quickstart.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Quickstart for MNIST-1D\n",
    "Sam Greydanus | 2020\n",
    "\n",
    "This notebook shows how to build the MNIST-1D dataset and train some baselines*.\n",
    "\n",
    "_* If you're running this in Colab, it's best to use a GPU runtime._"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fatal: destination path 'mnist1d' already exists and is not an empty directory.\n"
     ]
    }
   ],
   "source": [
    "# Run this if you're in a Colab\n",
    "!git clone https://github.com/greydanus/mnist1d"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using: cuda\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import torch\n",
    "import random\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import sys ; sys.path.append('..'); sys.path.append('.')  # useful if you're running locally\n",
    "import mnist1d\n",
    "\n",
    "from mnist1d.data import get_templates, get_dataset_args, get_dataset\n",
    "from train import get_model_args, train_model\n",
    "from models import ConvBase, GRUBase, MLPBase, LinearBase\n",
    "from mnist1d.utils import set_seed, plot_signals, ObjectView, from_pickle\n",
    "\n",
    "# Try attaching to GPU\n",
    "DEVICE = str(torch.device('cuda' if torch.cuda.is_available() else 'cpu'))\n",
    "print('Using:', DEVICE)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Step 1: Visualize templates\n",
    "To construct MNIST-1D, we start with ten 1D signals, each of which resembles a digit in the range 0-9."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Templates for the MNIST-1D dataset:\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 900x119.7 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "templates = get_templates()\n",
    "print(\"Templates for the MNIST-1D dataset:\")\n",
    "x = templates['x']\n",
    "t = templates['t']\n",
    "y = templates['y']\n",
    "fig = plot_signals(x, t, labels=y, ratio=1.33, dark_mode=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 900x119.7 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plot_signals(x, t, labels=y, ratio=1.33, dark_mode=False)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Step 2: Transform\n",
    "\n",
    "Now we are going to transform these signals in such a way that they become difficult to classify. Nonlinear representations and good spatial priors will become very important in order to classify them."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'num_samples': 5000,\n",
       " 'train_split': 0.8,\n",
       " 'template_len': 12,\n",
       " 'padding': [36, 60],\n",
       " 'scale_coeff': 0.4,\n",
       " 'max_translation': 48,\n",
       " 'corr_noise_scale': 0.25,\n",
       " 'iid_noise_scale': 0.02,\n",
       " 'shear_scale': 0.75,\n",
       " 'shuffle_seq': False,\n",
       " 'final_seq_length': 40,\n",
       " 'seed': 42,\n",
       " 'url': 'https://github.com/greydanus/mnist1d/raw/master/mnist1d_data.pkl'}"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "get_dataset_args(as_dict=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 900x198 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "arg_dict = get_dataset_args(as_dict=True)\n",
    "\n",
    "arg_dict['padding'] = [36, 60]\n",
    "arg_dict['max_translation'] = 48\n",
    "arg_dict['scale_coeff'] = 0\n",
    "arg_dict['corr_noise_scale'] = 0\n",
    "arg_dict['iid_noise_scale'] = 0\n",
    "arg_dict['shear_scale'] = 0\n",
    "args = ObjectView(arg_dict)\n",
    "\n",
    "x = templates['x']\n",
    "t = templates['t']\n",
    "y = templates['y']\n",
    "set_seed(args.seed)\n",
    "fig = plot_signals(x, t, labels=y, args=args, ratio=2.2, do_transform=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 900x198 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "args = get_dataset_args()\n",
    "\n",
    "x = templates['x']\n",
    "t = templates['t']\n",
    "y = templates['y']\n",
    "set_seed(args.seed)\n",
    "fig = plot_signals(x, t, labels=y, args=args, ratio=2.2, do_transform=True)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Step 3: Make a dataset\n",
    "Next we are going to construct a classification dataset, analogous to MNIST but much smaller, using these signals. Note that the dataset gets saved as a small pickle file, so you don't have to build it from scratch every time."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Did or could not load data from ./mnist1d_data.pkl. Rebuilding dataset...\n"
     ]
    }
   ],
   "source": [
    "args = get_dataset_args()\n",
