V4V+p+X5/ZV0HqZqleczFX1aNC/9/BJzf5vL832eZ8wfx+DcUFn5LBGYKpf3T5CVm0WoO5RZN88iNiq2yOt5F9GerXqyNn0tn676lFcWvZJfbXNGozPwqIe9R/cWqe64u/vdvND3BZrUaVLpbQT+/Mb8wsIXeHze48V+W69M9s2/+ijP3yLbk81fp/6VSasmccU5VxB9WjR92vWp9L+dJQJTpTbt3cSt397KotRF+esE4YzGZ3Bus3NpEN6AyWsmO3X7IoS7w4+71xvg3Mhz6X5adzbu2ciStCUo6tcLqL/Zt3VTEq96ue3b25i4aiIAdULqVPrno7REUOOmqjTVT97dNnVD6zJnyxxmbpyJS1y4xY2ihLhCuKnTTRzKPsS639exLn1d/n3XqkqXFl0Y2WMkYe6w/LthwtxhjB88vthqm8Ili5okpk0Mc4fNtW/r5jgucXFus3MRBEXJzM10Oq5V0WfEEoEpt+I6+UxYPoHRc0fn90JtHNGYx//0OHdH383W/VuLvegt3LaQSyZekl+3/+qAV/Nfb9uw7XHvqU0X0Jg2MTU6fuM/sVGxRIREkJmbiaKc0fiMKju2VQ2Zclm0bRF9Jzp1/i5x0aZhG5L3JxfZxoWLp2Kf4vE/P17m/qwe25jjxafEM/3X6bzxyxtc0PoC5twyp9Iaj61qyFRI3kW6Q2QHMrIymJ88n6nrp+b3GvWocy/6c72fo1FEIx7+8eH8apt+Z/Qr1zHsm7Exx8v7v2jbsC0jZoxgYtJEv99WClYiMIVk5mby5pI3GT1ndJGxUyLrRtKpeSd+2vYTXvX6tdeoMcZpPO71QS827tnI+nvXE1k38qT3aXcNmWLFp8Qzc+NMRIQ16Wv4YdMPHM45nP+6Cxf39ryX1wa+hktcdsE3pgqt2rWK8989n1s631IpPY+tasgcZ/KayQz9emh+z9zIupEM6zKM9k3a8+i8R/OreoZ2HJo/Do5V5xhTdTq16MRDMQ8x9uexDOsyzK93y1mJIAhN2zCN6yZflz8WvFvcPNP7Gf5x8T8Aq+oxpro4knOEjm93JMwdxsp7VhIeEn7C+7ISgQHA4/Xw+PzHefGnF+kQ2YHk/cn5t3D2juqdv5198zemeqgbWpe3L3ubSyddyt9m/I0zm5zply9olgiCRPrhdG74+gbm/jaX4ecP51+X/ovlO5bbN39jqrmBZw2k3xn9GL9iPC5xlTnHxYmwRFDLxafEMzFpIl+t/YoDWQf4YPAH3N7tdqDi3/zj4yEuDmJjIcbyhjFVpsdpPZizZQ5e9ZLtya70XseWCGqx+JR4Yj+KJduTjSB8MOQDbut6W/7rcXHOxb08F/affoLevSE3F0JCYNw4uO02aNjQEoQx/jbo7EG8vvh1vw2zYomgFpu0ahLZnmzAGctkx8Ed+a+NGQNjxzrP69SBuXNLvogfPgzDhztJAJzH++93fk4/HVJTQRXCw0vfT3VRUuKyhGaqK38Ps2KJoJZKyUjh89WfIwguceV/i/B6nSTw8ssF2x49Cp9/XvzFb8cOGDQI1q+H0FDweiEsDF58EQ4ehEmTwOMp2M+ECSd3Ea3oRbqiF+8FC6B/f8jJAbcbLrvMSYTbtsHixU5CCwtzElqvE5wqQNV5f0KCJRVTefx5E4clglroYNZBLv/scrI92Xx85cekZKQQGxVL18gYrrsOvv4arrgCfvgBsrKci/t//gOdO8Nf/wp5Q5usWuVcKPfuhWnToGnT4y+6ffs6P1lZzgXwvfecBPHKK9C6dfljVoW33nJKGR4PuFzQrx+ceSYcOABffumURNxuGDIE6tWD5GSnyiovOc2ZAxdfXPz+vV4n2d17L2Q7hSRyc50L9qmnwpEjzjbgnMugQXDffTBsGOzcWXoS6tHDiXfJEvjlF1i4EPbscbYJD4f58088GVRm6aWi71m4EGbMgAsvhO7dC9YvXeokucsvtyRXa6hqjfrp3r27mpLleHL0L5P+ou6n3Tpr46z89Tt3qvbsqSqi+s9/qnq9qosWqb7wgup336n266cKqjffrDpnjuptt6nWrat62mmqy5aVfsy8/cybp/rEE6rh4c5777lH9ZlnnNdLcviw6rvvqnbs6By/8E/jxqrNmqm63UXX16mjGhWl2qJF0fUNGqg+/7zq7t0F+/d6Vb//XrVrV2ebs85SDQtz9lmnTkFsixY5y26383p0tPO7AlWXy3keEqJ69dWqw4ap/ulPzvpjY27f3jlW3ntBtXNn1U2bSv8dzpun+sADqh98oDp/vurUqc7vMjS04NjXXKN6661Fj+12q/7tb6rTpzvHyM0t+Hv89JPqjh2qiYmqY8cW7Cs8XPWTT1T37Sv6OZgyRfWrr1Qffrjg91Xaj8ul+tBDqnv3ln5upnoAErWE62rAL+wV/bFEULp7Z9yrPIX+J+E/qur8k99/v+qppzoXum++Kf59ubnORVuk4CIm4lwcKmrLFudilXfBcLtVr71W9aWXVCdNUo2LU50wwdmmfn1nmy5dVMeMKbgYl3SRLu3i3aOHs6/wcCeRvfCC6hlnOOvatXMufh5PwYXv2AR17PqtW1X79y87CYk4x9uz5/i4QkKc2EJCVO+910m6zz/vxDJ+vOrw4U5yKuuiC6oREaqnn358Aiz8ExZWNAmV9RMefvz2YWGqrVsXTYRXXaX6/vvO47Hbh4erXnedk3AXLizf7/ZkVOa+atKxT5YlgiCwaNsivXzS5cpT6IM/PKiqzrfM0NCCf9j33y97P3feWfQC/sILJxbPCy8U/cYcHl78hcjtVn37beebqWr5L9IlrV+7VnXkSOeimXeM0FDV//3vxM6jPEmo8Pri4tq+3SkdFVeCaNjQSQSFL7p33ul8i//iC+c8yjr2rFnOuvffV/3jH4smqEGDnGQ+fnzBvsLDnaQ/bpxqr14F27tcqiNGqGZmlv+8J0xQHTVKtUmTouflcqn+4Q+qF1zgPOade0iI6qOPqi5erJqRUfbfdeFC1V9/dUpIY8eqXnppQQktLEx17tyy/4Y//lh2ybQsHo/qRx85nyWXq/i/eXVniaCWW7RtkYY9G6Y8hbqfduvsNfE6dqxqvXoVv6iXdYErd0zF7OfAAdV161T/+teiVRsnmmxK8/jjlXeMiiankjzwQNEL/t//XlBCKel3fiKJsSL7qqxjZ2aq3nRT0WTQvr3qgAHOY3mqmYpLHMf+5JUgC7/vz39Wfe451f/+10miDz3kJKd+/VQjI49PdEuXlq9kOGWKU1V3ww2qzZsfH0uHDqr/+pfqmjWqP/9csc/CokWqzz5btcnEEkEtN2bOGOWOC5U/Pa10e1/r1D+i4PxThYdX/KJeWcXfE7lYVZaqOEZlxhTIqpPK/HuXpxTxxRfON/wBA8pOHCJOu8zixar79xfdV3i46o03qnbrdvxFuk4dp53n2PaavJ8GDZx9lKftp0ULJ8k99phTqnK5nPe2alV0O5HyfdZmzSo4RlV+Ni0R1GJer1f/MPoOxZ2p4FVQ7XpBhi5Z4rxeXes0qyKu6nju1TGmylSRUkRlVLvlefTRoiXA558vfl/TpqlOnKh6/vnHJ46oKKctrfCF/f/+r6Dasrhj//ab6pVXFt3XY4+V/PtZudK5CaJwKcUfJeLiWCKoxd785U2lx7/zk4DL5a2yD5YxJ6uyqt0qq0rsREqSee/JS0TNmxf/vi+/dO6mi4wsaDMLCakeJQIbhroGW5e+jm6v90PfWkXOgca4XJLfGcru7zbBpqL9JPzRR6NFC3juOUhJcTpdPvCA893/8ced5ZgYpx9PcjLccYfTR2fHjoK+O6WZNAm2bnWGejmR/+/ShqEO+Df8iv5YicCRlZulXd/urqHnzNGwMK++/37trnIwpqbYt8+5zRacu7LOPtt5Pny406ie5733nPUrV5a9zylTtELtEMWhlBKB9SyuoZ6c/yQrvrgcNvTlP/9xvl0YYwKvUSP46it48EF47TVnXWioM0hjeKF5ZS67zHmcPt3p1V+at992HlWdnvFxcZVb6ndV3q6OJyIDRWSDiGwSkdHFvN5WROaLyHIRSRKRv/gzntpiwdYFjP1wJSx4gltvdQaEM8ZUHyLQrJkz9Ag4w5fExRXdpmVLZ3iS6dNL35eqM9aXy+UMsRIW5lRbVSa/lQhExA28BfQHUoEEEZmmqmsLbfYY8KWqviMifwBmAlH+iqk2+HHzj1z934eQb/5Hx07K22+Xr37RGFO1YmOdEkB2dskX70GD4MknYdcup32hOL/84gyKOGYM1K/vn4EM/Vki6AlsUtUtqpoNfA4MOWYbBRr4njcEtvsxnhpv0bZFDHjuBQ7+ZybqcfP4myupWzfQURljihMT49y48eyzJd/AMWiQ841/xoyS9/PRR84IuaNHO8nAHzeC+DMRtAJSCi2n+tYV9hRws4ik4pQGRvkxnhrv0Y+/Qz+aDQdbgSeMeatXBzokY0wpYmJKv3h36QJt2pRcPZSZ6Yyae9VV0KBB8dtUBr+2EZTDDcCHqtoa+AswUUSOi0lEhotIoogkpqenV3mQ1cG69HUsmNES1Feb53VB8p8DG5Qx5qSIOMN5z57tXPSPNW0a7N/vNDT7kz8TQRrQptBya9+6wu4AvgRQ1XggAog8dkeq+q6qRqtqdLNmzfwUbvV1NOcoQ78eimuXMyi8uLyEh7sYdsXpAY7MGHOyBg1y5sOYP//41z76yJnXo3dv/8bgz0SQALQXkXYiEgYMBaYds802oC+AiJyLkwiC8yt/KR6c/SBJiXXJ/e0ihg8Xnn/Oxfx5bus0Zkwt0Lu3M9HStGOujjt2wKxZcMstzt1C/uS3u4ZUNVdE7gV+ANzAB6q6RkSewenYMA14EHhPRP6O03B8m6/jg/H5eu3XvJPwDq2XbCKnBfzzn3DKKYGOyhhTWSIinOlTv/uOIncBTprk3HZ6663+j8GvHcpUdSZOI3DhdU8Uer4WOMGZYWu/5P3J3DHtDs7edx+/rjqTN9+0JGBMbTRoEEyZAitWQLduzp1EH37oTBN6zjn+P36gG4tNCRZuXcifJvyJnFwP7nkvccYZcNddgY7KGOMPl13mlATy7h5atgzWrPF/I3EeSwTVUHxKPH0+7kPKgRSyVl7JutXhPPus0ynFGFP7tGgBPXsWJIIPP3Q6o11/fdUc3xJBNTR782xyvbmQG4pn7pO0PGsXQ4cGOipjjD8NGgSJic4Io599Bldc4YxbVBUsEVRDzes1B0CWDYd9Z/LQ43vzxywxxtROgwY5jyNHwp49VdNInMdGH62GjuYehax61Fk0lrMvOMDfbzk30CEZY/ysUyenimjmTGja1LmTqKrY98wAi493JqyIjy9Yt3hbIvXmjufI/lMY8dcGNqicMUFg8WKnJACQkQEJCVV3bCsRBNBPP0GfPpCb6wwxe/bZzgdg+/aJOF0v4P77nW8K1nnMmNotLs7pNwDO7aOVPedAaaxEEEATJ0JOjvNH93ic57H9jkJUHIjzicibhMIYU7vlDVvtrzkHSmMlggDK6xyW94f/+GPYHxnHpy89Q/ikn8jNqfoPhDEmMPKGra7ofMmVwRJBAB065Ew0MXp0wYTUT8ctwdV2CTN+yGLJz3Wr/ANhjAmcmJjA/L9bIgigpCTo3h3+8Y+CdUu2L+EPzf5A3z/Vpe+fAhebMSZ4WBtBgHi9sGpV0UmrVZUlaUvocVqPwAVmjAk6lggC5Lff4PBh546gPFsztvL7kd/p2apn4AIzxgQdSwQBkpTkPBYuESxJWwJgicAYU6UsEQRIUpIz2uB55xWsW5K2hHB3OJ2adyr5jcYYU8ksEQRIUhKcdZYzM1GeJWlL6NayG6Hu0MAFZowJOpYIAiQpqWi1UK43l6U7ltLzNKsWMsZULUsEAXD4MGzeXDQRrEtfx5GcI9Y+YIypcpYIAmDNGmdYieIainu0sltHjTFVyxJBAJR0x1CjiEac1eSswARljAlalggCICnJGWcoKqpgXcL2BHqc1gOX2J/EGFO17KoTAElJTkeyvFnHjuYcJWlXkvUoNsYEhCWCKqZ6/B1Dy3cux6Meayg2xgSEJYIqlpYG+/YVHVrCehQbYwLJEkEVK66hOGF7Aq0btKZl/ZaBCcoYE9QsEVSxvERwbInA2geMMYFiiaCKJSVB27bQqJGzvPfoXjbt3WTVQsaYgLFEUMWObShOSEsArH3AGBM4lgiqUFYWrF9/fPuAIHRv2T1wgRljgpolgiq0fj14PMf3KD4n8hwaRjQMXGDGmKBmiaAKHXvHkKry07afqBNSh/iU+MAFZowJapYIqlBSEoSHQ/v2zvLU9VPZl7mPFTtX0PfjvpYMjDEBYYmgCiUlOTOShYQ4y1+v+xoARcn2ZBOXHBe44IwxQcuviUBEBorIBhHZJCKjS9jmOhFZKyJrRORTf8YTaMfeMRQREgGAW9yEucOIjYoNTGDGmKAW4q8di4gbeAvoD6QCCSIyTVXXFtqmPTAG6KWq+0Skub/iCbTdu2HnzqIdyfZl7qNV/Vb8rcffiI2KJaZNTOACNMYErTITgYgMAmaoqreC++4JbFLVLb79fA4MAdYW2uYu4C1V3QegqrsreIwaY9Uq57FwiWDlrpVc2PpCxlw8JjBBGWMM5asauh7YKCIvi0iHCuy7FZBSaDnVt66ws4GzReRnEVksIgOL25GIDBeRRBFJTE9Pr0AI1cexdwwdyj7E5r2b6dyic8lvMsaYKlBmIlDVm4FuwGbgQxGJ912Y61fC8UOA9kAscAPwnog0KiaGd1U1WlWjmzVrVgmHrXpJSdCiBTT3VX6t3r0aRenSoktgAzPGBL1yNRar6gHgK+BzoCVwJbBMREaV8rY0oE2h5da+dYWlAtNUNUdVfwN+xUkMtc6xDcUrd64EoMuplgiMMYFVZiIQkcEi8i0QB4QCPVX1UqAL8GApb00A2otIOxEJA4YC047ZZgpOaQARicSpKtpSwXOo9nJznQnrj20faBDegNMbnh64wIwxhvLdNXQ18JqqLii8UlWPiMgdJb1JVXNF5F7gB8ANfKCqa0TkGSBRVaf5XrtERNYCHuBhVd1zoidTXU2e7IwzVLduwbqVu1bSuUVnRCRwgRljDCCqWvoGIu2AHaqa6VuuA7RQ1WT/h3e86OhoTUxMDMShT0h8PMTGQna206t4/ny44EIvjcY2YliXYbz5lzcDHaIxJgiIyFJVjS7utfK0EUwGCt866vGtM+UQF+ckAXCqiOLiIHl/MgezD1pDsTGmWihPIghR1ey8Bd/zMP+FVLvktQuIQFiYUzqwhmJjTHVSnkSQLiKD8xZEZAjwu/9Cql3W+rrP3X8/zJ0LMTGQtCsJQTiv2XmBDc4YYyhfY/E9wCQReRMQnE5iw/waVS2hCuPHQ69e8OqrBetX7lpJ+6btqRdWL3DBGWOMT5mJQFU3AxeKyCm+5UN+j6qW+Pln2LABHnmk6PqVu1bajGTGmGqjXIPOichlwHlARN7tjqr6jB/jqhXGj4f69eHaawvWHcg6wJZ9W/hr178GLjBjjCmkPB3K/oMz3tAonKqhawHrBVWGAwfgyy9h6FA45ZSC9at3rwawMYaMMdVGeRqLL1LVYcA+VX0aiMHpAWxK8fnncOQI3Hln0fV2x5AxpropTyLI9D0eEZHTgByc8YZMKd5/Hzp2hB49iq5fuWsljSIa0aZBm+LfaIwxVaw8iWC6b0TQV4BlQDJQq2cSO1lJSZCQ4JQGjh1BYuWulXRp0cWGljDGVBulJgIRcQFzVXW/qn6N0zbQQVWfqJLoaqjx453OYzffXHS9V72s2rXK2geMMdVKqYnANyvZW4WWs1Q1w+9R1WBZWfDJJ3DlldC0adHXtuzbwuGcwza0hDGmWilP1dBcEblarC6jXKZMgb174Y5ixmW1hmJjTHVUnkRwN84gc1kickBEDorIAT/HVWO9/z6cfjr07Xv8ayt3rcQlLhtawhhTrZSnZ3FlTEkZFL7+GubMcRqJXcWk2KRdSZzd9GzqhNap+uCMMaYEZSYCEflTceuPnagm2MXHO53HwGkj+OtfnQHmClu5ayUXtLqg6vFqziMAAB1CSURBVIMzxphSlGeIiYcLPY8AegJLgT5+iaiGiotz5hsAyMlxlgsngozMDJL3JzP8/OGBCM8YY0pUnqqhQYWXRaQN8LrfIqqhYmML+gzkzTtQWNKuJMAaio0x1U95GouPlQqcW9mB1HTdujnDTvfpUzDvQGF5icD6EBhjqpvytBH8G8ib2NgFdMXpYWwK2bjRebzzzuOTADjtA03qNKFV/VZVG5gxxpShPG0EhWeKzwU+U9Wf/RRPjbVhg/N4zjnFv25DSxhjqqvyJIKvgExV9QCIiFtE6qrqEf+GVrOsX+88nl3MuKwer4dVu1Zxd/e7qzYoY4wph3L1LAYK3/heB5jjn3Bqrg0boE0bqFfM7JNfrf2Ko7lHqRtat+oDM8aYMpQnEUQUnp7S99yuaMdYvx46dDh+fXxKPMOmOFM8/zP+n8SnxFdxZMYYU7ryJILDInJ+3oKIdAeO+i+kmkfVKREU1z4QlxxHjicHgFxvLnHJcVUbnDHGlKE8bQT3A5NFZDvOVJWn4kxdaXx27ICDB4svEcRGxSIiqCph7jBio2KrPD5jjClNeTqUJYhIByDv++4GVc3xb1g1S2l3DF3Q+gIiQiLo2Kwjrw98nZg2xdxbaowxAVSeyev/BtRT1dWquho4RURG+j+0miMvERRXIvh1z68cyTnCPdH3WBIwxlRL5WkjuEtV9+ctqOo+4C7/hVTzrF/v3C3Uqpi+YglpCQD0aNXj+BeNMaYaKE8icBeelEZE3ECY/0KqefIaiovrK5awPYF6ofU4N9JG5TDGVE/lSQSzgC9EpK+I9AU+A773b1g1y/r1JfcoTtiewPktz8ftcldtUMYYU07lSQSPAPOAe3w/qyjawSyoHT0KW7cW3z6Q48lhxc4V9DjNqoWMMdVXmYnAN4H9L0AyzlwEfYB15dm5iAwUkQ0isklERpey3dUioiISXb6wq4+NG51+BMWVCFbvXk1mbqa1DxhjqrUSbx8VkbOBG3w/vwNfAKhq7/Ls2NeW8BbQH2fo6gQRmaaqa4/Zrj5wH06yqXFKu3U0YbuvodhKBMaYaqy0EsF6nG//l6vqH1X134CnAvvuCWxS1S2qmg18DgwpZrtngZeAzArsu9oobbC5JWlLaFKnCWc0PqNqgzLGmAooLRFcBewA5ovIe76G4oqModwKSCm0nOpbl883dEUbVZ1R2o5EZLiIJIpIYnp6egVC8L8NG6BtW6hbzOhLCdsTiD4t2oaeNsZUayUmAlWdoqpDgQ7AfJyhJpqLyDsicsnJHlhEXMCrwINlbauq76pqtKpGN2vW7GQPXalKGmzuSM4R1uxeY9VCxphqrzyNxYdV9VPf3MWtgeU4dxKVJQ1oU2i5tW9dnvpARyBORJKBC4FpNanBuLTB5pbvWI5HPfRs1bPqAzPGmAqo0JzFqrrP9+28bzk2TwDai0g7EQkDhgLTCu0rQ1UjVTVKVaOAxcBgVU0sfnfVz44dcOhQ8SUCayg2xtQUJzJ5fbmoai5wL/ADzu2mX6rqGhF5RkQG++u4VSmvobikO4Za1W9Fy/otqzYoY4ypoPIMQ33CVHUmMPOYdU+UsG2sP2Pxh9IGm0tIS7D+A8aYGsFvJYJgsH49nHIKnHZa0fX7ju5j496NVi1kjKkRLBGchJIGm0vc7jRzWCIwxtQElghOQkmDzeU1FEefVmNugDLGBDFLBCfoyBHYtq3kO4bOanIWjes0rvrAjDGmgiwRnKDSBptLSEuwaiFjTI1hieAElTTY3I6DO0g7mGaJwBhTY1giOEHr1zuNxO3bF12f1z5gPYqNMTWFJYITVNJgcwlpCbjFTbeW3QITmDHGVJAlghO0YUPxDcVLti/hvObnUTe0mOFIjTGmGrJEcAJKGmxOVUncnmjtA8aYGsUSwQnYvr34wea27NvC3qN7LREYY2oUSwRAfEo8Ly58kfiU+HKtL2mwuUmrJgEQHhLut1iNMaay+XXQuZpg4daF9P24L7neXEJcIYyIHkFUoyiS9yfzTuI7eNVLmDuMucPmEtMmBih+sLn4lHie/d+zAIycMZJzmp6Tv70xxlRnQZMI4uMhLg5iYyGm0PX5wxUfkrO1OyTHkhMVxxveNwpeTLkQkmPJareQuOS4/Av7/PkQFgbJyQUDzn219ityNReAbE92ke2NMaY6C4pEEB8PvXtDdja43XD99dDKN3vyL0vvhfkdwesGl4c+g9LpdGZTVm3ew7zpzcDrxuvyEPf7fvZ/B2lp8PXXToNxv34wdy5ceKGyYNsCANziJswdRmxUbOBO2BhjKiAoEkFcnJMEVCE3F774AkJ8Z56d0xm8LkDAKyyYcRqLQiA39zTwav76ed9FsiDEeb+q773Zzr5TGkwmcXsiD1z4AJF1I4mNirXSgDGmxgiKxuLYWIiIcEoDderAggVw9KjzM+jFcRCa6XtN8l9bsMBZdrsVQo7S99lnCq139hUWBj0uOsz9s+7n/Jbn83L/lxlz8RhLAsaYGiUoSgQxMU4VTnFtBDmtFtD+72u5vdFHRV4reI+wsf4kJux5ih82XciAmAFF9vVlxmPsPLSTKUOn4Ha5q/zcjDHmZInm1XPUENHR0ZqYWHnz23f9T1faNGzD9Buml7hNZm4mXf/TlSxPFqtHrKZeWD0AVuxcQfd3uzP8/OG8c/k7lRaTMcZUNhFZqqrFTpISFFVDpUk9kErr+q1L3SYiJIL3Br1H8v5knox7EgCvehk5YyRN6zTlhb4vVEWoxhjjF0FRNVSSzNxM9hzdQ6sGrcrc9uLTL+bu7nfz2uLXuKHjDSzfuZz41Hg+uuIjm4DGGFOjBXUiSDuQBkDrBqWXCPK81O8lpm2YxnVfXceOgzvo0qILt3S+xZ8hGmOM3wV11VDqgVQAWtUvu0QA0DCiIaN6jmLLvi0czT3Khj0bWJy62J8hGmOM3wV1Ikg7WLESQR5BAMjx5BCXHFfZYRljTJUK6kSQVyKoSCKIjYolIiTCehAbY2qNoG8jqB9Wn/rh9cv9npg2McwdNpe45DjrQWyMqRWCOhGkHkytcLUQOMnAEoAxprYI+qqhE0kExhhTmwR1Ikg7kFauPgTGGFObBW0iyPXmsuPQjjJ7FRtjTG0XtIlg16FdeNVrJQJjTNAL2kRwIreOGmNMbeTXRCAiA0Vkg4hsEpHRxbz+gIisFZEkEZkrIqf7M57CTrQzmTHG1DZ+SwQi4gbeAi4F/gDcICJ/OGaz5UC0qnYGvgJe9lc8x6ro8BLGGFNb+bNE0BPYpKpbVDUb+BwYUngDVZ2vqkd8i4uBKvt6nnoglTB3GJF1I6vqkMYYUy35MxG0AlIKLaf61pXkDuB7P8ZTRNrBNFrVb4WIVNUhjTGmWqoWPYtF5GYgGvhzCa8PB4YDtG3btlKOaZ3JjDHG4c8SQRrQptBya9+6IkSkH/AoMFhVs4rbkaq+q6rRqhrdrFmzygnuQJolAmOMwb+JIAFoLyLtRCQMGApMK7yBiHQD/ouTBHb7MZYiVJXUA6nWUGyMMfgxEahqLnAv8AOwDvhSVdeIyDMiMti32SvAKcBkEVkhItNK2F2l2nN0D1meLCsRGGMMfm4jUNWZwMxj1j1R6Hk/fx6/JBWdotIYY2qzoOxZnN+HwIaXMMaY4EwE1qvYGGMKVIvbR6ta6oFUXOLi1FNODXQoxlSZnJwcUlNTyczMDHQoxo8iIiJo3bo1oaGh5X5P0CaCU085lRBXUJ6+CVKpqanUr1+fqKgo60hZS6kqe/bsITU1lXbt2pX7fUFbNWTVQibYZGZm0rRpU0sCtZiI0LRp0wqX+oIyEVgfAhOsLAnUfifyNw7KRGC9io0xpkDQJYJD2YfIyMqwEoExVWz//v28/fbbJ/Tev/zlL+zfv7/UbZ544gnmzJlzQvsPdkGXCKwzmTHlF58Sz4sLXyQ+Jf6k91VaIsjNzS31vTNnzqRRo0albvPMM8/Qr19A+qiesLLOu6oE3W0zNkWlMXD/rPtZsXNFqdtkZGWQtCsJr3pxiYvOLTrTMLxhidt3PbUrrw98vcTXR48ezebNm+natSv9+/fnsssu4/HHH6dx48asX7+eX3/9lSuuuIKUlBQyMzO57777GD58OABRUVEkJiZy6NAhLr30Uv74xz+yaNEiWrVqxdSpU6lTpw633XYbl19+Oddccw1RUVHceuutTJ8+nZycHCZPnkyHDh1IT0/nxhtvZPv27cTExPDjjz+ydOlSIiOLzksyYsQIEhISOHr0KNdccw1PP/00AAkJCdx3330cPnyY8PBw5s6dS926dXnkkUeYNWsWLpeLu+66i1GjRuXHHBkZSWJiIg899BBxcXE89dRTbN68mS1bttC2bVtefPFFbrnlFg4fPgzAm2++yUUXXQTASy+9xCeffILL5eLSSy/lrrvu4tprr2XZsmUAbNy4keuvvz5/+UQFbSKwXsXGlC4jMwOvegHwqpeMzIxSE0FZxo4dy+rVq1mxwklAcXFxLFu2jNWrV+ff6vjBBx/QpEkTjh49So8ePbj66qtp2rRpkf1s3LiRzz77jPfee4/rrruOr7/+mptvvvm440VGRrJs2TLefvttxo0bx/vvv8/TTz9Nnz59GDNmDLNmzWL8+PHFxvr888/TpEkTPB4Pffv2JSkpiQ4dOnD99dfzxRdf0KNHDw4cOECdOnV49913SU5OZsWKFYSEhLB3794yfxdr167lp59+ok6dOhw5coQff/yRiIgINm7cyA033EBiYiLff/89U6dO5ZdffqFu3brs3buXJk2a0LBhQ1asWEHXrl2ZMGECt99+e0X/FMcJukSQ16vY2ghMMCvtm3ue+JR4+n7cl2xPNmHuMCZdNYmYNjGVGkfPnj2L3O/+xhtv8O233wKQkpLCxo0bj0sE7dq1o2vXrgB0796d5OTkYvd91VVX5W/zzTffAPDTTz/l73/gwIE0bty42Pd++eWXvPvuu+Tm5rJjxw7Wrl2LiNCyZUt69OgBQIMGDQCYM2cO99xzDyEhzuW0SZMmZZ734MGDqVOnDuB09Lv33ntZsWIFbrebX3/9NX+/t99+O3Xr1i2y3zvvvJMJEybw6quv8sUXX7BkyZIyj1eWoEsEqQdSaVqnKXVC6wQ6FGOqtZg2McwdNpe45Dhio2IrPQkA1KtXL/95XFwcc+bMIT4+nrp16xIbG1vs/fDh4eH5z91uN0ePHi1233nbud3uCtXF//bbb4wbN46EhAQaN27MbbfddkK9sUNCQvB6nRLVse8vfN6vvfYaLVq0YOXKlXi9XiIiIkrd79VXX51fsunevftxifJEBF9j8cE0qxYyppxi2sQw5uIxlZIE6tevz8GDB0t8PSMjg8aNG1O3bl3Wr1/P4sWLT/qYx+rVqxdffvklALNnz2bfvn3HbXPgwAHq1atHw4YN2bVrF99/78yge84557Bjxw4SEhIAOHjwILm5ufTv35///ve/+ckmr2ooKiqKpUuXAvD111+XGFNGRgYtW7bE5XIxceJEPB4PAP3792fChAkcOXKkyH4jIiIYMGAAI0aMqJRqIQjCRGBTVBoTGE2bNqVXr1507NiRhx9++LjXBw4cSG5uLueeey6jR4/mwgsvrPQYnnzySWbPnk3Hjh2ZPHkyp556KvXr1y+yTZcuXejWrRsdOnTgxhtvpFevXgCEhYXxxRdfMGrUKLp06UL//v3JzMzkzjvvpG3btnTu3JkuXbrw6aef5h/rvvvuIzo6GrfbXWJMI0eO5KOPPqJLly6sX78+v7QwcOBABg8eTHR0NF27dmXcuHH577nppptwuVxccskllfJ7EVWtlB1VlejoaE1MTDzh97cY14Ih5wzh3UHvVmJUxlR/69at49xzzw10GAGVlZWF2+0mJCSE+Ph4RowYkd94XZOMGzeOjIwMnn322WJfL+5vLSJLVTW6uO2Dqo0g25PN7sO7rURgTJDatm0b1113HV6vl7CwMN57771Ah1RhV155JZs3b2bevHmVts+gSgTbD24HrA+BMcGqffv2LF++PNBhnJS8u54qU1C1EeT1KrZbR40xpkBQJQLrVWyMMccLykRgt48aY0yBoEoEaQfTqBda76S6yRtjTG0TVIkgrw+BTc5hTNU7mWGoAV5//fX8zlWmcgVVIrBexcZUTHw8vPii83iyakMiqC7DRle2oLp9NPVAKrFRsYEOw5iAu/9+KKsfVUYGJCWB1wsuF3TuDA1LqVXt2hVeL2Usu2OHoX7llVd45ZVX+PLLL8nKyuLKK6/k6aef5vDhw1x33XWkpqbi8Xh4/PHH2bVrF9u3b6d3795ERkYyf/78Ivt+5plnmD59OkePHuWiiy7iv//9LyLCpk2buOeee0hPT8ftdjN58mTOPPPM44Z3Hjt2LLGxsYwbN47o6Gh+//13oqOjSU5O5sMPP+Sbb77h0KFDeDweZsyYwZAhQ9i3bx85OTk899xzDBkyBICPP/6YcePGISJ07tyZt99+m86dO/Prr78SGhrKgQMH6NKlS/5ydRE0icCrXrYf3E7r+nbHkDHlkZHhJAFwHjMySk8EZTl2GOrZs2ezceNGlixZgqoyePBgFixYQHp6OqeddhozZszwxZFBw4YNefXVV5k/f/5xcwcA3HvvvTzxxBMA3HLLLXz33XcMGjSIm266idGjR3PllVeSmZmJ1+stdnjnsixbtoykpCSaNGlCbm4u3377LQ0aNOD333/nwgsvZPDgwaxdu5bnnnuORYsWERkZyd69e6lfvz6xsbHMmDGDK664gs8//5yrrrqqWiUBCKJEsPvwbnK9uVY1ZAylf3PPEx8PfftCdjaEhcGkSRBTiQOQzp49m9mzZ9OtWzcADh06xMaNG7n44ot58MEHeeSRR7j88su5+OKLy9zX/Pnzefnllzly5Ah79+7lvPPOIzY2lrS0NK688kqA/FE9SxreuTT9+/fP305V+cc//sGCBQtwuVykpaWxa9cu5s2bx7XXXpufqAoPG/3yyy9zxRVXMGHChGrZmzloEoH1ITCmYmJiYO5ciIuD2NjKTQLgXFDHjBnD3Xfffdxry5YtY+bMmTz22GP07ds3/9t+cTIzMxk5ciSJiYm0adOGp556yq/DRk+aNIn09HSWLl1KaGgoUVFRpR6vV69eJCcnExcXh8fjoWPHjhWOzd+CprF47pa5AOw5sifAkRhTc8TEwJgxlZMEjh2GesCAAXzwwQccOnQIgLS0NHbv3s327dupW7cuN998Mw8//HD+NIwlDWOddxGOjIzk0KFDfPXVV/nbt27dmilTpgDOgHNHjhwpcXjnwsNG5+2jOBkZGTRv3pzQ0FDmz5/P1q1bAejTpw+TJ09mz549RfYLMGzYMG688cZKGza6sgVFIohPiefx+Y8DMHLmyEqZiNsYUzHHDkN9ySWXcOONNxITE0OnTp245pprOHjwIKtWraJnz5507dqVp59+msceewyA4cOHM3DgQHr37l1kv40aNeKuu+6iY8eODBgwIH8GMYCJEyfyxhtv0LlzZy666CJ27txZ4vDODz30EO+88w7dunXj999/L/E8brrpJhITE+nUqRMff/wxHTp0AOC8887j0Ucf5c9//jNdunThgQceKPKeffv2ccMNN1Ta77MyBcUw1C8ufJHH5j2GFy9ucfNs72cZc/EYP0VoTPVkw1AHzldffcXUqVOZOHFilRzPhqEuRmxULOEh4flzr9otpMaYqjJq1Ci+//57Zs6cGehQShQUiaAq5l41xpji/Pvf/w50CGXyayIQkYHAvwA38L6qjj3m9XDgY6A7sAe4XlWT/RFLTJsYSwAm6KmqDbFSy51Idb/fGotFxA28BVwK/AG4QUT+cMxmdwD7VPUs4DXgJX/FY0ywi4iIYM+ePSd0oTA1g6qyZ8+e/D4T5eXPEkFPYJOqbgEQkc+BIcDaQtsMAZ7yPf8KeFNERO2Takyla926NampqaSnpwc6FONHERERtG5dsf5S/kwErYCUQsupwAUlbaOquSKSATQFity7JSLDgeEAbdu29Ve8xtRqoaGhtGvXLtBhmGqoRvQjUNV3VTVaVaObNWsW6HCMMaZW8WciSAPaFFpu7VtX7DYiEgI0xGk0NsYYU0X8mQgSgPYi0k5EwoChwLRjtpkG3Op7fg0wz9oHjDGmavm1Z7GI/AV4Hef20Q9U9XkReQZIVNVpIhIBTAS6AXuBoXmNy6XsMx3YeoIhRXJM+0OQCNbzhuA9dzvv4FKe8z5dVYutW69xQ0ycDBFJLKmLdW0WrOcNwXvudt7B5WTPu0Y0FhtjjPEfSwTGGBPkgi0RvBvoAAIkWM8bgvfc7byDy0mdd1C1ERhjjDlesJUIjDHGHMMSgTHGBLmgSQQiMlBENojIJhEZHeh4/EVEPhCR3SKyutC6JiLyo4hs9D02DmSM/iAibURkvoisFZE1InKfb32tPncRiRCRJSKy0nfeT/vWtxORX3yf9y98nTprHRFxi8hyEfnOt1zrz1tEkkVklYisEJFE37qT+pwHRSIo55DYtcWHwMBj1o0G5qpqe2Cub7m2yQUeVNU/ABcCf/P9jWv7uWcBfVS1C9AVGCgiF+IM6f6ab4j3fThDvtdG9wHrCi0Hy3n3VtWuhfoOnNTnPCgSAYWGxFbVbCBvSOxaR1UX4PTSLmwI8JHv+UfAFVUaVBVQ1R2qusz3/CDOxaEVtfzc1XHItxjq+1GgD87Q7lALzxtARFoDlwHv+5aFIDjvEpzU5zxYEkFxQ2K3ClAsgdBCVXf4nu8EWgQyGH8TkSicYUt+IQjO3Vc9sgLYDfwIbAb2q2qub5Pa+nl/Hfh/gNe33JTgOG8FZovIUt8Q/XCSn/OgmLPYFFBVFZFae8+wiJwCfA3cr6oHCk/LWFvPXVU9QFcRaQR8C3QIcEh+JyKXA7tVdamIxAY6nir2R1VNE5HmwI8isr7wiyfyOQ+WEkF5hsSuzXaJSEsA3+PuAMfjFyISipMEJqnqN77VQXHuAKq6H5gPxACNfEO7Q+38vPcCBotIMk5Vbx+c+dFr+3mjqmm+x904ib8nJ/k5D5ZEUJ4hsWuzwsN93wpMDWAsfuGrHx4PrFPVVwu9VKvPXUSa+UoCiEgdoD9O+8h8nKHdoRaet6qOUdXWqhqF8/88T1Vvopaft4jUE5H6ec+BS4DVnOTnPGh6Fhc3JHaAQ/ILEfkMiMUZlnYX8CQwBfgSaIszhPd1qnpsg3KNJiJ/BBYCqyioM/4HTjtBrT13EemM0zjoxvli96WqPiMiZ+B8U24CLAduVtWswEXqP76qoYdU9fLaft6+8/vWtxgCfOob3r8pJ/E5D5pEYIwxpnjBUjVkjDGmBJYIjDEmyFkiMMaYIGeJwBhjgpwlAmOMCXKWCIzxERGPb0THvJ9KG6BORKIKjwhrTHViQ0wYU+CoqnYNdBDGVDUrERhTBt/47y/7xoBfIiJn+dZHicg8EUkSkbki0ta3voWIfOubI2CliFzk25VbRN7zzRsw29cTGBH5P988Ckki8nmATtMEMUsExhSoc0zV0PWFXstQ1U7Amzg91AH+DXykqp2BScAbvvVvAP/zzRFwPrDGt7498JaqngfsB672rR8NdPPt5x5/nZwxJbGexcb4iMghVT2lmPXJOJO/bPENbLdTVZuKyO9AS1XN8a3foaqRIpIOtC48tIFvaOwffROHICKPAKGq+pyIzAIO4QwFMqXQ/ALGVAkrERhTPlrC84ooPOaNh4I2ustwZtA7H0goNHqmMVXCEoEx5XN9ocd43/NFOCNfAtyEM+gdOFMFjoD8SWMalrRTEXEBbVR1PvAI0BA4rlRijD/ZNw9jCtTxzfSVZ5aq5t1C2lhEknC+1d/gWzcKmCAiDwPpwO2+9fcB74rIHTjf/EcAOyieG/jElywEeMM3r4AxVcbaCIwpg6+NIFpVfw90LMb4g1UNGWNMkLMSgTHGBDkrERhjTJCzRGCMMUHOEoExxgQ5SwTGGBPkLBEYY0yQ+//O0Nmqzu7w6gAAAABJRU5ErkJggg==\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plot_model(model4.model, to_file='/content/GRU_plot.png')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 758
        },
        "id": "kydczahcLk6m",
        "outputId": "9b929906-1254-46ee-ffcb-b44a42ba13a7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<IPython.core.display.Image object>"
            ],
            "image/png": "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\n"
          },
          "metadata": {},
          "execution_count": 51
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Model 2**"
      ],
      "metadata": {
        "id": "7U606x3cGDUv"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "with tf.device('/GPU:0'):\n",
        "  model5 = RNN(X_train_list, y_train, model_num=5)\n",
        "  model5.train((X_test_list, y_test))\n",
        "  model5.evaluate(X_test_list, y_test)\n",
        "  model5.outputResult()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "7SedfjCOGK09",
        "outputId": "e7821242-d25e-40fd-d137-0cd4abaf29ca"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "78/78 - 4s - loss: 0.5195 - accuracy: 0.0000e+00 - val_loss: 0.4593 - val_accuracy: 0.0000e+00 - 4s/epoch - 55ms/step\n",
            "Epoch 2/50\n",
            "78/78 - 1s - loss: 0.4408 - accuracy: 0.2313 - val_loss: 0.4235 - val_accuracy: 0.3871 - 917ms/epoch - 12ms/step\n",
            "Epoch 3/50\n",
            "78/78 - 1s - loss: 0.3940 - accuracy: 0.5474 - val_loss: 0.3740 - val_accuracy: 0.5887 - 933ms/epoch - 12ms/step\n",
            "Epoch 4/50\n",
            "78/78 - 1s - loss: 0.3524 - accuracy: 0.6770 - val_loss: 0.3502 - val_accuracy: 0.7065 - 906ms/epoch - 12ms/step\n",
            "Epoch 5/50\n",
            "78/78 - 1s - loss: 0.3229 - accuracy: 0.7885 - val_loss: 0.3279 - val_accuracy: 0.7903 - 951ms/epoch - 12ms/step\n",
            "Epoch 6/50\n",
            "78/78 - 1s - loss: 0.2996 - accuracy: 0.8514 - val_loss: 0.3140 - val_accuracy: 0.8500 - 915ms/epoch - 12ms/step\n",
            "Epoch 7/50\n",
            "78/78 - 1s - loss: 0.2801 - accuracy: 0.8966 - val_loss: 0.3106 - val_accuracy: 0.8613 - 1s/epoch - 13ms/step\n",
            "Epoch 8/50\n",
            "78/78 - 1s - loss: 0.2660 - accuracy: 0.9237 - val_loss: 0.3196 - val_accuracy: 0.8516 - 822ms/epoch - 11ms/step\n",
            "Epoch 9/50\n",
            "78/78 - 1s - loss: 0.2569 - accuracy: 0.9346 - val_loss: 0.3123 - val_accuracy: 0.8548 - 841ms/epoch - 11ms/step\n",
            "Epoch 10/50\n",
            "78/78 - 1s - loss: 0.2393 - accuracy: 0.9548 - val_loss: 0.3119 - val_accuracy: 0.8532 - 816ms/epoch - 10ms/step\n",
            "Epoch 11/50\n",
            "78/78 - 1s - loss: 0.2301 - accuracy: 0.9560 - val_loss: 0.3111 - val_accuracy: 0.8435 - 847ms/epoch - 11ms/step\n",
            "Epoch 12/50\n",
            "78/78 - 1s - loss: 0.2221 - accuracy: 0.9621 - val_loss: 0.3283 - val_accuracy: 0.8371 - 826ms/epoch - 11ms/step\n",
            "Epoch 13/50\n",
            "78/78 - 1s - loss: 0.2177 - accuracy: 0.9548 - val_loss: 0.3352 - val_accuracy: 0.8290 - 816ms/epoch - 10ms/step\n",
            "Epoch 14/50\n",
            "78/78 - 1s - loss: 0.2002 - accuracy: 0.9669 - val_loss: 0.3483 - val_accuracy: 0.8290 - 825ms/epoch - 11ms/step\n",
            "Epoch 15/50\n",
            "78/78 - 1s - loss: 0.1910 - accuracy: 0.9685 - val_loss: 0.3482 - val_accuracy: 0.8177 - 817ms/epoch - 10ms/step\n",
            "Epoch 16/50\n",
            "78/78 - 1s - loss: 0.1806 - accuracy: 0.9661 - val_loss: 0.3653 - val_accuracy: 0.8339 - 807ms/epoch - 10ms/step\n",
            "Epoch 17/50\n",
            "78/78 - 1s - loss: 0.1741 - accuracy: 0.9637 - val_loss: 0.3719 - val_accuracy: 0.8226 - 812ms/epoch - 10ms/step\n",
            "Epoch 18/50\n",
            "78/78 - 1s - loss: 0.1590 - accuracy: 0.9750 - val_loss: 0.3986 - val_accuracy: 0.7935 - 782ms/epoch - 10ms/step\n",
            "Epoch 19/50\n",
            "78/78 - 1s - loss: 0.1467 - accuracy: 0.9705 - val_loss: 0.4174 - val_accuracy: 0.8242 - 801ms/epoch - 10ms/step\n",
            "Epoch 20/50\n",
            "78/78 - 1s - loss: 0.1372 - accuracy: 0.9600 - val_loss: 0.4583 - val_accuracy: 0.8290 - 815ms/epoch - 10ms/step\n",
            "Epoch 21/50\n",
            "78/78 - 1s - loss: 0.1368 - accuracy: 0.9491 - val_loss: 0.4696 - val_accuracy: 0.7710 - 823ms/epoch - 11ms/step\n",
            "Epoch 22/50\n",
            "78/78 - 1s - loss: 0.1239 - accuracy: 0.9512 - val_loss: 0.4663 - val_accuracy: 0.8016 - 821ms/epoch - 11ms/step\n",
            "Epoch 23/50\n",
            "78/78 - 1s - loss: 0.1142 - accuracy: 0.9556 - val_loss: 0.5060 - val_accuracy: 0.7710 - 815ms/epoch - 10ms/step\n",
            "Epoch 24/50\n",
            "78/78 - 1s - loss: 0.1015 - accuracy: 0.9451 - val_loss: 0.5374 - val_accuracy: 0.7645 - 796ms/epoch - 10ms/step\n",
            "Epoch 25/50\n",
            "78/78 - 1s - loss: 0.0945 - accuracy: 0.9495 - val_loss: 0.5523 - val_accuracy: 0.7403 - 801ms/epoch - 10ms/step\n",
            "Epoch 26/50\n",
            "78/78 - 1s - loss: 0.0870 - accuracy: 0.9415 - val_loss: 0.5917 - val_accuracy: 0.7290 - 821ms/epoch - 11ms/step\n",
            "Epoch 27/50\n",
            "78/78 - 1s - loss: 0.0846 - accuracy: 0.9306 - val_loss: 0.6109 - val_accuracy: 0.7645 - 800ms/epoch - 10ms/step\n",
            "Epoch 28/50\n",
            "78/78 - 1s - loss: 0.0790 - accuracy: 0.9338 - val_loss: 0.6193 - val_accuracy: 0.7710 - 800ms/epoch - 10ms/step\n",
            "Epoch 29/50\n",
            "78/78 - 1s - loss: 0.0711 - accuracy: 0.9338 - val_loss: 0.6507 - val_accuracy: 0.7581 - 816ms/epoch - 10ms/step\n",
            "Epoch 30/50\n",
            "78/78 - 1s - loss: 0.0617 - accuracy: 0.9382 - val_loss: 0.6837 - val_accuracy: 0.7645 - 801ms/epoch - 10ms/step\n",
            "Epoch 31/50\n",
            "78/78 - 1s - loss: 0.0591 - accuracy: 0.9334 - val_loss: 0.6935 - val_accuracy: 0.7597 - 825ms/epoch - 11ms/step\n",
            "Epoch 32/50\n",
            "78/78 - 1s - loss: 0.0702 - accuracy: 0.9092 - val_loss: 0.7378 - val_accuracy: 0.7387 - 803ms/epoch - 10ms/step\n",
            "Epoch 33/50\n",
            "78/78 - 1s - loss: 0.0608 - accuracy: 0.9092 - val_loss: 0.7355 - val_accuracy: 0.7613 - 806ms/epoch - 10ms/step\n",
            "Epoch 34/50\n",
            "78/78 - 1s - loss: 0.0478 - accuracy: 0.9298 - val_loss: 0.7819 - val_accuracy: 0.7355 - 800ms/epoch - 10ms/step\n",
            "Epoch 35/50\n",
            "78/78 - 1s - loss: 0.0442 - accuracy: 0.9302 - val_loss: 0.7902 - val_accuracy: 0.7565 - 798ms/epoch - 10ms/step\n",
            "Epoch 36/50\n",
            "78/78 - 1s - loss: 0.0408 - accuracy: 0.9047 - val_loss: 0.8283 - val_accuracy: 0.7371 - 794ms/epoch - 10ms/step\n",
            "Epoch 37/50\n",
            "78/78 - 1s - loss: 0.0370 - accuracy: 0.9221 - val_loss: 0.8715 - val_accuracy: 0.7613 - 790ms/epoch - 10ms/step\n",
            "Epoch 38/50\n",
            "78/78 - 1s - loss: 0.0366 - accuracy: 0.9080 - val_loss: 0.8780 - val_accuracy: 0.7403 - 817ms/epoch - 10ms/step\n",
            "Epoch 39/50\n",
            "78/78 - 1s - loss: 0.0326 - accuracy: 0.9140 - val_loss: 0.8889 - val_accuracy: 0.7403 - 794ms/epoch - 10ms/step\n",
            "Epoch 40/50\n",
            "78/78 - 1s - loss: 0.0323 - accuracy: 0.9071 - val_loss: 0.9098 - val_accuracy: 0.7435 - 824ms/epoch - 11ms/step\n",
            "Epoch 41/50\n",
            "78/78 - 1s - loss: 0.0265 - accuracy: 0.9124 - val_loss: 0.9404 - val_accuracy: 0.7597 - 814ms/epoch - 10ms/step\n",
            "Epoch 42/50\n",
            "78/78 - 1s - loss: 0.0575 - accuracy: 0.8813 - val_loss: 0.8180 - val_accuracy: 0.7774 - 872ms/epoch - 11ms/step\n",
            "Epoch 43/50\n",
            "78/78 - 1s - loss: 0.0659 - accuracy: 0.8740 - val_loss: 0.7938 - val_accuracy: 0.7565 - 809ms/epoch - 10ms/step\n",
            "Epoch 44/50\n",
            "78/78 - 1s - loss: 0.0353 - accuracy: 0.8918 - val_loss: 0.8668 - val_accuracy: 0.7516 - 809ms/epoch - 10ms/step\n",
            "Epoch 45/50\n",
            "78/78 - 1s - loss: 0.0252 - accuracy: 0.9035 - val_loss: 0.9141 - val_accuracy: 0.7661 - 799ms/epoch - 10ms/step\n",
            "Epoch 46/50\n",
            "78/78 - 1s - loss: 0.0213 - accuracy: 0.8914 - val_loss: 0.9766 - val_accuracy: 0.7484 - 808ms/epoch - 10ms/step\n",
            "Epoch 47/50\n",
            "78/78 - 1s - loss: 0.0181 - accuracy: 0.8991 - val_loss: 1.0201 - val_accuracy: 0.7145 - 786ms/epoch - 10ms/step\n",
            "Epoch 48/50\n",
            "78/78 - 1s - loss: 0.0178 - accuracy: 0.8849 - val_loss: 1.0394 - val_accuracy: 0.7323 - 811ms/epoch - 10ms/step\n",
            "Epoch 49/50\n",
            "78/78 - 1s - loss: 0.0172 - accuracy: 0.8789 - val_loss: 1.0564 - val_accuracy: 0.7306 - 1s/epoch - 14ms/step\n",
            "Epoch 50/50\n",
            "78/78 - 1s - loss: 0.0147 - accuracy: 0.8930 - val_loss: 1.1059 - val_accuracy: 0.7242 - 1s/epoch - 13ms/step\n",
            "\n",
            "best model based on min training set loss:  acc= 0.8930157423019409  loss= 0.014732182025909424\n",
            "best model based on min test set loss:  acc= 0.8612903356552124  loss= 0.3105645477771759\n",
            "best model based on max training set accuracy:  acc= 0.974969744682312  loss= 0.15904803574085236\n",
            "best model based on max test set accuracy:  acc= 0.8612903356552124  loss= 0.3105645477771759\n",
            "\n",
            "evaluate dataset with best model based on maximum test set accuracy\n",
            "evaluate train set=  [0.2701292335987091, 0.9228906035423279]\n",
            "evaluate test set=  [0.3105645477771759, 0.8612903356552124]\n",
