LSTM: Multi Input to Single Output Regression - Wrong Input form

Hello,

i'm trying to map several inputs (human biosignals and stats) to a single value (concerning the heartrate).

I'm trying to understand what form my inputs needs to be, but i feel really stuck, since my lstm works with "dummy" input- matrices, but it seems to me, that i'm still missing something completely.

layers = [ ...
    sequenceInputLayer(7)
    lstmLayer(numHiddenUnits_lstm,'OutputMode','sequence')
    lstmLayer(numHiddenUnits_lstm,'OutputMode','sequence')
    fullyConnectedLayer(1)
    regressionLayer
    ];

maxEpochs = 15;
miniBatchSize = 1;
options = trainingOptions('adam', ...
    'MaxEpochs',maxEpochs, ...
    'MiniBatchSize',miniBatchSize, ...
    'InitialLearnRate',0.01, ...
    'GradientThreshold',1, ...
    'Shuffle','never', ...
    'Plots','training-progress',...
    'ValidationFrequency',4, ...
    'Verbose',0);

net = trainNetwork(LSTMARR3,LSTMARR_targets,layers,options);

My Input : LSTMARR3

My Output : LSTMARR_targets

Below i tried to show you how i formatted my data, so can you please help me how to feed my data to my network?

[Shuvasish Paul]

I also got the same problem anyone please help.

Answer 1

[Shuhei IKEDA]

It is not an answer. but I have a same ploblem!!

Could anyone help us?? ;(

Answer 2

[Linyu Lin]

First of all, I am not sure if you are having 29516 features with a single time sequence as inputs

or

You are having 1 feature with 29516 time sequence?

If it is the first case, all you need to do if swith the dimension from 1x29516 to 29516x1. It should solve the problem easily.

If it is the second case, I am afraid that MATLAB currently can not have different sequence length in the input and output for sequence regression problem.