    "data = get_dataset(args, path='./mnist1d_data.pkl', download=False, regenerate=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Successfully loaded data from ./mnist1d_data.pkl\n",
      "Examples in training set: 4000\n",
      "Examples in test set: 1000\n",
      "Length of each example: 40\n",
      "Number of classes: 10\n"
     ]
    }
   ],
   "source": [
    "args = get_dataset_args()\n",
    "data = get_dataset(args, path='./mnist1d_data.pkl', download=False, regenerate=False)\n",
    "\n",
    "print(\"Examples in training set: {}\".format(len(data['y'])))\n",
    "print(\"Examples in test set: {}\".format(len(data['y_test'])))\n",
    "print(\"Length of each example: {}\".format(data['x'].shape[-1]))\n",
    "print(\"Number of classes: {}\".format(len(data['templates']['y'])))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Step 3b: Download the baseline dataset\n",
    "It's always possible that the open-source libraries that we used to synthesize this dataset will change. Even if they change slightly, the dataset one obtains from running the code above may change slightly, making it hard for you and I to compare results.\n",
    "\n",
    "If you want to use MNIST-1D as a baseline, we suggest that you load it using the pickle file provided in the repo. The dataset contained in that file will serve as the one-and-only, original MNIST1D dataset. The link is https://github.com/greydanus/mnist1d/raw/master/mnist1d_data.pkl. The `mnist1d.get_dataset` function loads this file when `download` is set to `True`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "File already exists. Skipping download.\n",
      "Successfully loaded data from ./mnist1d_data.pkl\n"
     ]
    }
   ],
   "source": [
    "args = get_dataset_args()\n",
    "data = get_dataset(args, path='./mnist1d_data.pkl', download=True) # This is the default setting"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Step 4: Run baselines*\n",
    "\n",
    "We'll see that:\n",
    "* Nonlinear classifiers do better than linear classifiers.\n",
    "* Classifiers with transformation invariance and local connectivity (CNNs) do better than ones without these properties (MLPs).\n",
    "\n",
    "An additional note:\n",
    "* Humans do just a bit better than the other baselines (human performance is 95.8% on 500 test set examples). This suggests that shortcut learning and memorization don't help much.\n",
    "\n",
    "_*You may see slightly different numbers than the ones from the default run. This is due to the fact that you are using a GPU instead of CPU: PyTorch's random seeding function ``torch.cuda.manual_seed_all(seed)`` causes this difference. Your results will be reproducible between GPUs._"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'input_size': 40,\n",
       " 'output_size': 10,\n",
       " 'hidden_size': 256,\n",
       " 'learning_rate': 0.01,\n",
       " 'weight_decay': 0,\n",
       " 'batch_size': 100,\n",
       " 'total_steps': 8000,\n",
       " 'print_every': 1000,\n",
       " 'eval_every': 250,\n",
       " 'checkpoint_every': 1000,\n",
       " 'device': 'cpu',\n",
       " 'seed': 42}"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "get_model_args(as_dict=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initialized ConvBase model with 37785 parameters\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 1000, dt 2.53s, train_loss 3.141e-02, test_loss 2.531e-01, train_acc 99.0, test_acc 93.9\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 2000, dt 2.41s, train_loss 1.555e-01, test_loss 3.304e-01, train_acc 98.2, test_acc 92.4\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 3000, dt 2.42s, train_loss 1.127e-05, test_loss 2.167e-01, train_acc 100.0, test_acc 96.0\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 4000, dt 2.42s, train_loss 4.670e-06, test_loss 2.338e-01, train_acc 100.0, test_acc 96.1\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 5000, dt 2.40s, train_loss 2.403e-06, test_loss 2.463e-01, train_acc 100.0, test_acc 96.1\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 6000, dt 2.42s, train_loss 1.304e-06, test_loss 2.569e-01, train_acc 100.0, test_acc 96.1\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 7000, dt 2.41s, train_loss 7.033e-07, test_loss 2.661e-01, train_acc 100.0, test_acc 96.1\n",
      "0.01\n",
      "0.01\n",