            "\n",
            " ------------------------------ metrics for traning set ------------------------------\n",
            "confusion matrix: \n",
            " [[[1894   58]\n",
            "  [   4  521]]\n",
            "\n",
            " [[2144   18]\n",
            "  [   8  307]]\n",
            "\n",
            " [[2389    7]\n",
            "  [  22   59]]\n",
            "\n",
            " [[2128   15]\n",
            "  [  10  324]]\n",
            "\n",
            " [[1985   32]\n",
            "  [  32  428]]\n",
            "\n",
            " [[1656   59]\n",
            "  [ 113  649]]\n",
            "\n",
            " [[1912  224]\n",
            "  [ 228  113]]\n",
            "\n",
            " [[1507  452]\n",
            "  [ 382  136]]\n",
            "\n",
            " [[1505  337]\n",
            "  [ 447  188]]\n",
            "\n",
            " [[1691  334]\n",
            "  [ 307  145]]\n",
            "\n",
            " [[1594  352]\n",
            "  [ 335  196]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.900     0.992     0.944       525\n",
            "  Women/Girl      0.945     0.975     0.959       315\n",
            "       Child      0.894     0.728     0.803        81\n",
            "      Animal      0.956     0.970     0.963       334\n",
            "       Other      0.930     0.930     0.930       460\n",
            "          NA      0.917     0.852     0.883       762\n",
            "      Score1      0.335     0.331     0.333       341\n",
            "      Score2      0.231     0.263     0.246       518\n",
            "      Score3      0.358     0.296     0.324       635\n",
            "      Score4      0.303     0.321     0.311       452\n",
            "      Score5      0.358     0.369     0.363       531\n",
            "\n",
            "   micro avg      0.619     0.619     0.619      4954\n",
            "   macro avg      0.648     0.639     0.642      4954\n",
            "weighted avg      0.621     0.619     0.619      4954\n",
            " samples avg      0.619     0.619     0.619      4954\n",
            "\n",
            "------------------------------ metrics for test set ------------------------------\n",
            "confusion matrix: \n",
            " [[[454  22]\n",
            "  [  3 141]]\n",
            "\n",
            " [[541   4]\n",
            "  [  2  73]]\n",
            "\n",
            " [[599   3]\n",
            "  [  7  11]]\n",
            "\n",
            " [[518  12]\n",
            "  [  9  81]]\n",
            "\n",
            " [[494  18]\n",
            "  [ 11  97]]\n",
            "\n",
            " [[409  26]\n",
            "  [ 53 132]]\n",
            "\n",
            " [[476  51]\n",
            "  [ 69  24]]\n",
            "\n",
            " [[360 114]\n",
            "  [111  35]]\n",
            "\n",
            " [[368 100]\n",
            "  [121  31]]\n",
            "\n",
            " [[420 102]\n",
            "  [ 73  25]]\n",
            "\n",
            " [[375 114]\n",
            "  [107  24]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.865     0.979     0.919       144\n",
            "  Women/Girl      0.948     0.973     0.961        75\n",
            "       Child      0.786     0.611     0.688        18\n",
            "      Animal      0.871     0.900     0.885        90\n",
            "       Other      0.843     0.898     0.870       108\n",
            "          NA      0.835     0.714     0.770       185\n",
            "      Score1      0.320     0.258     0.286        93\n",
            "      Score2      0.235     0.240     0.237       146\n",
            "      Score3      0.237     0.204     0.219       152\n",
            "      Score4      0.197     0.255     0.222        98\n",
            "      Score5      0.174     0.183     0.178       131\n",
            "\n",
            "   micro avg      0.544     0.544     0.544      1240\n",
            "   macro avg      0.574     0.565     0.567      1240\n",
            "weighted avg      0.545     0.544     0.542      1240\n",
            " samples avg      0.544     0.544     0.544      1240\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **LSTM Models**"
      ],
      "metadata": {
        "id": "TW2MfzTBW_Hl"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Model 1**"
      ],
      "metadata": {
        "id": "V2mzDLZJLGxo"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "with tf.device('/GPU:0'):\n",
        "  model1 = RNN(X_train_list, y_train, model_num=1)\n",
        "  model1.train((X_test_list, y_test))\n",
        "  model1.evaluate(X_test_list, y_test)\n",
        "  model1.outputResult()"
      ],
      "metadata": {
        "id": "He21UkQqLGOT",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5efc350a-dea3-4c26-8998-5acecef7204d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "78/78 - 8s - loss: 0.5292 - accuracy: 0.0852 - val_loss: 0.4390 - val_accuracy: 0.0000e+00 - 8s/epoch - 97ms/step\n",
            "Epoch 2/50\n",
            "78/78 - 1s - loss: 0.4179 - accuracy: 0.1332 - val_loss: 0.3869 - val_accuracy: 0.2081 - 1s/epoch - 16ms/step\n",
            "Epoch 3/50\n",
            "78/78 - 1s - loss: 0.3855 - accuracy: 0.2939 - val_loss: 0.3667 - val_accuracy: 0.4742 - 1s/epoch - 16ms/step\n",
            "Epoch 4/50\n",
            "78/78 - 1s - loss: 0.3490 - accuracy: 0.5591 - val_loss: 0.3452 - val_accuracy: 0.6274 - 1s/epoch - 16ms/step\n",
            "Epoch 5/50\n",
            "78/78 - 1s - loss: 0.3351 - accuracy: 0.6455 - val_loss: 0.3426 - val_accuracy: 0.4984 - 1s/epoch - 15ms/step\n",
            "Epoch 6/50\n",
            "78/78 - 1s - loss: 0.3318 - accuracy: 0.6395 - val_loss: 0.3283 - val_accuracy: 0.6806 - 1s/epoch - 16ms/step\n",
            "Epoch 7/50\n",
            "78/78 - 1s - loss: 0.3107 - accuracy: 0.7505 - val_loss: 0.3181 - val_accuracy: 0.7339 - 1s/epoch - 16ms/step\n",
            "Epoch 8/50\n",
            "78/78 - 1s - loss: 0.3069 - accuracy: 0.7671 - val_loss: 0.3174 - val_accuracy: 0.7339 - 1s/epoch - 16ms/step\n",
            "Epoch 9/50\n",
            "78/78 - 1s - loss: 0.2999 - accuracy: 0.7784 - val_loss: 0.3183 - val_accuracy: 0.6790 - 1s/epoch - 16ms/step\n",
            "Epoch 10/50\n",
            "78/78 - 1s - loss: 0.3064 - accuracy: 0.7489 - val_loss: 0.3151 - val_accuracy: 0.7581 - 1s/epoch - 16ms/step\n",
            "Epoch 11/50\n",
            "78/78 - 1s - loss: 0.2896 - accuracy: 0.8123 - val_loss: 0.3072 - val_accuracy: 0.7742 - 1s/epoch - 19ms/step\n",
            "Epoch 12/50\n",
            "78/78 - 1s - loss: 0.2806 - accuracy: 0.8385 - val_loss: 0.3072 - val_accuracy: 0.7726 - 1s/epoch - 16ms/step\n",
            "Epoch 13/50\n",
            "78/78 - 1s - loss: 0.2753 - accuracy: 0.8446 - val_loss: 0.3046 - val_accuracy: 0.7661 - 1s/epoch - 15ms/step\n",
            "Epoch 14/50\n",
            "78/78 - 1s - loss: 0.2663 - accuracy: 0.8611 - val_loss: 0.2898 - val_accuracy: 0.8210 - 1s/epoch - 16ms/step\n",
            "Epoch 15/50\n",
            "78/78 - 1s - loss: 0.2570 - accuracy: 0.8797 - val_loss: 0.2863 - val_accuracy: 0.8065 - 1s/epoch - 15ms/step\n",
            "Epoch 16/50\n",
            "78/78 - 1s - loss: 0.2583 - accuracy: 0.8668 - val_loss: 0.2926 - val_accuracy: 0.7984 - 1s/epoch - 15ms/step\n",
            "Epoch 17/50\n",
            "78/78 - 1s - loss: 0.2694 - accuracy: 0.8506 - val_loss: 0.2953 - val_accuracy: 0.8081 - 1s/epoch - 15ms/step\n",
            "Epoch 18/50\n",
            "78/78 - 1s - loss: 0.2634 - accuracy: 0.8724 - val_loss: 0.2928 - val_accuracy: 0.8210 - 1s/epoch - 15ms/step\n",
            "Epoch 19/50\n",
            "78/78 - 1s - loss: 0.2609 - accuracy: 0.8817 - val_loss: 0.2837 - val_accuracy: 0.8452 - 1s/epoch - 16ms/step\n",
            "Epoch 20/50\n",
            "78/78 - 1s - loss: 0.2454 - accuracy: 0.9164 - val_loss: 0.2815 - val_accuracy: 0.8484 - 1s/epoch - 16ms/step\n",
            "Epoch 21/50\n",
            "78/78 - 1s - loss: 0.2428 - accuracy: 0.9241 - val_loss: 0.2813 - val_accuracy: 0.8355 - 1s/epoch - 16ms/step\n",
            "Epoch 22/50\n",
            "78/78 - 1s - loss: 0.2379 - accuracy: 0.9310 - val_loss: 0.2971 - val_accuracy: 0.8323 - 1s/epoch - 15ms/step\n",
            "Epoch 23/50\n",
            "78/78 - 1s - loss: 0.2367 - accuracy: 0.9374 - val_loss: 0.2776 - val_accuracy: 0.8468 - 1s/epoch - 15ms/step\n",
            "Epoch 24/50\n",
            "78/78 - 1s - loss: 0.2295 - accuracy: 0.9532 - val_loss: 0.2831 - val_accuracy: 0.8565 - 1s/epoch - 16ms/step\n",
            "Epoch 25/50\n",
            "78/78 - 1s - loss: 0.2249 - accuracy: 0.9568 - val_loss: 0.2920 - val_accuracy: 0.8565 - 1s/epoch - 15ms/step\n",
            "Epoch 26/50\n",
            "78/78 - 1s - loss: 0.2313 - accuracy: 0.9443 - val_loss: 0.2752 - val_accuracy: 0.8758 - 1s/epoch - 17ms/step\n",
            "Epoch 27/50\n",
            "78/78 - 1s - loss: 0.2238 - accuracy: 0.9580 - val_loss: 0.2789 - val_accuracy: 0.8710 - 1s/epoch - 16ms/step\n",
            "Epoch 28/50\n",
            "78/78 - 2s - loss: 0.2203 - accuracy: 0.9637 - val_loss: 0.2815 - val_accuracy: 0.8726 - 2s/epoch - 21ms/step\n",
            "Epoch 29/50\n",
            "78/78 - 1s - loss: 0.2246 - accuracy: 0.9463 - val_loss: 0.2927 - val_accuracy: 0.8371 - 1s/epoch - 17ms/step\n",
            "Epoch 30/50\n",
            "78/78 - 1s - loss: 0.2250 - accuracy: 0.9439 - val_loss: 0.2920 - val_accuracy: 0.8500 - 1s/epoch - 15ms/step\n",
            "Epoch 31/50\n",
            "78/78 - 1s - loss: 0.2263 - accuracy: 0.9443 - val_loss: 0.3387 - val_accuracy: 0.7065 - 1s/epoch - 15ms/step\n",
            "Epoch 32/50\n",
            "78/78 - 1s - loss: 0.2555 - accuracy: 0.8740 - val_loss: 0.3138 - val_accuracy: 0.7839 - 1s/epoch - 15ms/step\n",
            "Epoch 33/50\n",
            "78/78 - 1s - loss: 0.2376 - accuracy: 0.9217 - val_loss: 0.2989 - val_accuracy: 0.8290 - 1s/epoch - 16ms/step\n",
            "Epoch 34/50\n",
            "78/78 - 1s - loss: 0.2256 - accuracy: 0.9471 - val_loss: 0.2934 - val_accuracy: 0.8306 - 1s/epoch - 15ms/step\n",
            "Epoch 35/50\n",
            "78/78 - 1s - loss: 0.2198 - accuracy: 0.9560 - val_loss: 0.3007 - val_accuracy: 0.8226 - 1s/epoch - 16ms/step\n",
            "Epoch 36/50\n",
            "78/78 - 1s - loss: 0.2145 - accuracy: 0.9661 - val_loss: 0.3047 - val_accuracy: 0.8355 - 1s/epoch - 15ms/step\n",
            "Epoch 37/50\n",
            "78/78 - 1s - loss: 0.2135 - accuracy: 0.9657 - val_loss: 0.2947 - val_accuracy: 0.8532 - 1s/epoch - 15ms/step\n",
            "Epoch 38/50\n",
            "78/78 - 1s - loss: 0.2101 - accuracy: 0.9701 - val_loss: 0.3080 - val_accuracy: 0.8355 - 1s/epoch - 15ms/step\n",
            "Epoch 39/50\n",
            "78/78 - 1s - loss: 0.2141 - accuracy: 0.9592 - val_loss: 0.3173 - val_accuracy: 0.8435 - 1s/epoch - 15ms/step\n",
            "Epoch 40/50\n",
            "78/78 - 1s - loss: 0.2083 - accuracy: 0.9681 - val_loss: 0.3143 - val_accuracy: 0.8339 - 1s/epoch - 15ms/step\n",
            "Epoch 41/50\n",
            "78/78 - 1s - loss: 0.2052 - accuracy: 0.9721 - val_loss: 0.3103 - val_accuracy: 0.8516 - 1s/epoch - 15ms/step\n",
            "Epoch 42/50\n",
            "78/78 - 1s - loss: 0.2062 - accuracy: 0.9750 - val_loss: 0.3136 - val_accuracy: 0.8355 - 1s/epoch - 15ms/step\n",
            "Epoch 43/50\n",
            "78/78 - 1s - loss: 0.2002 - accuracy: 0.9818 - val_loss: 0.3281 - val_accuracy: 0.8323 - 1s/epoch - 15ms/step\n",
            "Epoch 44/50\n",
            "78/78 - 1s - loss: 0.1991 - accuracy: 0.9782 - val_loss: 0.3293 - val_accuracy: 0.8355 - 1s/epoch - 15ms/step\n",
            "Epoch 45/50\n",
            "78/78 - 1s - loss: 0.1962 - accuracy: 0.9766 - val_loss: 0.3431 - val_accuracy: 0.8177 - 1s/epoch - 15ms/step\n",
            "Epoch 46/50\n",
            "78/78 - 1s - loss: 0.1995 - accuracy: 0.9762 - val_loss: 0.3312 - val_accuracy: 0.8403 - 1s/epoch - 16ms/step\n",
            "Epoch 47/50\n",
            "78/78 - 1s - loss: 0.1949 - accuracy: 0.9782 - val_loss: 0.3334 - val_accuracy: 0.8258 - 1s/epoch - 15ms/step\n",
            "Epoch 48/50\n",
            "78/78 - 1s - loss: 0.2007 - accuracy: 0.9661 - val_loss: 0.3537 - val_accuracy: 0.8032 - 1s/epoch - 15ms/step\n",
            "Epoch 49/50\n",
            "78/78 - 1s - loss: 0.1962 - accuracy: 0.9750 - val_loss: 0.3159 - val_accuracy: 0.8597 - 1s/epoch - 15ms/step\n",
            "Epoch 50/50\n",
            "78/78 - 1s - loss: 0.1909 - accuracy: 0.9814 - val_loss: 0.3390 - val_accuracy: 0.8484 - 1s/epoch - 15ms/step\n",
            "\n",
            "best model based on min training set loss:  acc= 0.9814291596412659  loss= 0.19092489778995514\n",
            "best model based on min test set loss:  acc= 0.875806450843811  loss= 0.27524009346961975\n",
            "best model based on max training set accuracy:  acc= 0.9818328619003296  loss= 0.20017215609550476\n",
            "best model based on max test set accuracy:  acc= 0.875806450843811  loss= 0.27524009346961975\n",
            "\n",
            "evaluate dataset with best model based on maximum test set accuracy\n",
            "evaluate train set=  [0.21996881067752838, 0.9656842947006226]\n",
            "evaluate test set=  [0.27524009346961975, 0.875806450843811]\n",
            "\n",
            " ------------------------------ metrics for traning set ------------------------------\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.985     0.987     0.986       532\n",
            "  Women/Girl      0.984     0.987     0.986       312\n",
            "       Child      0.965     0.953     0.959        86\n",
            "      Animal      0.979     0.968     0.974       345\n",
            "       Other      0.969     0.944     0.956       465\n",
