Conv LSTM input size mismatch error

I have a cell array of dimension tIN{2000,1} and each element in the cell array is a 100x21 the ouput to that is tOUT{2000,1} with each element also of dimensino 100x21

when training I get the error Invalid training data. Responses must be a matrix of numeric responses, or a N-by-1 cell array of sequences, where N is

the number of sequences. The feature dimension of all sequences must be the same.

How do I fix this?

Error in CONv (line 23)

net = trainNetwork(tIN,tOUT,lgraph,options);

The current code is:

% Define Network Layers

layers = [sequenceInputLayer([100,21,1],'Name','input')

sequenceFoldingLayer('Name','fold')

convolution2dLayer(5,20,'Name','Conv')

maxPooling2dLayer([4 4],'Stride',2,'Name','max')

sequenceUnfoldingLayer('Name','unfold')

flattenLayer('Name','flat')

lstmLayer(100,'Name','lstm','OutputMode','Sequence')

dropoutLayer(0.2,'Name','drop')

fullyConnectedLayer(1,'Name','fc2')

regressionLayer('Name','output')];

lgraph = layerGraph(layers);

lgraph = connectLayers(lgraph,'fold/miniBatchSize','unfold/miniBatchSize');

%Training Options

options = trainingOptions('adam', ...

'InitialLearnRate',0.0001, ....

'ExecutionEnvironment','cpu', ...

'MaxEpochs',100, ...

'Plots','training-progress','Shuffle','every-epoch','L2Regularization',0.0005);

net = trainNetwork(tIN,tOUT,lgraph,options);

Answer 1

[Harsha Priya Daggubati]

Hi,

I guess you are trying to do sequence to sequence classification of your data using LSTM's. From the data you provided, I can infer you have a training set with 2000 samples, where each sample has 100 features, with 21 values for each feature. Similarly the Responses/labels is also a 2000 x 1 cell array.

I doubt the issue is with the elements of your responses being a cell array of 100 X 21. It is usually expected to be 1 X 21.

You can refer to HumanActivityTrain dataset in MATLAB to help you organise your data.

https://www.mathworks.com/help/deeplearning/ug/sequence-to-sequence-classification-using-deep-learning.html

[Shraddha Naidu]

I am actually trying to do a sequence to sequence regression so I need a 100x21 output. Is there a way around it?

Answer 2

[Harsha Priya Daggubati]

As far as I know, you will be able to assign one class at each time step based on the feature values. So you would need 100 X 21 response for each sample.

This example speaks the same too.

https://www.mathworks.com/help/deeplearning/ug/sequence-to-sequence-regression-using-deep-learning.html

[Shraddha Naidu]

This link follows a regression and not classification.

But it does not implement convolution?

Answer 3

[NGR MNFD]

Hello . I hope you have a good day. I sent the article to your service. I implemented the coding part in the MATLAB software, but to implement my network, two lines of setlayers, training MATLAB 2014 give me an error. What other function do you think I should replace? Do you think the codes I wrote are correct?( I used gait-in-neurodegenerative-disease-database in physionet website.) Thanks a lot