      "Epoch    32: reducing learning rate of group 0 to 9.0000e-03.\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 8000, dt 2.37s, train_loss 4.053e-07, test_loss 2.746e-01, train_acc 100.0, test_acc 96.3\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 9000, dt 2.42s, train_loss 2.432e-07, test_loss 2.820e-01, train_acc 100.0, test_acc 96.4\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 10000, dt 2.42s, train_loss 1.454e-07, test_loss 2.899e-01, train_acc 100.0, test_acc 96.4\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 11000, dt 2.41s, train_loss 8.583e-08, test_loss 2.984e-01, train_acc 100.0, test_acc 96.3\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 12000, dt 2.46s, train_loss 5.245e-08, test_loss 3.069e-01, train_acc 100.0, test_acc 96.2\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "Epoch    53: reducing learning rate of group 0 to 8.1000e-03.\n",
      "0.008100000000000001\n",
      "step 13000, dt 2.45s, train_loss 2.980e-08, test_loss 3.155e-01, train_acc 100.0, test_acc 96.1\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 14000, dt 2.39s, train_loss 1.788e-08, test_loss 3.235e-01, train_acc 100.0, test_acc 96.1\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 15000, dt 2.35s, train_loss 1.073e-08, test_loss 3.318e-01, train_acc 100.0, test_acc 96.1\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 16000, dt 2.36s, train_loss 5.960e-09, test_loss 3.397e-01, train_acc 100.0, test_acc 96.2\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 17000, dt 2.41s, train_loss 3.576e-09, test_loss 3.471e-01, train_acc 100.0, test_acc 96.0\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 18000, dt 2.44s, train_loss 2.384e-09, test_loss 3.539e-01, train_acc 100.0, test_acc 96.1\n",
      "Epoch    74: reducing learning rate of group 0 to 7.2900e-03.\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 19000, dt 2.41s, train_loss 1.192e-09, test_loss 3.580e-01, train_acc 100.0, test_acc 96.0\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 20000, dt 2.41s, train_loss 0.000e+00, test_loss 3.605e-01, train_acc 100.0, test_acc 96.0\n",
      "\n",
      "Initialized GRUBase model with 5134 parameters\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 1000, dt 3.31s, train_loss 9.236e-02, test_loss 3.444e-01, train_acc 95.0, test_acc 88.0\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 2000, dt 3.24s, train_loss 6.171e-02, test_loss 4.258e-01, train_acc 97.3, test_acc 88.3\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 3000, dt 3.36s, train_loss 2.165e-02, test_loss 4.546e-01, train_acc 98.5, test_acc 89.7\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 4000, dt 3.26s, train_loss 1.407e-02, test_loss 5.326e-01, train_acc 99.0, test_acc 88.1\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 5000, dt 3.29s, train_loss 2.277e-02, test_loss 4.744e-01, train_acc 99.8, test_acc 90.1\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "Epoch    25: reducing learning rate of group 0 to 9.0000e-03.\n",
      "0.009000000000000001\n",
      "step 6000, dt 3.26s, train_loss 4.753e-03, test_loss 5.166e-01, train_acc 99.8, test_acc 90.6\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 7000, dt 3.24s, train_loss 2.660e-03, test_loss 5.459e-01, train_acc 99.8, test_acc 90.6\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 8000, dt 3.29s, train_loss 1.286e-03, test_loss 5.703e-01, train_acc 99.8, test_acc 90.7\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 9000, dt 3.26s, train_loss 6.517e-04, test_loss 5.909e-01, train_acc 100.0, test_acc 91.1\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 10000, dt 3.28s, train_loss 3.690e-04, test_loss 6.193e-01, train_acc 100.0, test_acc 90.9\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 11000, dt 3.30s, train_loss 2.276e-04, test_loss 6.537e-01, train_acc 100.0, test_acc 91.1\n",
      "Epoch    46: reducing learning rate of group 0 to 8.1000e-03.\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 12000, dt 3.29s, train_loss 1.470e-04, test_loss 6.903e-01, train_acc 100.0, test_acc 91.0\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 13000, dt 3.30s, train_loss 1.021e-04, test_loss 7.329e-01, train_acc 100.0, test_acc 90.7\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 14000, dt 3.28s, train_loss 1.307e-02, test_loss 8.072e-01, train_acc 97.8, test_acc 88.5\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 15000, dt 3.54s, train_loss 1.146e-04, test_loss 5.621e-01, train_acc 100.0, test_acc 92.0\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 16000, dt 3.26s, train_loss 9.763e-05, test_loss 5.752e-01, train_acc 100.0, test_acc 92.1\n",
      "0.008100000000000001\n",