            "          NA      0.939     0.958     0.948       737\n",
            "      Score1      0.250     0.008     0.015       128\n",
            "      Score2      0.182     0.009     0.018       217\n",
            "      Score3      0.344     0.116     0.174       653\n",
            "      Score4      0.436     0.926     0.593      1048\n",
            "      Score5      0.389     0.016     0.031       431\n",
            "\n",
            "   micro avg      0.696     0.696     0.696      4954\n",
            "   macro avg      0.675     0.625     0.604      4954\n",
            "weighted avg      0.669     0.696     0.636      4954\n",
            " samples avg      0.696     0.696     0.696      4954\n",
            "\n",
            "------------------------------ metrics for test set ------------------------------\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.903     0.949     0.925       137\n",
            "  Women/Girl      0.873     0.885     0.879        78\n",
            "       Child      0.688     0.846     0.759        13\n",
            "      Animal      0.935     0.911     0.923        79\n",
            "       Other      0.906     0.932     0.919       103\n",
            "          NA      0.843     0.795     0.819       210\n",
            "      Score1      0.000     0.000     0.000        32\n",
            "      Score2      0.250     0.018     0.034        55\n",
            "      Score3      0.154     0.042     0.066       144\n",
            "      Score4      0.456     0.935     0.613       278\n",
            "      Score5      0.250     0.009     0.017       111\n",
            "\n",
            "   micro avg      0.656     0.656     0.656      1240\n",
            "   macro avg      0.569     0.575     0.541      1240\n",
            "weighted avg      0.593     0.656     0.587      1240\n",
            " samples avg      0.656     0.656     0.656      1240\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "with tf.device('/GPU:0'):\n",
        "  model11 = RNN(X_train_list, y_train, model_num=1)\n",
        "  model11.train((X_test_list, y_test))\n",
        "  model11.evaluate(X_test_list, y_test)\n",
        "  model11.outputResult()"
      ],
      "metadata": {
        "id": "-w-w4u-grU15",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "446a6bfb-25cc-4493-9f98-bb049860af24"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "78/78 - 8s - loss: 0.5307 - accuracy: 0.1655 - val_loss: 0.4807 - val_accuracy: 0.1758 - 8s/epoch - 96ms/step\n",
            "Epoch 2/50\n",
            "78/78 - 1s - loss: 0.4628 - accuracy: 0.2717 - val_loss: 0.4620 - val_accuracy: 0.2855 - 1s/epoch - 17ms/step\n",
            "Epoch 3/50\n",
            "78/78 - 1s - loss: 0.4498 - accuracy: 0.3016 - val_loss: 0.4482 - val_accuracy: 0.3371 - 1s/epoch - 17ms/step\n",
            "Epoch 4/50\n",
            "78/78 - 1s - loss: 0.4229 - accuracy: 0.3783 - val_loss: 0.4243 - val_accuracy: 0.3629 - 1s/epoch - 17ms/step\n",
            "Epoch 5/50\n",
            "78/78 - 1s - loss: 0.3966 - accuracy: 0.5071 - val_loss: 0.3957 - val_accuracy: 0.5952 - 1s/epoch - 16ms/step\n",
            "Epoch 6/50\n",
            "78/78 - 2s - loss: 0.3615 - accuracy: 0.6928 - val_loss: 0.3671 - val_accuracy: 0.6984 - 2s/epoch - 22ms/step\n",
            "Epoch 7/50\n",
            "78/78 - 2s - loss: 0.3325 - accuracy: 0.7566 - val_loss: 0.3473 - val_accuracy: 0.7081 - 2s/epoch - 21ms/step\n",
            "Epoch 8/50\n",
            "78/78 - 1s - loss: 0.3128 - accuracy: 0.8131 - val_loss: 0.3418 - val_accuracy: 0.7210 - 1s/epoch - 16ms/step\n",
            "Epoch 9/50\n",
            "78/78 - 2s - loss: 0.3034 - accuracy: 0.8434 - val_loss: 0.3455 - val_accuracy: 0.7661 - 2s/epoch - 22ms/step\n",
            "Epoch 10/50\n",
            "78/78 - 1s - loss: 0.2940 - accuracy: 0.8716 - val_loss: 0.3313 - val_accuracy: 0.8081 - 1s/epoch - 16ms/step\n",
            "Epoch 11/50\n",
            "78/78 - 1s - loss: 0.2850 - accuracy: 0.8975 - val_loss: 0.3276 - val_accuracy: 0.8226 - 1s/epoch - 16ms/step\n",
            "Epoch 12/50\n",
            "78/78 - 2s - loss: 0.2766 - accuracy: 0.9144 - val_loss: 0.3318 - val_accuracy: 0.8242 - 2s/epoch - 20ms/step\n",
            "Epoch 13/50\n",
            "78/78 - 1s - loss: 0.2709 - accuracy: 0.9253 - val_loss: 0.3403 - val_accuracy: 0.8032 - 1s/epoch - 18ms/step\n",
            "Epoch 14/50\n",
            "78/78 - 1s - loss: 0.2674 - accuracy: 0.9273 - val_loss: 0.3339 - val_accuracy: 0.8226 - 1s/epoch - 15ms/step\n",
            "Epoch 15/50\n",
            "78/78 - 1s - loss: 0.2660 - accuracy: 0.9318 - val_loss: 0.3313 - val_accuracy: 0.8161 - 1s/epoch - 15ms/step\n",
            "Epoch 16/50\n",
            "78/78 - 1s - loss: 0.2618 - accuracy: 0.9362 - val_loss: 0.3481 - val_accuracy: 0.8097 - 1s/epoch - 15ms/step\n",
            "Epoch 17/50\n",
            "78/78 - 1s - loss: 0.2567 - accuracy: 0.9415 - val_loss: 0.3345 - val_accuracy: 0.8323 - 1s/epoch - 16ms/step\n",
            "Epoch 18/50\n",
            "78/78 - 1s - loss: 0.2556 - accuracy: 0.9467 - val_loss: 0.3416 - val_accuracy: 0.8161 - 1s/epoch - 15ms/step\n",
            "Epoch 19/50\n",
            "78/78 - 1s - loss: 0.2549 - accuracy: 0.9471 - val_loss: 0.3379 - val_accuracy: 0.8242 - 1s/epoch - 15ms/step\n",
            "Epoch 20/50\n",
            "78/78 - 1s - loss: 0.2487 - accuracy: 0.9576 - val_loss: 0.3587 - val_accuracy: 0.7903 - 1s/epoch - 15ms/step\n",
            "Epoch 21/50\n",
            "78/78 - 1s - loss: 0.2464 - accuracy: 0.9552 - val_loss: 0.3501 - val_accuracy: 0.8048 - 1s/epoch - 15ms/step\n",
            "Epoch 22/50\n",
            "78/78 - 1s - loss: 0.2494 - accuracy: 0.9524 - val_loss: 0.3491 - val_accuracy: 0.8145 - 1s/epoch - 15ms/step\n",
            "Epoch 23/50\n",
            "78/78 - 2s - loss: 0.2551 - accuracy: 0.9338 - val_loss: 0.3472 - val_accuracy: 0.8145 - 2s/epoch - 25ms/step\n",
            "Epoch 24/50\n",
            "78/78 - 1s - loss: 0.2500 - accuracy: 0.9479 - val_loss: 0.3489 - val_accuracy: 0.8177 - 1s/epoch - 16ms/step\n",
            "Epoch 25/50\n",
            "78/78 - 1s - loss: 0.2496 - accuracy: 0.9423 - val_loss: 0.3468 - val_accuracy: 0.8194 - 1s/epoch - 15ms/step\n",
            "Epoch 26/50\n",
            "78/78 - 1s - loss: 0.2408 - accuracy: 0.9621 - val_loss: 0.3548 - val_accuracy: 0.8081 - 1s/epoch - 15ms/step\n",
            "Epoch 27/50\n",
            "78/78 - 1s - loss: 0.2378 - accuracy: 0.9625 - val_loss: 0.3566 - val_accuracy: 0.8177 - 1s/epoch - 15ms/step\n",
            "Epoch 28/50\n",
            "78/78 - 1s - loss: 0.2351 - accuracy: 0.9673 - val_loss: 0.3516 - val_accuracy: 0.8435 - 1s/epoch - 16ms/step\n",
            "Epoch 29/50\n",
            "78/78 - 1s - loss: 0.2299 - accuracy: 0.9725 - val_loss: 0.3537 - val_accuracy: 0.8355 - 1s/epoch - 15ms/step\n",
            "Epoch 30/50\n",
            "78/78 - 1s - loss: 0.2247 - accuracy: 0.9746 - val_loss: 0.3729 - val_accuracy: 0.8290 - 1s/epoch - 15ms/step\n",
            "Epoch 31/50\n",
            "78/78 - 1s - loss: 0.2343 - accuracy: 0.9608 - val_loss: 0.3634 - val_accuracy: 0.8210 - 1s/epoch - 15ms/step\n",
            "Epoch 32/50\n",
            "78/78 - 1s - loss: 0.2243 - accuracy: 0.9730 - val_loss: 0.3780 - val_accuracy: 0.8210 - 1s/epoch - 15ms/step\n",
            "Epoch 33/50\n",
            "78/78 - 1s - loss: 0.2227 - accuracy: 0.9725 - val_loss: 0.4041 - val_accuracy: 0.7968 - 1s/epoch - 16ms/step\n",
            "Epoch 34/50\n",
            "78/78 - 1s - loss: 0.2237 - accuracy: 0.9705 - val_loss: 0.3996 - val_accuracy: 0.8113 - 1s/epoch - 15ms/step\n",
            "Epoch 35/50\n",
            "78/78 - 1s - loss: 0.2156 - accuracy: 0.9810 - val_loss: 0.4013 - val_accuracy: 0.8016 - 1s/epoch - 15ms/step\n",
            "Epoch 36/50\n",
            "78/78 - 1s - loss: 0.2152 - accuracy: 0.9762 - val_loss: 0.4140 - val_accuracy: 0.8065 - 1s/epoch - 15ms/step\n",
            "Epoch 37/50\n",
            "78/78 - 1s - loss: 0.2142 - accuracy: 0.9758 - val_loss: 0.4067 - val_accuracy: 0.8048 - 1s/epoch - 15ms/step\n",
            "Epoch 38/50\n",
            "78/78 - 1s - loss: 0.2135 - accuracy: 0.9810 - val_loss: 0.4141 - val_accuracy: 0.8194 - 1s/epoch - 15ms/step\n",
            "Epoch 39/50\n",
            "78/78 - 1s - loss: 0.2162 - accuracy: 0.9693 - val_loss: 0.3995 - val_accuracy: 0.8194 - 1s/epoch - 15ms/step\n",
            "Epoch 40/50\n",
            "78/78 - 1s - loss: 0.2107 - accuracy: 0.9746 - val_loss: 0.4239 - val_accuracy: 0.8161 - 1s/epoch - 15ms/step\n",
            "Epoch 41/50\n",
            "78/78 - 1s - loss: 0.2083 - accuracy: 0.9730 - val_loss: 0.4223 - val_accuracy: 0.8065 - 1s/epoch - 15ms/step\n",
            "Epoch 42/50\n",
            "78/78 - 1s - loss: 0.2121 - accuracy: 0.9681 - val_loss: 0.4183 - val_accuracy: 0.8177 - 1s/epoch - 15ms/step\n",
            "Epoch 43/50\n",
            "78/78 - 1s - loss: 0.2005 - accuracy: 0.9834 - val_loss: 0.4386 - val_accuracy: 0.8000 - 1s/epoch - 15ms/step\n",
            "Epoch 44/50\n",
            "78/78 - 1s - loss: 0.1950 - accuracy: 0.9859 - val_loss: 0.4489 - val_accuracy: 0.8032 - 1s/epoch - 15ms/step\n",
            "Epoch 45/50\n",
            "78/78 - 1s - loss: 0.1992 - accuracy: 0.9705 - val_loss: 0.4493 - val_accuracy: 0.7952 - 1s/epoch - 15ms/step\n",
            "Epoch 46/50\n",
            "78/78 - 1s - loss: 0.2065 - accuracy: 0.9657 - val_loss: 0.4599 - val_accuracy: 0.7774 - 1s/epoch - 15ms/step\n",
            "Epoch 47/50\n",
            "78/78 - 1s - loss: 0.2204 - accuracy: 0.9487 - val_loss: 0.4211 - val_accuracy: 0.8113 - 1s/epoch - 15ms/step\n",
            "Epoch 48/50\n",
            "78/78 - 1s - loss: 0.2068 - accuracy: 0.9580 - val_loss: 0.4244 - val_accuracy: 0.8177 - 1s/epoch - 15ms/step\n",
            "Epoch 49/50\n",
            "78/78 - 1s - loss: 0.1923 - accuracy: 0.9730 - val_loss: 0.4376 - val_accuracy: 0.8145 - 1s/epoch - 17ms/step\n",
            "Epoch 50/50\n",
            "78/78 - 1s - loss: 0.1863 - accuracy: 0.9746 - val_loss: 0.4564 - val_accuracy: 0.8065 - 1s/epoch - 16ms/step\n",
            "\n",
            "best model based on min training set loss:  acc= 0.9745659828186035  loss= 0.1862754225730896\n",
            "best model based on min test set loss:  acc= 0.8225806355476379  loss= 0.32761088013648987\n",
            "best model based on max training set accuracy:  acc= 0.9858700037002563  loss= 0.19504934549331665\n",
            "best model based on max test set accuracy:  acc= 0.8435483574867249  loss= 0.35157209634780884\n",
            "\n",
            "evaluate dataset with best model based on maximum test set accuracy\n",
            "evaluate train set=  [0.22911228239536285, 0.9685102701187134]\n",
            "evaluate test set=  [0.35157209634780884, 0.8435483574867249]\n",
            "\n",
            " ------------------------------ metrics for traning set ------------------------------\n",
            "confusion matrix: \n",
            " [[[1911   21]\n",
            "  [   4  541]]\n",
            "\n",
            " [[2178    3]\n",
            "  [   1  295]]\n",
            "\n",
            " [[2393    8]\n",
            "  [   1   75]]\n",
            "\n",
            " [[2137    9]\n",
            "  [   3  328]]\n",
            "\n",
            " [[2009    9]\n",
            "  [  21  438]]\n",
            "\n",
            " [[1681   26]\n",
            "  [  46  724]]\n",
            "\n",
            " [[1918  218]\n",
            "  [ 261   80]]\n",
            "\n",
            " [[1492  463]\n",
            "  [ 284  238]]\n",
            "\n",
            " [[1410  427]\n",
            "  [ 314  326]]\n",
            "\n",
            " [[1911  130]\n",
            "  [ 329  107]]\n",
            "\n",
            " [[1595  344]\n",
            "  [ 394  144]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.963     0.993     0.977       545\n",
            "  Women/Girl      0.990     0.997     0.993       296\n",
            "       Child      0.904     0.987     0.943        76\n",
            "      Animal      0.973     0.991     0.982       331\n",
            "       Other      0.980     0.954     0.967       459\n",
            "          NA      0.965     0.940     0.953       770\n",
            "      Score1      0.268     0.235     0.250       341\n",
            "      Score2      0.340     0.456     0.389       522\n",
            "      Score3      0.433     0.509     0.468       640\n",
            "      Score4      0.451     0.245     0.318       436\n",
            "      Score5      0.295     0.268     0.281       538\n",
            "\n",
            "   micro avg      0.665     0.665     0.665      4954\n",
            "   macro avg      0.687     0.689     0.684      4954\n",
            "weighted avg      0.667     0.665     0.662      4954\n",
            " samples avg      0.665     0.665     0.665      4954\n",
            "\n",
            "------------------------------ metrics for test set ------------------------------\n",
            "confusion matrix: \n",
            " [[[482  14]\n",
            "  [  6 118]]\n",
            "\n",
            " [[516  10]\n",
            "  [ 10  84]]\n",
            "\n",
            " [[589   8]\n",
            "  [  4  19]]\n",
            "\n",
            " [[514  13]\n",
            "  [  6  87]]\n",
            "\n",
            " [[500  11]\n",
            "  [ 25  84]]\n",
            "\n",
            " [[403  40]\n",
            "  [ 45 132]]\n",
            "\n",
            " [[472  55]\n",
            "  [ 83  10]]\n",
            "\n",
            " [[339 139]\n",
            "  [110  32]]\n",
            "\n",
            " [[320 153]\n",
            "  [ 94  53]]\n",
