Array inputs have incompatible channel dimensions.

im error using trainNetwork. i have no idea, im newbie here. im using matlab 2022a

layers = [ ...
    sequenceInputLayer([n_samples,numFeatures])
    convolution1dLayer(filterSize,numFilters,'Padding','same')    
    eluLayer
    maxPooling1dLayer(3, 'Stride', 2)
    flattenLayer
    fullyConnectedLayer(numClasses)
    softmaxLayer
    classificationLayer];

miniBatchSize = 50;

options = trainingOptions("adam", ...
    MiniBatchSize=miniBatchSize, ...
    MaxEpochs=10, ...
    L2Regularization=0.00005,...
    Shuffle='every-epoch',...
    ValidationFrequency=10,...
    ValidationPatience=20,...% for early stop
    SequencePaddingDirection="left", ...
    ValidationData={XValidation,TValidation}, ...
    Plots='none', ...% or "training-progress" , 'none'
    Verbose=1);

net = trainNetwork(XTrain,TTrain,layers,options);

Answer 1

[Govind KM]

Hi Ahmad,

A common reason for the mentioned error is a mismatch between the dimensions of input data and the expected input dimensions specified by the network. Here are potential steps to debug the issue:

• Verify that the inputs to the trainnetwork function are passed in the required format, i.e format for sequences and format for responses.
• Ensure that the input data XTrain matches the dimensions specified in sequenceInputLayer, i.e. n_samples-by-numFeatures.
• Ensure that the filterSize and numFilters are appropriate for the data.
• Ensure that the sequencePaddingDirection has been set correctly.

A related MATLAB Answers post about the same error mentions that the issue is fixed in R2024a:

https://www.mathworks.com/matlabcentral/answers/2118461-c-rnn-regression-error-array-inputs-have-incompatible-channel-dimensions

If the issue still persists, providing the data used would help in debugging the issue further.

Hope this is helpful!