Consultation on AR-LSTM prediction issues

[Reo-yan]

Hi, teacher marcopeix：
I'm currently studying the book "Time-Series-Forecasting in-Python". I really like the time series prediction method and thanks for your github code sharing.

 However, there are also some questions I need during my study process, which puzzles me a lot.

 Taking Chapter 17 of this book, AR-LSTM, as an example, we gave a model evaluation on the data and proved that AR-LSTM is a relatively appropriate model. This is comforting. But the problem that follows is, how can I use the AR-LSTM model to predict the test dataset and getting the predicted values.

After running the code in CH17.ipynb,I continued to run the following code:

The inverse minmaxscaler method I used is as follows：

scaler = MinMaxScaler()
scaler.fit(train_df)
Restore using the scaler.inverse_transform() method.

The prediction method for the test set is as follows:

test_predictions_scaled = AR_LSTM.predict(multi_window.test)

The resulting test_predictions_scaled is a numpy array with dimensions of: (1709, 24, 5)

test_predictions_scaled sample data:

array([[[0.09379353, 0.07652892, 0.09657007, 0.11072036, 0.11131918],
[0.04895833, 0.06484775, 0.03994162, 0.0662628 , 0.05923389],
[0.03269311, 0.06338501, 0.03825416, 0.06969497, 0.06171504],
...,
[0.24172495, 0.24662721, 0.25657123, 0.2510248 , 0.25501385],
[0.16440384, 0.16854419, 0.18737686, 0.17829569, 0.17556214],
[0.09430677, 0.09476103, 0.11893123, 0.10730372, 0.10237038]],
...,
[0.6324826 , 0.62707186, 0.62948847, 0.6140796 , 0.62807786],
[0.6725644 , 0.6729528 , 0.65185994, 0.66380966, 0.67474675],
[0.6586775 , 0.6655749 , 0.6467061 , 0.658816 , 0.6668321 ]]],
shape=(1709, 24, 5), dtype=float32)

But in fact, our test set is a dataframe with a dimension of 1756 rows and 5 columns.

I very much hope that the prediction result of AR_LSTM is a value of 1756 rows and add 'AR-LSTM' column. However, it seems that ar_lstm.predict cannot return what I want. Is there any way to achieve this?

I guess that by doing some kind of combinatorial transformation on test_predictions_scaled, combined with scaler.inverse_transform(), I should be able to get the result I want, but I have never found it.

I hope you can help me solve my doubts. Thank you again.