            "\n",
            " [[456  50]\n",
            "  [106   8]]\n",
            "\n",
            " [[401  95]\n",
            "  [ 99  25]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.894     0.952     0.922       124\n",
            "  Women/Girl      0.894     0.894     0.894        94\n",
            "       Child      0.704     0.826     0.760        23\n",
            "      Animal      0.870     0.935     0.902        93\n",
            "       Other      0.884     0.771     0.824       109\n",
            "          NA      0.767     0.746     0.756       177\n",
            "      Score1      0.154     0.108     0.127        93\n",
            "      Score2      0.187     0.225     0.204       142\n",
            "      Score3      0.257     0.361     0.300       147\n",
            "      Score4      0.138     0.070     0.093       114\n",
            "      Score5      0.208     0.202     0.205       124\n",
            "\n",
            "   micro avg      0.526     0.526     0.526      1240\n",
            "   macro avg      0.542     0.553     0.544      1240\n",
            "weighted avg      0.520     0.526     0.520      1240\n",
            " samples avg      0.526     0.526     0.526      1240\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Model 2**"
      ],
      "metadata": {
        "id": "iPFr5_uOXVLk"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "with tf.device('/GPU:0'):\n",
        "  model21 = RNN(X_train_list, y_train, model_num=2)\n",
        "  model21.train((X_test_list, y_test))\n",
        "  model21.evaluate(X_test_list, y_test)\n",
        "  model21.outputResult()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "CVj3hsH0upnF",
        "outputId": "bff7fbc5-2ab2-4eeb-967d-ab453e2f8b5e"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "78/78 - 8s - loss: 0.6015 - accuracy: 0.0327 - val_loss: 0.5339 - val_accuracy: 0.0371 - 8s/epoch - 104ms/step\n",
            "Epoch 2/50\n",
            "78/78 - 1s - loss: 0.4966 - accuracy: 0.1340 - val_loss: 0.4735 - val_accuracy: 0.2065 - 1s/epoch - 17ms/step\n",
            "Epoch 3/50\n",
            "78/78 - 1s - loss: 0.4318 - accuracy: 0.3940 - val_loss: 0.4123 - val_accuracy: 0.5419 - 1s/epoch - 16ms/step\n",
            "Epoch 4/50\n",
            "78/78 - 1s - loss: 0.3843 - accuracy: 0.6270 - val_loss: 0.3801 - val_accuracy: 0.5339 - 1s/epoch - 16ms/step\n",
            "Epoch 5/50\n",
            "78/78 - 2s - loss: 0.3542 - accuracy: 0.6948 - val_loss: 0.3651 - val_accuracy: 0.6452 - 2s/epoch - 29ms/step\n",
            "Epoch 6/50\n",
            "78/78 - 1s - loss: 0.3371 - accuracy: 0.7364 - val_loss: 0.3552 - val_accuracy: 0.6629 - 1s/epoch - 19ms/step\n",
            "Epoch 7/50\n",
            "78/78 - 1s - loss: 0.3148 - accuracy: 0.8179 - val_loss: 0.3469 - val_accuracy: 0.7597 - 1s/epoch - 17ms/step\n",
            "Epoch 8/50\n",
            "78/78 - 1s - loss: 0.2956 - accuracy: 0.8692 - val_loss: 0.3241 - val_accuracy: 0.8097 - 1s/epoch - 16ms/step\n",
            "Epoch 9/50\n",
            "78/78 - 1s - loss: 0.2835 - accuracy: 0.8966 - val_loss: 0.3191 - val_accuracy: 0.8371 - 1s/epoch - 16ms/step\n",
            "Epoch 10/50\n",
            "78/78 - 1s - loss: 0.2740 - accuracy: 0.9152 - val_loss: 0.3154 - val_accuracy: 0.8500 - 1s/epoch - 16ms/step\n",
            "Epoch 11/50\n",
            "78/78 - 1s - loss: 0.2657 - accuracy: 0.9342 - val_loss: 0.3140 - val_accuracy: 0.8710 - 1s/epoch - 16ms/step\n",
            "Epoch 12/50\n",
            "78/78 - 1s - loss: 0.2594 - accuracy: 0.9479 - val_loss: 0.3156 - val_accuracy: 0.8565 - 1s/epoch - 16ms/step\n",
            "Epoch 13/50\n",
            "78/78 - 1s - loss: 0.2580 - accuracy: 0.9455 - val_loss: 0.3147 - val_accuracy: 0.8500 - 1s/epoch - 16ms/step\n",
            "Epoch 14/50\n",
            "78/78 - 1s - loss: 0.2566 - accuracy: 0.9447 - val_loss: 0.3099 - val_accuracy: 0.8774 - 1s/epoch - 16ms/step\n",
            "Epoch 15/50\n",
            "78/78 - 1s - loss: 0.2468 - accuracy: 0.9608 - val_loss: 0.3155 - val_accuracy: 0.8629 - 1s/epoch - 15ms/step\n",
            "Epoch 16/50\n",
            "78/78 - 1s - loss: 0.2436 - accuracy: 0.9649 - val_loss: 0.3174 - val_accuracy: 0.8565 - 1s/epoch - 15ms/step\n",
            "Epoch 17/50\n",
            "78/78 - 1s - loss: 0.2390 - accuracy: 0.9713 - val_loss: 0.3224 - val_accuracy: 0.8548 - 1s/epoch - 16ms/step\n",
            "Epoch 18/50\n",
            "78/78 - 1s - loss: 0.2360 - accuracy: 0.9730 - val_loss: 0.3292 - val_accuracy: 0.8500 - 1s/epoch - 15ms/step\n",
            "Epoch 19/50\n",
            "78/78 - 1s - loss: 0.2334 - accuracy: 0.9701 - val_loss: 0.3380 - val_accuracy: 0.8355 - 1s/epoch - 16ms/step\n",
            "Epoch 20/50\n",
            "78/78 - 1s - loss: 0.2354 - accuracy: 0.9669 - val_loss: 0.3281 - val_accuracy: 0.8468 - 1s/epoch - 16ms/step\n",
            "Epoch 21/50\n",
            "78/78 - 1s - loss: 0.2278 - accuracy: 0.9774 - val_loss: 0.3326 - val_accuracy: 0.8532 - 1s/epoch - 16ms/step\n",
            "Epoch 22/50\n",
            "78/78 - 1s - loss: 0.2262 - accuracy: 0.9746 - val_loss: 0.3365 - val_accuracy: 0.8323 - 1s/epoch - 17ms/step\n",
            "Epoch 23/50\n",
            "78/78 - 1s - loss: 0.2236 - accuracy: 0.9717 - val_loss: 0.3373 - val_accuracy: 0.8629 - 1s/epoch - 16ms/step\n",
            "Epoch 24/50\n",
            "78/78 - 1s - loss: 0.2190 - accuracy: 0.9782 - val_loss: 0.3458 - val_accuracy: 0.8403 - 1s/epoch - 16ms/step\n",
            "Epoch 25/50\n",
            "78/78 - 1s - loss: 0.2164 - accuracy: 0.9782 - val_loss: 0.3506 - val_accuracy: 0.8532 - 1s/epoch - 17ms/step\n",
            "Epoch 26/50\n",
            "78/78 - 1s - loss: 0.2090 - accuracy: 0.9847 - val_loss: 0.3518 - val_accuracy: 0.8468 - 1s/epoch - 16ms/step\n",
            "Epoch 27/50\n",
            "78/78 - 1s - loss: 0.2081 - accuracy: 0.9830 - val_loss: 0.3682 - val_accuracy: 0.8339 - 1s/epoch - 16ms/step\n",
            "Epoch 28/50\n",
            "78/78 - 1s - loss: 0.2050 - accuracy: 0.9847 - val_loss: 0.3702 - val_accuracy: 0.8435 - 1s/epoch - 16ms/step\n",
            "Epoch 29/50\n",
            "78/78 - 1s - loss: 0.1983 - accuracy: 0.9847 - val_loss: 0.3840 - val_accuracy: 0.8323 - 1s/epoch - 15ms/step\n",
            "Epoch 30/50\n",
            "78/78 - 1s - loss: 0.1982 - accuracy: 0.9826 - val_loss: 0.3836 - val_accuracy: 0.8468 - 1s/epoch - 16ms/step\n",
            "Epoch 31/50\n",
            "78/78 - 1s - loss: 0.1946 - accuracy: 0.9851 - val_loss: 0.3885 - val_accuracy: 0.8516 - 1s/epoch - 16ms/step\n",
            "Epoch 32/50\n",
            "78/78 - 1s - loss: 0.1930 - accuracy: 0.9847 - val_loss: 0.3923 - val_accuracy: 0.8419 - 1s/epoch - 15ms/step\n",
            "Epoch 33/50\n",
            "78/78 - 1s - loss: 0.1937 - accuracy: 0.9798 - val_loss: 0.3876 - val_accuracy: 0.8435 - 1s/epoch - 15ms/step\n",
            "Epoch 34/50\n",
            "78/78 - 1s - loss: 0.1896 - accuracy: 0.9867 - val_loss: 0.3968 - val_accuracy: 0.8532 - 1s/epoch - 16ms/step\n",
            "Epoch 35/50\n",
            "78/78 - 1s - loss: 0.1830 - accuracy: 0.9843 - val_loss: 0.4074 - val_accuracy: 0.8419 - 1s/epoch - 15ms/step\n",
            "Epoch 36/50\n",
            "78/78 - 1s - loss: 0.1789 - accuracy: 0.9867 - val_loss: 0.4064 - val_accuracy: 0.8484 - 1s/epoch - 15ms/step\n",
            "Epoch 37/50\n",
            "78/78 - 1s - loss: 0.1736 - accuracy: 0.9851 - val_loss: 0.4160 - val_accuracy: 0.8484 - 1s/epoch - 15ms/step\n",
            "Epoch 38/50\n",
            "78/78 - 1s - loss: 0.1699 - accuracy: 0.9859 - val_loss: 0.4353 - val_accuracy: 0.8258 - 1s/epoch - 16ms/step\n",
            "Epoch 39/50\n",
            "78/78 - 1s - loss: 0.1700 - accuracy: 0.9790 - val_loss: 0.4378 - val_accuracy: 0.8306 - 1s/epoch - 16ms/step\n",
            "Epoch 40/50\n",
            "78/78 - 1s - loss: 0.1693 - accuracy: 0.9766 - val_loss: 0.4511 - val_accuracy: 0.8258 - 1s/epoch - 16ms/step\n",
            "Epoch 41/50\n",
            "78/78 - 1s - loss: 0.1616 - accuracy: 0.9762 - val_loss: 0.4504 - val_accuracy: 0.8210 - 1s/epoch - 16ms/step\n",
            "Epoch 42/50\n",
            "78/78 - 1s - loss: 0.1567 - accuracy: 0.9810 - val_loss: 0.4556 - val_accuracy: 0.8290 - 1s/epoch - 16ms/step\n",
            "Epoch 43/50\n",
            "78/78 - 1s - loss: 0.1543 - accuracy: 0.9798 - val_loss: 0.4862 - val_accuracy: 0.8290 - 1s/epoch - 16ms/step\n",
            "Epoch 44/50\n",
            "78/78 - 1s - loss: 0.1517 - accuracy: 0.9746 - val_loss: 0.4816 - val_accuracy: 0.7984 - 1s/epoch - 16ms/step\n",
            "Epoch 45/50\n",
            "78/78 - 2s - loss: 0.1486 - accuracy: 0.9802 - val_loss: 0.4940 - val_accuracy: 0.8242 - 2s/epoch - 21ms/step\n",
            "Epoch 46/50\n",
            "78/78 - 1s - loss: 0.1463 - accuracy: 0.9701 - val_loss: 0.5103 - val_accuracy: 0.8194 - 1s/epoch - 19ms/step\n",
            "Epoch 47/50\n",
            "78/78 - 1s - loss: 0.1514 - accuracy: 0.9625 - val_loss: 0.5053 - val_accuracy: 0.8161 - 1s/epoch - 16ms/step\n",
            "Epoch 48/50\n",
            "78/78 - 1s - loss: 0.1481 - accuracy: 0.9584 - val_loss: 0.5197 - val_accuracy: 0.8113 - 1s/epoch - 16ms/step\n",
            "Epoch 49/50\n",
            "78/78 - 1s - loss: 0.1339 - accuracy: 0.9762 - val_loss: 0.5209 - val_accuracy: 0.8177 - 1s/epoch - 16ms/step\n",
            "Epoch 50/50\n",
            "78/78 - 1s - loss: 0.1267 - accuracy: 0.9713 - val_loss: 0.5380 - val_accuracy: 0.8113 - 1s/epoch - 16ms/step\n",
            "\n",
            "best model based on min training set loss:  acc= 0.971336305141449  loss= 0.12668797373771667\n",
            "best model based on min test set loss:  acc= 0.8774193525314331  loss= 0.30993834137916565\n",
            "best model based on max training set accuracy:  acc= 0.9866774082183838  loss= 0.18963345885276794\n",
            "best model based on max test set accuracy:  acc= 0.8774193525314331  loss= 0.30993834137916565\n",
            "\n",
            "evaluate dataset with best model based on maximum test set accuracy\n",
            "evaluate train set=  [0.24609872698783875, 0.9612434506416321]\n",
            "evaluate test set=  [0.30993834137916565, 0.8774193525314331]\n",
            "\n",
            " ------------------------------ metrics for traning set ------------------------------\n",
            "confusion matrix: \n",
            " [[[1931    1]\n",
            "  [  12  533]]\n",
            "\n",
            " [[2168   13]\n",
            "  [   6  290]]\n",
            "\n",
            " [[2387   14]\n",
            "  [   0   76]]\n",
            "\n",
            " [[2134   12]\n",
            "  [   2  329]]\n",
            "\n",
            " [[1999   19]\n",
            "  [  17  442]]\n",
            "\n",
            " [[1673   34]\n",
            "  [  56  714]]\n",
            "\n",
            " [[2032  104]\n",
            "  [ 271   70]]\n",
            "\n",
            " [[1839  116]\n",
            "  [ 475   47]]\n",
            "\n",
            " [[1233  604]\n",
            "  [ 366  274]]\n",
            "\n",
            " [[2026   15]\n",
            "  [ 431    5]]\n",
            "\n",
            " [[1021  918]\n",
            "  [ 214  324]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.998     0.978     0.988       545\n",
            "  Women/Girl      0.957     0.980     0.968       296\n",
            "       Child      0.844     1.000     0.916        76\n",
            "      Animal      0.965     0.994     0.979       331\n",
            "       Other      0.959     0.963     0.961       459\n",
            "          NA      0.955     0.927     0.941       770\n",
            "      Score1      0.402     0.205     0.272       341\n",
            "      Score2      0.288     0.090     0.137       522\n",
            "      Score3      0.312     0.428     0.361       640\n",
            "      Score4      0.250     0.011     0.022       436\n",
            "      Score5      0.261     0.602     0.364       538\n",
            "\n",
            "   micro avg      0.627     0.627     0.627      4954\n",
            "   macro avg      0.654     0.653     0.628      4954\n",
            "weighted avg      0.630     0.627     0.603      4954\n",
            " samples avg      0.627     0.627     0.627      4954\n",
            "\n",
            "------------------------------ metrics for test set ------------------------------\n",
            "confusion matrix: \n",
            " [[[491   5]\n",
            "  [  9 115]]\n",
            "\n",
            " [[520   6]\n",
            "  [ 10  84]]\n",
            "\n",
            " [[597   0]\n",
            "  [  5  18]]\n",
            "\n",
            " [[515  12]\n",
            "  [  4  89]]\n",
            "\n",
            " [[500  11]\n",
            "  [ 19  90]]\n",
            "\n",
            " [[403  40]\n",
            "  [ 27 150]]\n",
            "\n",
            " [[497  30]\n",
            "  [ 80  13]]\n",
            "\n",
            " [[457  21]\n",
            "  [135   7]]\n",
            "\n",
            " [[288 185]\n",
            "  [ 93  54]]\n",
            "\n",
            " [[505   1]\n",
            "  [113   1]]\n",
            "\n",
            " [[256 240]\n",
            "  [ 56  68]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.958     0.927     0.943       124\n",
            "  Women/Girl      0.933     0.894     0.913        94\n",
            "       Child      1.000     0.783     0.878        23\n",
            "      Animal      0.881     0.957     0.918        93\n",
            "       Other      0.891     0.826     0.857       109\n",
            "          NA      0.789     0.847     0.817       177\n",
