How do I fix this error?

I keep getting this error

"Incompatible input and output sequence lengths. The network must return sequences with the same length as the

input data or a sequence with length one."

when I try to run this 1D CNN:

(For reference I am trying to analyse 320 signals with 2000 time points each)

inputSize = size(xTrain());
numClasses = 2; % Cracked or not cracked
layers = [
    sequenceInputLayer(time_points,'Name', 'input','MinLength', 2000)
    convolution1dLayer(32,20,'Padding', 'same')
    batchNormalizationLayer
    reluLayer
    maxPooling1dLayer(2,'Stride',2)
    convolution1dLayer(64,10,'Padding', 'same')
    batchNormalizationLayer
    reluLayer
    maxPooling1dLayer(2,'Stride',2)
    fullyConnectedLayer(256)
    reluLayer
    dropoutLayer(0.5)
    fullyConnectedLayer(numClasses)
    softmaxLayer 
    classificationLayer];
%----------------------Define training options-----------------------------
options = trainingOptions('adam', ...
    'InitialLearnRate',0.01, ...
    'MaxEpochs',50, ...
    'MiniBatchSize',128, ...
    'Shuffle','every-epoch', ...
    'ValidationData',{xTest,yTest}, ...
    'ValidationFrequency',5, ...
    'Verbose',false, ...
    'Plots','training-progress');
% Train the network
net = trainNetwork(xTrain, yTrain, layers, options);

This is my first time using these neural networks so I have left all the layers and options in case there is an error in those. Any help would be useful.

Thank you

Answer 1

[Matt J]

Shouldn't your input layer be,

sequenceInputLayer(inputSize,'Name', 'input','MinLength', time_points)

[Nikolas Katsaros]

Thank you that seems to have fixed it. Can you explain why that worked? 'time_points' is set equal to 2000 and inputSize is equal to [320, 1] now. I thought inputSize is supposed to be equal to the number of features in the sequence?

[Matt J]

Yes, you have 320 features and 2000 time points.

[Nikolas Katsaros]

Okay, I am now getting an error saying that the input data to my fullyConnectedLayer is invalid as it cannot have both spatial and temporal dimensions. How can this be possible if each sequence is only 1D (2000,1)? I have tried taking the real component of each value but this doesn't change anything

[Matt J]

Seems like a good question. You should post it! However, if you do, you should attach an example of the training input that triggers the error, so that people can reproduce and study it.

[Nikolas Katsaros]

Will do - thank you for your help!