Replace a layer on LSTM

Dear all,

I am trying to create a weithed LSTM to Sequence-to-sequence classification. So first I created de LSTM.

numFeatures = 1;
numHiddenUnits = 200;
numClasses = 11;

layers = [ ...
    sequenceInputLayer(numFeatures)
    lstmLayer(numHiddenUnits,'OutputMode','sequence')
    fullyConnectedLayer(numClasses)
    softmaxLayer
    classificationLayer('Name','classoutput')];

Then I created the weighted layer as I found in:

https://es.mathworks.com/help/deeplearning/ug/create-custom-weighted-cross-entropy-classification-layer.html

https://es.mathworks.com/matlabcentral/answers/434918-weighted-classification-layer-for-time-series-lstm?s_tid=mlc_ans_email_view#comment_649462

classdef weightedClassificationLayer  < nnet.layer.ClassificationLayer
        
    properties
        % (Optional) Layer properties.
        ClassWeights
    end
 
    methods
        function layer = weightedClassificationLayer(classWeights,name)           
            % layer = weightedClassificationLayer(classWeights) creates a
            % weighted cross entropy loss layer. classWeights is a row
            % vector of weights corresponding to the classes in the order
            % that they appear in the training data.
            % 
            % layer = weightedClassificationLayer(classWeights, name)
            % additionally specifies the layer name. 

            % Set class weights
            layer.ClassWeights = classWeights;

            % Set layer name
            if nargin == 2
                layer.Name = name;
            end

            % Set layer description
            layer.Description = 'Weighted cross entropy';
        end

        function loss = forwardLoss(layer, Y, T)
            % loss = forwardLoss(layer, Y, T) returns the weighted cross
            % entropy loss between the predictions Y and the training
            % targets T.
            % Find observation and sequence dimensions of Y
            [~, N, S] = size(Y);
            
            % Reshape ClassWeights to KxNxS
            W = repmat(layer.ClassWeights(:), 1, N, S);
            
            % Compute the loss
            loss = -sum( W(:).*T(:).*log(Y(:)) )/N;
        end
        
        function dLdY = backwardLoss(layer, Y, T)
            % dLdY = backwardLoss(layer, Y, T) returns the derivatives of
            % the weighted cross entropy loss with respect to the
            % predictions Y.
            % Find observation and sequence dimensions of Y
            [~, N, S] = size(Y);
            
            % Reshape ClassWeights to KxNxS
            W = repmat(layer.ClassWeights(:), 1, N, S);
            
            % Compute the derivative
            dLdY = -(W.*T./Y)/N;
        end
    end
end

And

classWeights =[0.05 0.05 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1];
wLayer = weightedClassificationLayer(classWeights);

But when I try to replace the classification layer following the CNN example:

https://es.mathworks.com/help/vision/examples/semantic-segmentation-using-deep-learning.html

layers = replaceLayer(layers,"classoutput",wLayer);

It appears the error:

Check for missing argument or incorrect argument data type in call to function 'replaceLayer'

Can anyone help me?

Answer 1

[Ayush Laddha]

From your explanation, I do infer that you wanted to replace the layer using replaceLayer function. The error message clearly states the reason for error as well. You receive that error because the first argument to the replaceLayer function should be a layerGraph whereas you are trying to provide a layers array to it. You should convert your layers array to a layerGraph object and then use the replaceLayer function.

You can follow the documentation for further reference –

https://in.mathworks.com/help/deeplearning/ref/nnet.cnn.layergraph.html

https://in.mathworks.com/help/deeplearning/ref/replacelayer.html

[Jesus Balado Frias]

Yes, thanks