            "      Score1      0.302     0.140     0.191        93\n",
            "      Score2      0.250     0.049     0.082       142\n",
            "      Score3      0.226     0.367     0.280       147\n",
            "      Score4      0.500     0.009     0.017       114\n",
            "      Score5      0.221     0.548     0.315       124\n",
            "\n",
            "   micro avg      0.556     0.556     0.556      1240\n",
            "   macro avg      0.632     0.577     0.565      1240\n",
            "weighted avg      0.588     0.556     0.531      1240\n",
            " samples avg      0.556     0.556     0.556      1240\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Model 3**"
      ],
      "metadata": {
        "id": "u7C-zaeUh0oL"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "with tf.device('/GPU:0'):\n",
        "  model3 = RNN(X_train_list, y_train, model_num=3)\n",
        "  model3.train((X_test_list, y_test))\n",
        "  model3.evaluate(X_test_list, y_test)\n",
        "  model3.outputResult()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DV6Tql1fh3GM",
        "outputId": "31e20afb-ef5a-45f8-c049-b80b750d9375"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "78/78 - 8s - loss: 0.5966 - accuracy: 0.0000e+00 - val_loss: 0.5383 - val_accuracy: 0.0000e+00 - 8s/epoch - 104ms/step\n",
            "Epoch 2/50\n",
            "78/78 - 1s - loss: 0.4967 - accuracy: 0.0763 - val_loss: 0.4659 - val_accuracy: 0.1597 - 1s/epoch - 16ms/step\n",
            "Epoch 3/50\n",
            "78/78 - 1s - loss: 0.4404 - accuracy: 0.3690 - val_loss: 0.4298 - val_accuracy: 0.4081 - 1s/epoch - 15ms/step\n",
            "Epoch 4/50\n",
            "78/78 - 1s - loss: 0.4034 - accuracy: 0.5329 - val_loss: 0.4208 - val_accuracy: 0.5097 - 1s/epoch - 16ms/step\n",
            "Epoch 5/50\n",
            "78/78 - 1s - loss: 0.3886 - accuracy: 0.5793 - val_loss: 0.4109 - val_accuracy: 0.4919 - 1s/epoch - 14ms/step\n",
            "Epoch 6/50\n",
            "78/78 - 1s - loss: 0.3724 - accuracy: 0.6145 - val_loss: 0.3977 - val_accuracy: 0.5403 - 1s/epoch - 17ms/step\n",
            "Epoch 7/50\n",
            "78/78 - 1s - loss: 0.3476 - accuracy: 0.7231 - val_loss: 0.3818 - val_accuracy: 0.6532 - 1s/epoch - 15ms/step\n",
            "Epoch 8/50\n",
            "78/78 - 1s - loss: 0.3309 - accuracy: 0.7662 - val_loss: 0.3652 - val_accuracy: 0.6661 - 1s/epoch - 15ms/step\n",
            "Epoch 9/50\n",
            "78/78 - 1s - loss: 0.3176 - accuracy: 0.7840 - val_loss: 0.3676 - val_accuracy: 0.6806 - 1s/epoch - 16ms/step\n",
            "Epoch 10/50\n",
            "78/78 - 1s - loss: 0.3158 - accuracy: 0.7727 - val_loss: 0.3468 - val_accuracy: 0.7161 - 1s/epoch - 16ms/step\n",
            "Epoch 11/50\n",
            "78/78 - 1s - loss: 0.3021 - accuracy: 0.8066 - val_loss: 0.3549 - val_accuracy: 0.6952 - 1s/epoch - 15ms/step\n",
            "Epoch 12/50\n",
            "78/78 - 1s - loss: 0.2907 - accuracy: 0.8692 - val_loss: 0.3428 - val_accuracy: 0.8194 - 1s/epoch - 15ms/step\n",
            "Epoch 13/50\n",
            "78/78 - 1s - loss: 0.2821 - accuracy: 0.9180 - val_loss: 0.3419 - val_accuracy: 0.8306 - 1s/epoch - 16ms/step\n",
            "Epoch 14/50\n",
            "78/78 - 1s - loss: 0.2771 - accuracy: 0.9318 - val_loss: 0.3365 - val_accuracy: 0.8355 - 1s/epoch - 15ms/step\n",
            "Epoch 15/50\n",
            "78/78 - 1s - loss: 0.2702 - accuracy: 0.9342 - val_loss: 0.3432 - val_accuracy: 0.8339 - 1s/epoch - 14ms/step\n",
            "Epoch 16/50\n",
            "78/78 - 1s - loss: 0.2704 - accuracy: 0.9310 - val_loss: 0.3525 - val_accuracy: 0.8339 - 1s/epoch - 15ms/step\n",
            "Epoch 17/50\n",
            "78/78 - 1s - loss: 0.2640 - accuracy: 0.9394 - val_loss: 0.3294 - val_accuracy: 0.8484 - 1s/epoch - 16ms/step\n",
            "Epoch 18/50\n",
            "78/78 - 1s - loss: 0.2632 - accuracy: 0.9411 - val_loss: 0.3460 - val_accuracy: 0.8435 - 1s/epoch - 15ms/step\n",
            "Epoch 19/50\n",
            "78/78 - 1s - loss: 0.2549 - accuracy: 0.9487 - val_loss: 0.3433 - val_accuracy: 0.8387 - 1s/epoch - 15ms/step\n",
            "Epoch 20/50\n",
            "78/78 - 1s - loss: 0.2547 - accuracy: 0.9467 - val_loss: 0.3510 - val_accuracy: 0.8194 - 1s/epoch - 14ms/step\n",
            "Epoch 21/50\n",
            "78/78 - 1s - loss: 0.2522 - accuracy: 0.9499 - val_loss: 0.3464 - val_accuracy: 0.8403 - 1s/epoch - 14ms/step\n",
            "Epoch 22/50\n",
            "78/78 - 1s - loss: 0.2529 - accuracy: 0.9419 - val_loss: 0.3496 - val_accuracy: 0.8306 - 1s/epoch - 15ms/step\n",
            "Epoch 23/50\n",
            "78/78 - 1s - loss: 0.2469 - accuracy: 0.9532 - val_loss: 0.3472 - val_accuracy: 0.8306 - 1s/epoch - 14ms/step\n",
            "Epoch 24/50\n",
            "78/78 - 1s - loss: 0.2402 - accuracy: 0.9621 - val_loss: 0.3440 - val_accuracy: 0.8532 - 1s/epoch - 15ms/step\n",
            "Epoch 25/50\n",
            "78/78 - 1s - loss: 0.2417 - accuracy: 0.9580 - val_loss: 0.3506 - val_accuracy: 0.8355 - 1s/epoch - 14ms/step\n",
            "Epoch 26/50\n",
            "78/78 - 1s - loss: 0.2497 - accuracy: 0.9455 - val_loss: 0.3475 - val_accuracy: 0.8355 - 1s/epoch - 14ms/step\n",
            "Epoch 27/50\n",
            "78/78 - 1s - loss: 0.2383 - accuracy: 0.9641 - val_loss: 0.3597 - val_accuracy: 0.8145 - 1s/epoch - 14ms/step\n",
            "Epoch 28/50\n",
            "78/78 - 1s - loss: 0.2370 - accuracy: 0.9580 - val_loss: 0.3494 - val_accuracy: 0.8468 - 1s/epoch - 14ms/step\n",
            "Epoch 29/50\n",
            "78/78 - 1s - loss: 0.2288 - accuracy: 0.9766 - val_loss: 0.3594 - val_accuracy: 0.8355 - 1s/epoch - 14ms/step\n",
            "Epoch 30/50\n",
            "78/78 - 1s - loss: 0.2254 - accuracy: 0.9778 - val_loss: 0.3676 - val_accuracy: 0.8452 - 1s/epoch - 14ms/step\n",
            "Epoch 31/50\n",
            "78/78 - 1s - loss: 0.2219 - accuracy: 0.9798 - val_loss: 0.3793 - val_accuracy: 0.8355 - 1s/epoch - 14ms/step\n",
            "Epoch 32/50\n",
            "78/78 - 1s - loss: 0.2211 - accuracy: 0.9786 - val_loss: 0.3861 - val_accuracy: 0.8226 - 1s/epoch - 15ms/step\n",
            "Epoch 33/50\n",
            "78/78 - 1s - loss: 0.2184 - accuracy: 0.9766 - val_loss: 0.3894 - val_accuracy: 0.8258 - 1s/epoch - 19ms/step\n",
            "Epoch 34/50\n",
            "78/78 - 1s - loss: 0.2285 - accuracy: 0.9604 - val_loss: 0.4038 - val_accuracy: 0.8306 - 1s/epoch - 16ms/step\n",
            "Epoch 35/50\n",
            "78/78 - 1s - loss: 0.2254 - accuracy: 0.9693 - val_loss: 0.3828 - val_accuracy: 0.8290 - 1s/epoch - 14ms/step\n",
            "Epoch 36/50\n",
            "78/78 - 1s - loss: 0.2192 - accuracy: 0.9734 - val_loss: 0.3827 - val_accuracy: 0.8419 - 1s/epoch - 14ms/step\n",
            "Epoch 37/50\n",
            "78/78 - 1s - loss: 0.2155 - accuracy: 0.9762 - val_loss: 0.3855 - val_accuracy: 0.8290 - 1s/epoch - 14ms/step\n",
            "Epoch 38/50\n",
            "78/78 - 1s - loss: 0.2157 - accuracy: 0.9778 - val_loss: 0.3912 - val_accuracy: 0.8274 - 1s/epoch - 14ms/step\n",
            "Epoch 39/50\n",
            "78/78 - 1s - loss: 0.2141 - accuracy: 0.9758 - val_loss: 0.3970 - val_accuracy: 0.8371 - 1s/epoch - 14ms/step\n",
            "Epoch 40/50\n",
            "78/78 - 1s - loss: 0.2090 - accuracy: 0.9814 - val_loss: 0.4038 - val_accuracy: 0.8242 - 1s/epoch - 14ms/step\n",
            "Epoch 41/50\n",
            "78/78 - 1s - loss: 0.2090 - accuracy: 0.9818 - val_loss: 0.4199 - val_accuracy: 0.8274 - 1s/epoch - 14ms/step\n",
            "Epoch 42/50\n",
            "78/78 - 1s - loss: 0.2029 - accuracy: 0.9822 - val_loss: 0.4241 - val_accuracy: 0.8258 - 1s/epoch - 14ms/step\n",
            "Epoch 43/50\n",
            "78/78 - 1s - loss: 0.2011 - accuracy: 0.9843 - val_loss: 0.4421 - val_accuracy: 0.8258 - 1s/epoch - 14ms/step\n",
            "Epoch 44/50\n",
            "78/78 - 1s - loss: 0.1960 - accuracy: 0.9867 - val_loss: 0.4419 - val_accuracy: 0.8355 - 1s/epoch - 14ms/step\n",
            "Epoch 45/50\n",
            "78/78 - 1s - loss: 0.1981 - accuracy: 0.9847 - val_loss: 0.4505 - val_accuracy: 0.8403 - 1s/epoch - 14ms/step\n",
            "Epoch 46/50\n",
            "78/78 - 1s - loss: 0.2125 - accuracy: 0.9604 - val_loss: 0.4283 - val_accuracy: 0.8210 - 1s/epoch - 14ms/step\n",
            "Epoch 47/50\n",
            "78/78 - 1s - loss: 0.2005 - accuracy: 0.9822 - val_loss: 0.4520 - val_accuracy: 0.8258 - 1s/epoch - 14ms/step\n",
            "Epoch 48/50\n",
            "78/78 - 1s - loss: 0.1950 - accuracy: 0.9786 - val_loss: 0.4780 - val_accuracy: 0.8210 - 1s/epoch - 14ms/step\n",
            "Epoch 49/50\n",
            "78/78 - 1s - loss: 0.2100 - accuracy: 0.9685 - val_loss: 0.4523 - val_accuracy: 0.8145 - 1s/epoch - 15ms/step\n",
            "Epoch 50/50\n",
            "78/78 - 1s - loss: 0.1936 - accuracy: 0.9806 - val_loss: 0.4674 - val_accuracy: 0.8129 - 1s/epoch - 14ms/step\n",
            "\n",
            "best model based on min training set loss:  acc= 0.9806216955184937  loss= 0.19357730448246002\n",
            "best model based on min test set loss:  acc= 0.8483871221542358  loss= 0.32938310503959656\n",
            "best model based on max training set accuracy:  acc= 0.9866774082183838  loss= 0.19602550566196442\n",
            "best model based on max test set accuracy:  acc= 0.853225827217102  loss= 0.3439982831478119\n",
            "\n",
            "evaluate dataset with best model based on maximum test set accuracy\n",
            "evaluate train set=  [0.23254036903381348, 0.971336305141449]\n",
            "evaluate test set=  [0.3439982831478119, 0.853225827217102]\n",
            "\n",
            " ------------------------------ metrics for traning set ------------------------------\n",
            "confusion matrix: \n",
            " [[[1923    9]\n",
            "  [   1  544]]\n",
            "\n",
            " [[2167   14]\n",
            "  [   3  293]]\n",
            "\n",
            " [[2384   17]\n",
            "  [   1   75]]\n",
            "\n",
            " [[2139    7]\n",
            "  [  10  321]]\n",
            "\n",
            " [[2016    2]\n",
            "  [  14  445]]\n",
            "\n",
            " [[1691   16]\n",
            "  [  36  734]]\n",
            "\n",
            " [[1859  277]\n",
            "  [ 232  109]]\n",
            "\n",
            " [[1926   29]\n",
            "  [ 501   21]]\n",
            "\n",
            " [[1135  702]\n",
            "  [ 331  309]]\n",
            "\n",
            " [[1859  182]\n",
            "  [ 334  102]]\n",
            "\n",
            " [[1476  463]\n",
            "  [ 255  283]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.984     0.998     0.991       545\n",
            "  Women/Girl      0.954     0.990     0.972       296\n",
            "       Child      0.815     0.987     0.893        76\n",
            "      Animal      0.979     0.970     0.974       331\n",
            "       Other      0.996     0.969     0.982       459\n",
            "          NA      0.979     0.953     0.966       770\n",
            "      Score1      0.282     0.320     0.300       341\n",
            "      Score2      0.420     0.040     0.073       522\n",
            "      Score3      0.306     0.483     0.374       640\n",
            "      Score4      0.359     0.234     0.283       436\n",
            "      Score5      0.379     0.526     0.441       538\n",
            "\n",
            "   micro avg      0.653     0.653     0.653      4954\n",
            "   macro avg      0.678     0.679     0.659      4954\n",
            "weighted avg      0.663     0.653     0.637      4954\n",
            " samples avg      0.653     0.653     0.653      4954\n",
            "\n",
            "------------------------------ metrics for test set ------------------------------\n",
            "confusion matrix: \n",
            " [[[480  16]\n",
            "  [  6 118]]\n",
            "\n",
            " [[518   8]\n",
            "  [ 11  83]]\n",
            "\n",
            " [[594   3]\n",
            "  [  8  15]]\n",
            "\n",
            " [[520   7]\n",
            "  [ 10  83]]\n",
            "\n",
            " [[503   8]\n",
            "  [ 26  83]]\n",
            "\n",
            " [[394  49]\n",
            "  [ 30 147]]\n",
            "\n",
            " [[460  67]\n",
            "  [ 71  22]]\n",
            "\n",
            " [[468  10]\n",
            "  [137   5]]\n",
            "\n",
            " [[274 199]\n",
            "  [ 81  66]]\n",
            "\n",
            " [[467  39]\n",
            "  [101  13]]\n",
            "\n",
            " [[334 162]\n",
            "  [ 87  37]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.881     0.952     0.915       124\n",
            "  Women/Girl      0.912     0.883     0.897        94\n",
            "       Child      0.833     0.652     0.732        23\n",
            "      Animal      0.922     0.892     0.907        93\n",
            "       Other      0.912     0.761     0.830       109\n",
            "          NA      0.750     0.831     0.788       177\n",
            "      Score1      0.247     0.237     0.242        93\n",
            "      Score2      0.333     0.035     0.064       142\n",
            "      Score3      0.249     0.449     0.320       147\n",
            "      Score4      0.250     0.114     0.157       114\n",
            "      Score5      0.186     0.298     0.229       124\n",
            "\n",
            "   micro avg      0.542     0.542     0.542      1240\n",
            "   macro avg      0.589     0.555     0.553      1240\n",
            "weighted avg      0.557     0.542     0.527      1240\n",
            " samples avg      0.542     0.542     0.542      1240\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Model 4**"
