Input shape for the LSTM model

XTrain dataset (a cell) has a shape of 12x1 and each of 12 elements has 55x1 data. YTrain data set (as a matrix) has a shape of 12x55. How can I match the dimension of these datasets? The error msg: error msg " 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" is alway there

%%Train the Layers
numFeatures = 55; 
numHiddenUnits = 300; 
numResponses = 1; 

LSTM = [ ...
    sequenceInputLayer(numFeatures)
    lstmLayer(numHiddenUnits,'OutputMode','last')
    fullyConnectedLayer(numResponses)
    regressionLayer];

maxEpochs = 10
miniBatchSize = 100

options = trainingOptions('adam', ...
    'ExecutionEnvironment','auto', ...
    'MaxEpochs',maxEpochs, ...
    'MiniBatchSize',miniBatchSize, ...
    'GradientThreshold',1, ...
    'Verbose',false, ...
    'Plots','training-progress');

net = trainNetwork(XTrain,YTrain,LSTM,options)

Answer 1

[Milan Bansal]

Hi,

I understand you are facing an error when passing a "cell array" as input to the Long Short Term Memory (LSTM) Model.

According to given code, the input layer expects a sequence of features of dimension 55. Ensure that each element in the input array i.e "XTrain" has dimension of (55 × 1) and cell array has the dimension of (N × 1), where N is the number of sequences.

You may also convert the cell array to matrix using the below code.

A = cell2mat(C) % C is the cell array

The variable "numResponses" has value 1, which means that the output layer is expecting each output response to have dimension of 1, but each element in "YTrain" has a dimension of 55. Set the value of "numResponses" to 55.

Refer to the documentation link to know more about LSTM Models.

https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.lstmlayer.html