Try to make Time Series Prediction with simple neurons network supported by Keras.
Thanks to a wonderful tutorial published by machinlearningmaster,The code(Time Series prediction with Keras-predictmorethan1time-Copy1.ipynb) I published was mainly copied from that.
In that tutorial, the prediction time lenght is 1 time unit.The dataframe like this (Xt-2,Xt-1,Xt,Yt+1),Yt+1 is for prediction.For expample , i use the passed 4days data to predict the 5th day.
What I try to explore is that,I want to use the passed 4days data to predict the next 2 days'.
So the notebook shows the steps what i try to do.
However! If you look very carefully of the trainPredict data.
the first 3 array is: array([[ 128.60112 , 127.5030365 ], [ 121.16256714, 122.3662262 ], [ 144.46884155, 145.67802429]
the list inside [ 128.6,127.5 ] [121,2,122,3] does not like t+1 and t+2.
Instead, It looks like 2 probaly prediction for 1 unit.
What i means is [128.6,127.5] doesn't mean t+1 and t+2 prediction, it most possibly mean 2 possible prediction for t+1.
one output cell with 2dimension and 2 output cell with 1 dimension is different. The input dimension and the output dimension will be tricky for the NN.
It seems i should redisgn a new neurons network.seq2seq ? distributedtime wraper were the tools on hand right now. What your suggestion ? You can launch new issue and we discuss about it.
11.15 update Solution:I figured out how to implement the N time unit prediction, and mostly dropped that former code. Using keras time-distributed-Dense neurons structure, It performs acceptably to predict within 5 days. After 5days , the longer to predict, the lower accuracy. (CStimeSeries_TimeDistributed-step1shift24_LSTM-1115.ipynb). Using CNN1D + RNN could cause a more impressive accurate result.
An explicit explanation for the Timedistributed Dense is [here] (keras-team/keras#1029)