How to implement LSTM Time-series prediction using multi-features?

Hello everyone,

I have the attached code and the attached data file here. I want to modify that code to proceed time-series prediction for 1 output using 5 inputs.

When I the training finishes I get the following error:

The prediction sequences are of feature dimension 1 but the input layer expects sequences of feature dimension 4. 
Error in multi_lstmOMNI_noStand (line 110)
[net,YPred] = predictAndUpdateState(net,YTrain);  

Can you please tell me how to fix it?

I appreciate your help.

[OLUWAFEMI AJAYI]

Hello,

I am also having the same challenge using that code for time-series prediction for two input/one output. Please have you been able to fix the error?

[Mohamed Nedal]

Hi, have you been able to find a solution?

I'm still stuck with it ..

[Marcelo Olmedo]

Hello people; I found the problem; the key is in the correct loading of data as the published documents say. I attach my code and used tables so you don't have problems to run it; I upload data from excel to train and test. I do not use standardized data. The model fits quite well. Cheers

[Mohamed Nedal]

Great!, Thanks a lot

If you were going to forecast the future data (Y) that you don't have, and there are no data for input features yet (X; because they are in the future), What would you do in this case?

[Mohamed Nedal]

Hi @Marcelo,

I tried to add a few lines of code to predict new future values of the target output, here's what I added:

%% Forecast the Future 
net = resetState(net); 
Yforecast = []; 
numTimeStepsTest = numel(XTest) + 500;     % to forecast new 500 steps in the future 
for i = 1:numTimeStepsTest 
    [net, Yforecast(:,i)] = predictAndUpdateState(net, XTest(:,i), 'ExecutionEnvironment','cpu'); 
end 

but I got this error:

Conversion to double from cell is not possible.
Error in LSTM_multi_motores (line 82)
    [net, Yforecast(:,i)] = predictAndUpdateState(net, XTest(:,i),
    'ExecutionEnvironment','cpu'); 

Can you please tell me how to fix this part?

Answer 1

[Marcelo Olmedo]

Hello people; I found the problem; the key is in the correct loading of data as the published documents say. I attach my code and used tables so you don't have problems to run it; I upload data from excel to train and test. I do not use standardized data. The model fits quite well. Cheers

[Mohamed Nedal]

Great!, Thanks a lot

If you were going to forecast the future data (Y) that you don't have, and there are no data for input features yet (X; because they are in the future), What would you do in this case?

[Mohamed Nedal]

Hi @Marcelo,

I tried to add a few lines of code to predict new future values of the target output, here's what I added:

%% Forecast the Future 
net = resetState(net); 
Yforecast = []; 
numTimeStepsTest = numel(XTest) + 500;     % to forecast new 500 steps in the future 
for i = 1:numTimeStepsTest 
    [net, Yforecast(:,i)] = predictAndUpdateState(net, XTest(:,i), 'ExecutionEnvironment','cpu'); 
end 

but I got this error:

Conversion to double from cell is not possible.
Error in LSTM_multi_motores (line 82)
    [net, Yforecast(:,i)] = predictAndUpdateState(net, XTest(:,i),
    'ExecutionEnvironment','cpu'); 

Can you please tell me how to fix this part?

[Song Decn]

Thnks!!

[Asli]

Hi Marcelo, i am really new in this topic and i am trying to predict time series of Y depending on two external variables (X1 and X2). I have two questions. (i) Why don’t you use the “predictAndUpdateState” and why do you use the “predict” statement? Is it used for time series prediction? (ii) Why are the previous y values not input into this function? Only x values are input.