      "Epoch    67: reducing learning rate of group 0 to 7.2900e-03.\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 17000, dt 3.35s, train_loss 8.833e-05, test_loss 5.886e-01, train_acc 100.0, test_acc 92.2\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 18000, dt 3.41s, train_loss 7.845e-05, test_loss 6.032e-01, train_acc 100.0, test_acc 92.4\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 19000, dt 3.53s, train_loss 6.700e-05, test_loss 6.203e-01, train_acc 100.0, test_acc 92.1\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 20000, dt 3.38s, train_loss 5.528e-05, test_loss 6.397e-01, train_acc 100.0, test_acc 91.8\n",
      "\n",
      "Initialized MLPBase model with 15210 parameters\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 1000, dt 1.33s, train_loss 2.786e-01, test_loss 1.826e+00, train_acc 90.7, test_acc 60.6\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 2000, dt 1.35s, train_loss 1.520e-01, test_loss 2.483e+00, train_acc 97.9, test_acc 64.8\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 3000, dt 1.33s, train_loss 1.286e-01, test_loss 3.414e+00, train_acc 97.3, test_acc 64.0\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 4000, dt 1.33s, train_loss 1.861e-04, test_loss 3.534e+00, train_acc 100.0, test_acc 64.4\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 5000, dt 1.33s, train_loss 8.446e-05, test_loss 3.596e+00, train_acc 100.0, test_acc 64.4\n",
      "0.01\n",
      "Epoch    23: reducing learning rate of group 0 to 9.0000e-03.\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 6000, dt 1.31s, train_loss 5.193e-05, test_loss 3.652e+00, train_acc 100.0, test_acc 64.8\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 7000, dt 1.31s, train_loss 3.386e-05, test_loss 3.711e+00, train_acc 100.0, test_acc 64.7\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 8000, dt 1.31s, train_loss 2.204e-05, test_loss 3.772e+00, train_acc 100.0, test_acc 65.1\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 9000, dt 1.30s, train_loss 1.410e-05, test_loss 3.848e+00, train_acc 100.0, test_acc 65.0\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 10000, dt 1.31s, train_loss 9.136e-06, test_loss 3.926e+00, train_acc 100.0, test_acc 65.0\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "Epoch    44: reducing learning rate of group 0 to 8.1000e-03.\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 11000, dt 1.28s, train_loss 5.978e-06, test_loss 4.008e+00, train_acc 100.0, test_acc 64.6\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 12000, dt 1.26s, train_loss 4.021e-06, test_loss 4.089e+00, train_acc 100.0, test_acc 64.6\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 13000, dt 1.30s, train_loss 2.631e-06, test_loss 4.176e+00, train_acc 100.0, test_acc 64.7\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 14000, dt 1.29s, train_loss 1.568e-06, test_loss 4.266e+00, train_acc 100.0, test_acc 64.8\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 15000, dt 1.31s, train_loss 9.739e-07, test_loss 4.366e+00, train_acc 100.0, test_acc 64.7\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "Epoch    65: reducing learning rate of group 0 to 7.2900e-03.\n",
      "0.007290000000000001\n",
      "step 16000, dt 1.32s, train_loss 6.056e-07, test_loss 4.474e+00, train_acc 100.0, test_acc 65.1\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 17000, dt 1.31s, train_loss 3.850e-07, test_loss 4.575e+00, train_acc 100.0, test_acc 65.4\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 18000, dt 1.32s, train_loss 2.420e-07, test_loss 4.679e+00, train_acc 100.0, test_acc 65.4\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 19000, dt 1.32s, train_loss 1.442e-07, test_loss 4.790e+00, train_acc 100.0, test_acc 65.3\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 20000, dt 1.31s, train_loss 8.941e-08, test_loss 4.896e+00, train_acc 100.0, test_acc 65.2\n",
      "\n",
      "Initialized LinearBase model with 410 parameters\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 1000, dt 0.63s, train_loss 1.537e+00, test_loss 1.663e+00, train_acc 37.0, test_acc 32.4\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 2000, dt 0.61s, train_loss 1.533e+00, test_loss 1.667e+00, train_acc 37.1, test_acc 32.0\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 3000, dt 0.63s, train_loss 1.534e+00, test_loss 1.670e+00, train_acc 37.1, test_acc 31.6\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 4000, dt 0.63s, train_loss 1.534e+00, test_loss 1.671e+00, train_acc 37.1, test_acc 31.3\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "step 5000, dt 0.63s, train_loss 1.535e+00, test_loss 1.672e+00, train_acc 37.2, test_acc 31.7\n",
      "0.01\n",
      "0.01\n",