      ],
      "metadata": {
        "id": "PH61K3zUsinn"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "with tf.device('/GPU:0'):\n",
        "  model8 = RNN(X_train_list, y_train, model_num=8)\n",
        "  model8.train((X_test_list, y_test))\n",
        "  model8.evaluate(X_test_list, y_test)\n",
        "  model8.outputResult()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wL3p4IOUspRM",
        "outputId": "6fbefe3c-d777-4a49-b620-b937a6db7535"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "78/78 - 18s - loss: 0.5487 - accuracy: 0.1558 - val_loss: 0.4773 - val_accuracy: 0.3468 - 18s/epoch - 233ms/step\n",
            "Epoch 2/50\n",
            "78/78 - 5s - loss: 0.4647 - accuracy: 0.2842 - val_loss: 0.4286 - val_accuracy: 0.4774 - 5s/epoch - 69ms/step\n",
            "Epoch 3/50\n",
            "78/78 - 5s - loss: 0.4293 - accuracy: 0.4098 - val_loss: 0.3960 - val_accuracy: 0.6065 - 5s/epoch - 68ms/step\n",
            "Epoch 4/50\n",
            "78/78 - 5s - loss: 0.3915 - accuracy: 0.5672 - val_loss: 0.3679 - val_accuracy: 0.6419 - 5s/epoch - 70ms/step\n",
            "Epoch 5/50\n",
            "78/78 - 5s - loss: 0.3688 - accuracy: 0.6657 - val_loss: 0.3422 - val_accuracy: 0.7290 - 5s/epoch - 70ms/step\n",
            "Epoch 6/50\n",
            "78/78 - 5s - loss: 0.3473 - accuracy: 0.7368 - val_loss: 0.3273 - val_accuracy: 0.8048 - 5s/epoch - 69ms/step\n",
            "Epoch 7/50\n",
            "78/78 - 5s - loss: 0.3298 - accuracy: 0.7893 - val_loss: 0.3107 - val_accuracy: 0.8500 - 5s/epoch - 69ms/step\n",
            "Epoch 8/50\n",
            "78/78 - 5s - loss: 0.3197 - accuracy: 0.8317 - val_loss: 0.3102 - val_accuracy: 0.8484 - 5s/epoch - 68ms/step\n",
            "Epoch 9/50\n",
            "78/78 - 5s - loss: 0.3215 - accuracy: 0.8288 - val_loss: 0.3097 - val_accuracy: 0.8419 - 5s/epoch - 69ms/step\n",
            "Epoch 10/50\n",
            "78/78 - 6s - loss: 0.3057 - accuracy: 0.8583 - val_loss: 0.3098 - val_accuracy: 0.8532 - 6s/epoch - 71ms/step\n",
            "Epoch 11/50\n",
            "78/78 - 6s - loss: 0.3007 - accuracy: 0.8797 - val_loss: 0.3010 - val_accuracy: 0.8645 - 6s/epoch - 75ms/step\n",
            "Epoch 12/50\n",
            "78/78 - 6s - loss: 0.2943 - accuracy: 0.8894 - val_loss: 0.2966 - val_accuracy: 0.8758 - 6s/epoch - 72ms/step\n",
            "Epoch 13/50\n",
            "78/78 - 5s - loss: 0.2944 - accuracy: 0.8797 - val_loss: 0.2988 - val_accuracy: 0.8710 - 5s/epoch - 68ms/step\n",
            "Epoch 14/50\n",
            "78/78 - 5s - loss: 0.2944 - accuracy: 0.8914 - val_loss: 0.2975 - val_accuracy: 0.8726 - 5s/epoch - 68ms/step\n",
            "Epoch 15/50\n",
            "78/78 - 5s - loss: 0.2889 - accuracy: 0.9007 - val_loss: 0.2979 - val_accuracy: 0.8645 - 5s/epoch - 68ms/step\n",
            "Epoch 16/50\n",
            "78/78 - 5s - loss: 0.2851 - accuracy: 0.9051 - val_loss: 0.2951 - val_accuracy: 0.8726 - 5s/epoch - 68ms/step\n",
            "Epoch 17/50\n",
            "78/78 - 5s - loss: 0.2846 - accuracy: 0.9120 - val_loss: 0.2945 - val_accuracy: 0.8790 - 5s/epoch - 69ms/step\n",
            "Epoch 18/50\n",
            "78/78 - 5s - loss: 0.2775 - accuracy: 0.9229 - val_loss: 0.2965 - val_accuracy: 0.8726 - 5s/epoch - 69ms/step\n",
            "Epoch 19/50\n",
            "78/78 - 5s - loss: 0.2778 - accuracy: 0.9140 - val_loss: 0.3012 - val_accuracy: 0.8629 - 5s/epoch - 67ms/step\n",
            "Epoch 20/50\n",
            "78/78 - 5s - loss: 0.2764 - accuracy: 0.9201 - val_loss: 0.2981 - val_accuracy: 0.8806 - 5s/epoch - 69ms/step\n",
            "Epoch 21/50\n",
            "78/78 - 5s - loss: 0.2729 - accuracy: 0.9265 - val_loss: 0.3003 - val_accuracy: 0.8742 - 5s/epoch - 68ms/step\n",
            "Epoch 22/50\n",
            "78/78 - 5s - loss: 0.2718 - accuracy: 0.9269 - val_loss: 0.2997 - val_accuracy: 0.8597 - 5s/epoch - 68ms/step\n",
            "Epoch 23/50\n",
            "78/78 - 5s - loss: 0.2729 - accuracy: 0.9277 - val_loss: 0.2972 - val_accuracy: 0.8790 - 5s/epoch - 68ms/step\n",
            "Epoch 24/50\n",
            "78/78 - 5s - loss: 0.2635 - accuracy: 0.9423 - val_loss: 0.3021 - val_accuracy: 0.8790 - 5s/epoch - 68ms/step\n",
            "Epoch 25/50\n",
            "78/78 - 5s - loss: 0.2673 - accuracy: 0.9358 - val_loss: 0.2978 - val_accuracy: 0.8871 - 5s/epoch - 69ms/step\n",
            "Epoch 26/50\n",
            "78/78 - 5s - loss: 0.2652 - accuracy: 0.9382 - val_loss: 0.2991 - val_accuracy: 0.8806 - 5s/epoch - 68ms/step\n",
            "Epoch 27/50\n",
            "78/78 - 6s - loss: 0.2619 - accuracy: 0.9435 - val_loss: 0.3020 - val_accuracy: 0.8823 - 6s/epoch - 74ms/step\n",
            "Epoch 28/50\n",
            "78/78 - 5s - loss: 0.2641 - accuracy: 0.9386 - val_loss: 0.3017 - val_accuracy: 0.8790 - 5s/epoch - 68ms/step\n",
            "Epoch 29/50\n",
            "78/78 - 5s - loss: 0.2596 - accuracy: 0.9507 - val_loss: 0.3071 - val_accuracy: 0.8694 - 5s/epoch - 68ms/step\n",
            "Epoch 30/50\n",
            "78/78 - 5s - loss: 0.2574 - accuracy: 0.9483 - val_loss: 0.3028 - val_accuracy: 0.8790 - 5s/epoch - 69ms/step\n",
            "Epoch 31/50\n",
            "78/78 - 5s - loss: 0.2537 - accuracy: 0.9524 - val_loss: 0.2999 - val_accuracy: 0.8758 - 5s/epoch - 68ms/step\n",
            "Epoch 32/50\n",
            "78/78 - 5s - loss: 0.2499 - accuracy: 0.9560 - val_loss: 0.3011 - val_accuracy: 0.8694 - 5s/epoch - 68ms/step\n",
            "Epoch 33/50\n",
            "78/78 - 6s - loss: 0.2535 - accuracy: 0.9507 - val_loss: 0.3064 - val_accuracy: 0.8790 - 6s/epoch - 77ms/step\n",
            "Epoch 34/50\n",
            "78/78 - 5s - loss: 0.2499 - accuracy: 0.9564 - val_loss: 0.3031 - val_accuracy: 0.8726 - 5s/epoch - 68ms/step\n",
            "Epoch 35/50\n",
            "78/78 - 5s - loss: 0.2498 - accuracy: 0.9548 - val_loss: 0.3039 - val_accuracy: 0.8694 - 5s/epoch - 69ms/step\n",
            "Epoch 36/50\n",
            "78/78 - 5s - loss: 0.2513 - accuracy: 0.9516 - val_loss: 0.3077 - val_accuracy: 0.8726 - 5s/epoch - 67ms/step\n",
            "Epoch 37/50\n",
            "78/78 - 5s - loss: 0.2467 - accuracy: 0.9629 - val_loss: 0.3127 - val_accuracy: 0.8806 - 5s/epoch - 69ms/step\n",
            "Epoch 38/50\n",
            "78/78 - 5s - loss: 0.2472 - accuracy: 0.9584 - val_loss: 0.3125 - val_accuracy: 0.8855 - 5s/epoch - 68ms/step\n",
            "Epoch 39/50\n",
            "78/78 - 5s - loss: 0.2460 - accuracy: 0.9629 - val_loss: 0.3119 - val_accuracy: 0.8806 - 5s/epoch - 68ms/step\n",
            "Epoch 40/50\n",
            "78/78 - 5s - loss: 0.2480 - accuracy: 0.9552 - val_loss: 0.3049 - val_accuracy: 0.8871 - 5s/epoch - 68ms/step\n",
            "Epoch 41/50\n",
            "78/78 - 5s - loss: 0.2418 - accuracy: 0.9705 - val_loss: 0.3120 - val_accuracy: 0.8742 - 5s/epoch - 68ms/step\n",
            "Epoch 42/50\n",
            "78/78 - 5s - loss: 0.2387 - accuracy: 0.9721 - val_loss: 0.3236 - val_accuracy: 0.8806 - 5s/epoch - 68ms/step\n",
            "Epoch 43/50\n",
            "78/78 - 5s - loss: 0.2383 - accuracy: 0.9742 - val_loss: 0.3204 - val_accuracy: 0.8806 - 5s/epoch - 68ms/step\n",
            "Epoch 44/50\n",
            "78/78 - 5s - loss: 0.2368 - accuracy: 0.9750 - val_loss: 0.3232 - val_accuracy: 0.8774 - 5s/epoch - 69ms/step\n",
            "Epoch 45/50\n",
            "78/78 - 5s - loss: 0.2332 - accuracy: 0.9778 - val_loss: 0.3213 - val_accuracy: 0.8823 - 5s/epoch - 69ms/step\n",
            "Epoch 46/50\n",
            "78/78 - 5s - loss: 0.2347 - accuracy: 0.9730 - val_loss: 0.3361 - val_accuracy: 0.8710 - 5s/epoch - 68ms/step\n",
            "Epoch 47/50\n",
            "78/78 - 5s - loss: 0.2331 - accuracy: 0.9810 - val_loss: 0.3301 - val_accuracy: 0.8742 - 5s/epoch - 67ms/step\n",
            "Epoch 48/50\n",
            "78/78 - 5s - loss: 0.2315 - accuracy: 0.9766 - val_loss: 0.3311 - val_accuracy: 0.8774 - 5s/epoch - 67ms/step\n",
            "Epoch 49/50\n",
            "78/78 - 6s - loss: 0.2373 - accuracy: 0.9697 - val_loss: 0.3337 - val_accuracy: 0.8694 - 6s/epoch - 72ms/step\n",
            "Epoch 50/50\n",
            "78/78 - 6s - loss: 0.2354 - accuracy: 0.9742 - val_loss: 0.3205 - val_accuracy: 0.8790 - 6s/epoch - 75ms/step\n",
            "\n",
            "best model based on min training set loss:  acc= 0.9765845537185669  loss= 0.23149321973323822\n",
            "best model based on min test set loss:  acc= 0.8790322542190552  loss= 0.2944726347923279\n",
            "best model based on max training set accuracy:  acc= 0.9810254573822021  loss= 0.23311354219913483\n",
            "best model based on max test set accuracy:  acc= 0.8870967626571655  loss= 0.2977857291698456\n",
            "\n",
            "evaluate dataset with best model based on maximum test set accuracy\n",
            "evaluate train set=  [0.2515788674354553, 0.9547840356826782]\n",
            "evaluate test set=  [0.2977857291698456, 0.8870967626571655]\n",
            "\n",
            " ------------------------------ metrics for traning set ------------------------------\n",
            "confusion matrix: \n",
            " [[[1932   20]\n",
            "  [   6  519]]\n",
            "\n",
            " [[2152   10]\n",
            "  [   3  312]]\n",
            "\n",
            " [[2392    4]\n",
            "  [   4   77]]\n",
            "\n",
            " [[2132   11]\n",
            "  [   7  327]]\n",
            "\n",
            " [[2001   16]\n",
            "  [  38  422]]\n",
            "\n",
            " [[1664   51]\n",
            "  [  54  708]]\n",
            "\n",
            " [[1763  373]\n",
            "  [ 200  141]]\n",
            "\n",
            " [[1953    6]\n",
            "  [ 516    2]]\n",
            "\n",
            " [[ 834 1008]\n",
            "  [ 254  381]]\n",
            "\n",
            " [[1953   72]\n",
            "  [ 428   24]]\n",
            "\n",
            " [[1596  350]\n",
            "  [ 411  120]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.963     0.989     0.976       525\n",
            "  Women/Girl      0.969     0.990     0.980       315\n",
            "       Child      0.951     0.951     0.951        81\n",
            "      Animal      0.967     0.979     0.973       334\n",
            "       Other      0.963     0.917     0.940       460\n",
            "          NA      0.933     0.929     0.931       762\n",
            "      Score1      0.274     0.413     0.330       341\n",
            "      Score2      0.250     0.004     0.008       518\n",
            "      Score3      0.274     0.600     0.376       635\n",
            "      Score4      0.250     0.053     0.088       452\n",
            "      Score5      0.255     0.226     0.240       531\n",
            "\n",
            "   micro avg      0.612     0.612     0.612      4954\n",
            "   macro avg      0.641     0.641     0.617      4954\n",
            "weighted avg      0.608     0.612     0.583      4954\n",
            " samples avg      0.612     0.612     0.612      4954\n",
            "\n",
            "------------------------------ metrics for test set ------------------------------\n",
            "confusion matrix: \n",
            " [[[460  16]\n",
            "  [  8 136]]\n",
            "\n",
            " [[542   3]\n",
            "  [  2  73]]\n",
            "\n",
            " [[599   3]\n",
            "  [  3  15]]\n",
            "\n",
            " [[520  10]\n",
            "  [  9  81]]\n",
            "\n",
            " [[502  10]\n",
            "  [ 10  98]]\n",
            "\n",
            " [[408  27]\n",
            "  [ 37 148]]\n",
            "\n",
            " [[446  81]\n",
            "  [ 58  35]]\n",
            "\n",
            " [[471   3]\n",
            "  [145   1]]\n",
            "\n",
            " [[198 270]\n",
            "  [ 67  85]]\n",
            "\n",
            " [[505  17]\n",
            "  [ 94   4]]\n",
            "\n",
            " [[391  98]\n",
            "  [105  26]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.895     0.944     0.919       144\n",
            "  Women/Girl      0.961     0.973     0.967        75\n",
            "       Child      0.833     0.833     0.833        18\n",
            "      Animal      0.890     0.900     0.895        90\n",
            "       Other      0.907     0.907     0.907       108\n",
            "          NA      0.846     0.800     0.822       185\n",
            "      Score1      0.302     0.376     0.335        93\n",
            "      Score2      0.250     0.007     0.013       146\n",
            "      Score3      0.239     0.559     0.335       152\n",
            "      Score4      0.190     0.041     0.067        98\n",
            "      Score5      0.210     0.198     0.204       131\n",
            "\n",
            "   micro avg      0.566     0.566     0.566      1240\n",
            "   macro avg      0.593     0.595     0.573      1240\n",
            "weighted avg      0.563     0.566     0.539      1240\n",
            " samples avg      0.566     0.566     0.566      1240\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "with tf.device('/GPU:0'):\n",
        "  model81 = RNN(X_train_list, y_train, model_num=8)\n",
        "  model81.train((X_test_list, y_test))\n",
        "  model81.evaluate(X_test_list, y_test)\n",
        "  model81.outputResult()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Rm25vmi4LMIB",
        "outputId": "168f4ea4-16ba-4dc6-e535-2b97cfe0f376"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "78/78 - 12s - loss: 0.5100 - accuracy: 0.0832 - val_loss: 0.4290 - val_accuracy: 0.4258 - 12s/epoch - 152ms/step\n",
            "Epoch 2/50\n",
            "78/78 - 6s - loss: 0.4228 - accuracy: 0.4425 - val_loss: 0.3835 - val_accuracy: 0.6065 - 6s/epoch - 71ms/step\n",
            "Epoch 3/50\n",
            "78/78 - 6s - loss: 0.3853 - accuracy: 0.6165 - val_loss: 0.3631 - val_accuracy: 0.7210 - 6s/epoch - 71ms/step\n",
            "Epoch 4/50\n",
            "78/78 - 6s - loss: 0.3639 - accuracy: 0.7109 - val_loss: 0.3336 - val_accuracy: 0.8274 - 6s/epoch - 71ms/step\n",
            "Epoch 5/50\n",
            "78/78 - 5s - loss: 0.3386 - accuracy: 0.8099 - val_loss: 0.3227 - val_accuracy: 0.8274 - 5s/epoch - 69ms/step\n",
            "Epoch 6/50\n",
            "78/78 - 6s - loss: 0.3249 - accuracy: 0.8228 - val_loss: 0.3169 - val_accuracy: 0.8306 - 6s/epoch - 71ms/step\n",
            "Epoch 7/50\n",
            "78/78 - 6s - loss: 0.3138 - accuracy: 0.8494 - val_loss: 0.3070 - val_accuracy: 0.8548 - 6s/epoch - 71ms/step\n",
            "Epoch 8/50\n",
            "78/78 - 6s - loss: 0.3041 - accuracy: 0.8676 - val_loss: 0.3074 - val_accuracy: 0.8500 - 6s/epoch - 78ms/step\n",
            "Epoch 9/50\n",
            "78/78 - 5s - loss: 0.3002 - accuracy: 0.8785 - val_loss: 0.3006 - val_accuracy: 0.8710 - 5s/epoch - 69ms/step\n",
            "Epoch 10/50\n",
            "78/78 - 6s - loss: 0.2933 - accuracy: 0.8966 - val_loss: 0.3010 - val_accuracy: 0.8742 - 6s/epoch - 71ms/step\n",
            "Epoch 11/50\n",
            "78/78 - 5s - loss: 0.2866 - accuracy: 0.9051 - val_loss: 0.3047 - val_accuracy: 0.8516 - 5s/epoch - 69ms/step\n",
            "Epoch 12/50\n",
            "78/78 - 5s - loss: 0.2867 - accuracy: 0.9015 - val_loss: 0.3036 - val_accuracy: 0.8613 - 5s/epoch - 69ms/step\n",
            "Epoch 13/50\n",
            "78/78 - 5s - loss: 0.2820 - accuracy: 0.9104 - val_loss: 0.2982 - val_accuracy: 0.8694 - 5s/epoch - 69ms/step\n",
            "Epoch 14/50\n",
            "78/78 - 5s - loss: 0.2788 - accuracy: 0.9132 - val_loss: 0.3008 - val_accuracy: 0.8645 - 5s/epoch - 69ms/step\n",
            "Epoch 15/50\n",
            "78/78 - 6s - loss: 0.2769 - accuracy: 0.9172 - val_loss: 0.2931 - val_accuracy: 0.8806 - 6s/epoch - 71ms/step\n",
            "Epoch 16/50\n",
            "78/78 - 5s - loss: 0.2735 - accuracy: 0.9298 - val_loss: 0.2925 - val_accuracy: 0.8887 - 5s/epoch - 70ms/step\n",
            "Epoch 17/50\n",
            "78/78 - 5s - loss: 0.2688 - accuracy: 0.9334 - val_loss: 0.2959 - val_accuracy: 0.8774 - 5s/epoch - 69ms/step\n",
            "Epoch 18/50\n",
            "78/78 - 5s - loss: 0.2652 - accuracy: 0.9411 - val_loss: 0.3029 - val_accuracy: 0.8710 - 5s/epoch - 69ms/step\n",
            "Epoch 19/50\n",
            "78/78 - 5s - loss: 0.2649 - accuracy: 0.9330 - val_loss: 0.3398 - val_accuracy: 0.8258 - 5s/epoch - 69ms/step\n",
            "Epoch 20/50\n",
            "78/78 - 5s - loss: 0.2667 - accuracy: 0.9306 - val_loss: 0.2948 - val_accuracy: 0.8871 - 5s/epoch - 70ms/step\n",
            "Epoch 21/50\n",
            "78/78 - 5s - loss: 0.2615 - accuracy: 0.9386 - val_loss: 0.2997 - val_accuracy: 0.8823 - 5s/epoch - 69ms/step\n",
            "Epoch 22/50\n",
            "78/78 - 6s - loss: 0.2579 - accuracy: 0.9520 - val_loss: 0.2952 - val_accuracy: 0.8919 - 6s/epoch - 71ms/step\n",
            "Epoch 23/50\n",
            "78/78 - 5s - loss: 0.2545 - accuracy: 0.9524 - val_loss: 0.3005 - val_accuracy: 0.8742 - 5s/epoch - 69ms/step\n",
            "Epoch 24/50\n",
            "78/78 - 5s - loss: 0.2569 - accuracy: 0.9520 - val_loss: 0.3037 - val_accuracy: 0.8742 - 5s/epoch - 70ms/step\n",
            "Epoch 25/50\n",
            "78/78 - 5s - loss: 0.2561 - accuracy: 0.9503 - val_loss: 0.3133 - val_accuracy: 0.8548 - 5s/epoch - 69ms/step\n",
            "Epoch 26/50\n",
            "78/78 - 5s - loss: 0.2585 - accuracy: 0.9370 - val_loss: 0.3044 - val_accuracy: 0.8758 - 5s/epoch - 69ms/step\n",
            "Epoch 27/50\n",
            "78/78 - 5s - loss: 0.2507 - accuracy: 0.9616 - val_loss: 0.3133 - val_accuracy: 0.8677 - 5s/epoch - 70ms/step\n",
            "Epoch 28/50\n",
            "78/78 - 5s - loss: 0.2466 - accuracy: 0.9641 - val_loss: 0.3100 - val_accuracy: 0.8710 - 5s/epoch - 69ms/step\n",
            "Epoch 29/50\n",
            "78/78 - 5s - loss: 0.2464 - accuracy: 0.9637 - val_loss: 0.3094 - val_accuracy: 0.8806 - 5s/epoch - 69ms/step\n",
            "Epoch 30/50\n",
            "78/78 - 5s - loss: 0.2441 - accuracy: 0.9669 - val_loss: 0.3031 - val_accuracy: 0.8855 - 5s/epoch - 69ms/step\n",
            "Epoch 31/50\n",
            "78/78 - 5s - loss: 0.2429 - accuracy: 0.9612 - val_loss: 0.3156 - val_accuracy: 0.8758 - 5s/epoch - 70ms/step\n",
            "Epoch 32/50\n",
            "78/78 - 5s - loss: 0.2423 - accuracy: 0.9709 - val_loss: 0.3250 - val_accuracy: 0.8548 - 5s/epoch - 70ms/step\n",
            "Epoch 33/50\n",
            "78/78 - 5s - loss: 0.2453 - accuracy: 0.9629 - val_loss: 0.3107 - val_accuracy: 0.8823 - 5s/epoch - 70ms/step\n",
            "Epoch 34/50\n",
            "78/78 - 5s - loss: 0.2432 - accuracy: 0.9621 - val_loss: 0.3180 - val_accuracy: 0.8629 - 5s/epoch - 70ms/step\n",
            "Epoch 35/50\n",
            "78/78 - 5s - loss: 0.2403 - accuracy: 0.9713 - val_loss: 0.3211 - val_accuracy: 0.8710 - 5s/epoch - 69ms/step\n",
            "Epoch 36/50\n",
            "78/78 - 5s - loss: 0.2398 - accuracy: 0.9697 - val_loss: 0.3264 - val_accuracy: 0.8677 - 5s/epoch - 69ms/step\n",
            "Epoch 37/50\n",
            "78/78 - 5s - loss: 0.2393 - accuracy: 0.9677 - val_loss: 0.3322 - val_accuracy: 0.8516 - 5s/epoch - 69ms/step\n",
            "Epoch 38/50\n",
            "78/78 - 5s - loss: 0.2351 - accuracy: 0.9766 - val_loss: 0.3286 - val_accuracy: 0.8613 - 5s/epoch - 69ms/step\n",
            "Epoch 39/50\n",
            "78/78 - 5s - loss: 0.2535 - accuracy: 0.9475 - val_loss: 0.3246 - val_accuracy: 0.8661 - 5s/epoch - 69ms/step\n",
            "Epoch 40/50\n",
            "78/78 - 5s - loss: 0.2376 - accuracy: 0.9730 - val_loss: 0.3269 - val_accuracy: 0.8629 - 5s/epoch - 68ms/step\n",
            "Epoch 41/50\n",
            "78/78 - 5s - loss: 0.2373 - accuracy: 0.9669 - val_loss: 0.3282 - val_accuracy: 0.8677 - 5s/epoch - 69ms/step\n",
            "Epoch 42/50\n",
            "78/78 - 5s - loss: 0.2385 - accuracy: 0.9673 - val_loss: 0.3351 - val_accuracy: 0.8661 - 5s/epoch - 69ms/step\n",
            "Epoch 43/50\n",
            "78/78 - 5s - loss: 0.2332 - accuracy: 0.9746 - val_loss: 0.3354 - val_accuracy: 0.8710 - 5s/epoch - 69ms/step\n",
            "Epoch 44/50\n",
            "78/78 - 5s - loss: 0.2292 - accuracy: 0.9810 - val_loss: 0.3293 - val_accuracy: 0.8694 - 5s/epoch - 70ms/step\n",
            "Epoch 45/50\n",
            "78/78 - 5s - loss: 0.2273 - accuracy: 0.9847 - val_loss: 0.3285 - val_accuracy: 0.8661 - 5s/epoch - 68ms/step\n",
            "Epoch 46/50\n",
            "78/78 - 5s - loss: 0.2291 - accuracy: 0.9802 - val_loss: 0.3345 - val_accuracy: 0.8581 - 5s/epoch - 68ms/step\n",
            "Epoch 47/50\n",
            "78/78 - 5s - loss: 0.2308 - accuracy: 0.9766 - val_loss: 0.3290 - val_accuracy: 0.8839 - 5s/epoch - 68ms/step\n",
            "Epoch 48/50\n",
            "78/78 - 6s - loss: 0.2294 - accuracy: 0.9778 - val_loss: 0.3440 - val_accuracy: 0.8726 - 6s/epoch - 77ms/step\n",
            "Epoch 49/50\n",
            "78/78 - 5s - loss: 0.2274 - accuracy: 0.9766 - val_loss: 0.3567 - val_accuracy: 0.8565 - 5s/epoch - 69ms/step\n",
            "Epoch 50/50\n",
            "78/78 - 5s - loss: 0.2242 - accuracy: 0.9822 - val_loss: 0.3657 - val_accuracy: 0.8694 - 5s/epoch - 68ms/step\n",
            "\n",
            "best model based on min training set loss:  acc= 0.9822365641593933  loss= 0.2241699993610382\n",
            "best model based on min test set loss:  acc= 0.8887096643447876  loss= 0.29250574111938477\n",
            "best model based on max training set accuracy:  acc= 0.9846588373184204  loss= 0.2272823005914688\n",
            "best model based on max test set accuracy:  acc= 0.8919354677200317  loss= 0.2951980233192444\n",
            "\n",
            "evaluate dataset with best model based on maximum test set accuracy\n",
            "evaluate train set=  [0.24195407330989838, 0.9668954610824585]\n",
            "evaluate test set=  [0.2951980233192444, 0.8919354677200317]\n",
            "\n",
            " ------------------------------ metrics for traning set ------------------------------\n",
            "confusion matrix: \n",
            " [[[1944    8]\n",
            "  [   9  516]]\n",
            "\n",
            " [[2150   12]\n",
            "  [   3  312]]\n",
            "\n",
            " [[2388    8]\n",
            "  [   0   81]]\n",
            "\n",
            " [[2138    5]\n",
            "  [   4  330]]\n",
            "\n",
            " [[2013    4]\n",
            "  [  33  427]]\n",
            "\n",
            " [[1670   45]\n",
            "  [  33  729]]\n",
            "\n",
            " [[1880  256]\n",
            "  [ 231  110]]\n",
            "\n",
            " [[1712  247]\n",
            "  [ 418  100]]\n",
            "\n",
            " [[1096  746]\n",
            "  [ 322  313]]\n",
            "\n",
            " [[1845  180]\n",
            "  [ 397   55]]\n",
            "\n",
            " [[1624  322]\n",
            "  [ 383  148]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.985     0.983     0.984       525\n",
            "  Women/Girl      0.963     0.990     0.977       315\n",
            "       Child      0.910     1.000     0.953        81\n",
            "      Animal      0.985     0.988     0.987       334\n",
            "       Other      0.991     0.928     0.958       460\n",
            "          NA      0.942     0.957     0.949       762\n",
            "      Score1      0.301     0.323     0.311       341\n",
            "      Score2      0.288     0.193     0.231       518\n",
            "      Score3      0.296     0.493     0.370       635\n",
            "      Score4      0.234     0.122     0.160       452\n",
            "      Score5      0.315     0.279     0.296       531\n",
            "\n",
            "   micro avg      0.630     0.630     0.630      4954\n",
            "   macro avg      0.655     0.660     0.652      4954\n",
            "weighted avg      0.628     0.630     0.623      4954\n",
            " samples avg      0.630     0.630     0.630      4954\n",
            "\n",
            "------------------------------ metrics for test set ------------------------------\n",
            "confusion matrix: \n",
            " [[[464  12]\n",
            "  [  9 135]]\n",
            "\n",
            " [[539   6]\n",
            "  [  2  73]]\n",
            "\n",
            " [[598   4]\n",
            "  [  0  18]]\n",
            "\n",
            " [[522   8]\n",
            "  [  7  83]]\n",
            "\n",
            " [[503   9]\n",
            "  [ 12  96]]\n",
            "\n",
            " [[407  28]\n",
            "  [ 37 148]]\n",
            "\n",
            " [[476  51]\n",
            "  [ 73  20]]\n",
            "\n",
            " [[407  67]\n",
            "  [129  17]]\n",
            "\n",
            " [[256 212]\n",
            "  [ 87  65]]\n",
            "\n",
            " [[457  65]\n",
            "  [ 89   9]]\n",
            "\n",
            " [[398  91]\n",
            "  [108  23]]]\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "     Men/Boy      0.918     0.938     0.928       144\n",
            "  Women/Girl      0.924     0.973     0.948        75\n",
            "       Child      0.818     1.000     0.900        18\n",
            "      Animal      0.912     0.922     0.917        90\n",
            "       Other      0.914     0.889     0.901       108\n",
            "          NA      0.841     0.800     0.820       185\n",
            "      Score1      0.282     0.215     0.244        93\n",
            "      Score2      0.202     0.116     0.148       146\n",
            "      Score3      0.235     0.428     0.303       152\n",
            "      Score4      0.122     0.092     0.105        98\n",
            "      Score5      0.202     0.176     0.188       131\n",
            "\n",
            "   micro avg      0.554     0.554     0.554      1240\n",
            "   macro avg      0.579     0.595     0.582      1240\n",
            "weighted avg      0.550     0.554     0.547      1240\n",
            " samples avg      0.554     0.554     0.554      1240\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "showPlots(model81)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 541
        },
        "id": "6Y_rsbx6niZC",
        "outputId": "274393a0-5067-4b74-c403-50c97a7eb4d1"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plot_model(model8.model, to_file='/content/LSTM_plot.png')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "DjXXJRReNJ1A",
        "outputId": "dc64c1de-60ee-4553-bdee-0de1131238b6"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<IPython.core.display.Image object>"
            ],
            "image/png": "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\n"
          },
          "metadata": {},
          "execution_count": 54
        }
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [
        "fo1HnzyLo6bh",
        "Q1qXvuLDsHy-",
        "LXvHNESIZ_BV",
        "gzcKYBWePRVS",
        "LIA_l44HaJXD",
        "9rY02m-adIsR",
        "8bAfP_I6BJEP",
        "JtId7nYzVJGC",
        "43UxIhy88Xfm",
        "51aHU20h8kdg",
        "PBQmryY-9QMc",
        "XCpPYCUde4AK",
        "uRUJbQgFOtuT",
        "buodTk7DPAv6",
        "vMWnMMgLRlw7",
        "CseiX3GND2Wn",
        "xPG4nWnsF_dy",
        "V2mzDLZJLGxo",
        "iPFr5_uOXVLk",
        "u7C-zaeUh0oL",
        "PH61K3zUsinn"
      ],
      "provenance": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}