      "0.01\n",
      "Epoch    25: reducing learning rate of group 0 to 9.0000e-03.\n",
      "0.009000000000000001\n",
      "step 6000, dt 0.58s, train_loss 1.535e+00, test_loss 1.673e+00, train_acc 37.2, test_acc 31.7\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 7000, dt 0.60s, train_loss 1.533e+00, test_loss 1.671e+00, train_acc 37.4, test_acc 31.6\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 8000, dt 0.63s, train_loss 1.533e+00, test_loss 1.671e+00, train_acc 37.4, test_acc 31.5\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 9000, dt 0.62s, train_loss 1.534e+00, test_loss 1.671e+00, train_acc 37.4, test_acc 31.5\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 10000, dt 0.62s, train_loss 1.534e+00, test_loss 1.672e+00, train_acc 37.5, test_acc 31.5\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "0.009000000000000001\n",
      "step 11000, dt 0.63s, train_loss 1.534e+00, test_loss 1.672e+00, train_acc 37.4, test_acc 31.5\n",
      "Epoch    46: reducing learning rate of group 0 to 8.1000e-03.\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 12000, dt 0.62s, train_loss 1.533e+00, test_loss 1.670e+00, train_acc 37.9, test_acc 31.5\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 13000, dt 0.59s, train_loss 1.534e+00, test_loss 1.670e+00, train_acc 38.0, test_acc 31.5\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 14000, dt 0.62s, train_loss 1.534e+00, test_loss 1.670e+00, train_acc 38.0, test_acc 31.5\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 15000, dt 0.57s, train_loss 1.534e+00, test_loss 1.670e+00, train_acc 37.9, test_acc 31.5\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "0.008100000000000001\n",
      "step 16000, dt 0.60s, train_loss 1.534e+00, test_loss 1.670e+00, train_acc 37.9, test_acc 31.5\n",
      "0.008100000000000001\n",
      "Epoch    67: reducing learning rate of group 0 to 7.2900e-03.\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 17000, dt 0.63s, train_loss 1.535e+00, test_loss 1.669e+00, train_acc 37.9, test_acc 31.3\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 18000, dt 0.63s, train_loss 1.535e+00, test_loss 1.669e+00, train_acc 37.8, test_acc 31.3\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 19000, dt 0.63s, train_loss 1.535e+00, test_loss 1.669e+00, train_acc 37.8, test_acc 31.2\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "0.007290000000000001\n",
      "step 20000, dt 0.60s, train_loss 1.535e+00, test_loss 1.669e+00, train_acc 37.8, test_acc 31.2\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Note: if you run on a GPU, the walltimes will be _much_ lower\n",
    "args = get_model_args()\n",
    "args.total_steps = 20000\n",
    "args.device = DEVICE\n",
    "\n",
    "set_seed(args.seed)\n",
    "model = ConvBase(output_size=args.output_size)\n",
    "results_cnn = train_model(data, model, args)\n",
    "print()\n",
    "\n",
    "set_seed(args.seed)\n",
    "model = GRUBase(input_size=1, output_size=args.output_size)\n",
    "results_gru = train_model(data, model, args)\n",
    "print()\n",
    "\n",
    "set_seed(args.seed)\n",
    "model = MLPBase(args.input_size, args.output_size)\n",
    "results_mlp = train_model(data, model, args)\n",
    "print()\n",
    "\n",
    "set_seed(args.seed)\n",
    "model = LinearBase(args.input_size, args.output_size)\n",
    "results_lin = train_model(data, model, args)\n",
    "print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 615x450 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(4.1, 3), dpi=150)\n",
    "\n",
    "plt.subplot(1,1,1)\n",
    "t = range(0, args.total_steps+1, args.eval_every)\n",
    "plt.plot(t, results_lin['test_acc'], 'r-', label='logistic')\n",
    "plt.plot(t, results_mlp['test_acc'], 'g-', label='mlp')\n",
    "plt.plot(t, results_cnn['test_acc'], 'b-', label='cnn')\n",
    "plt.plot(t, results_gru['test_acc'], 'c-', label='gru')\n",
    "\n",
    "plt.plot(t, [95.8]*len(t), 'k-', label='human')\n",
    "\n",
    "plt.title('Test accuracy') ; plt.xlabel(\"Train step\") ; plt.legend(fontsize=6.5, ncol=5, loc='lower right')\n",
    "plt.xticks(fontsize=9) ; plt.yticks(fontsize=9)\n",
    "plt.ylim(0,105)\n",
    "plt.tight_layout() ; plt.show()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Step 5: Conclusions\n",
    "We believe this dataset is ideal for performing small-scale, \"science of deep learning\" experiments.\n",
    "\n",
    "Try it out, do something interesting, and share it with us via a Colab. We would like to feature your work in our README."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}