Function to trainNetwork returns an unexpected error

My code returns the following error for this function call - What is the fix for this?

net = trainNetwork(X_train, categorical(y_train), layers, options);

Error using trainNetwork (line 191)

Too many input arguments.

Error in LSTMGomz (line 63)

net = trainNetwork(X_train, categorical(y_train), layers, options);

Caused by:

Error using nnet.internal.cnn.trainNetwork.DLTInputParser>iParseInputArguments (line 75)

Too many input arguments.

[Matt J]

You would have to attach a .mat file providing inputs X_train, categorical(y_train), layers, options for us to run with.

[Ernest Modise - Kgamane]

Let me attach,

Answer 1

[Matt J]

Your X_train and y_train data were in some weird format that trainNetwork cannot recognize. Try this instead,

Xdata = num2cell(readmatrix('LSTMdataIn.xlsx')',1)';

N=200;
train_ratio=0.8;
split_index=round(train_ratio*N);

inputSize = height(Xdata{1});  % Number of features in the input data
numClasses = height(Xdata)/N;  % Number of categories

Xdata=reshape(Xdata,N,numClasses);
ydata=repmat(1:numClasses,N,1);

X_train=Xdata(1:split_index,:);
y_train=ydata(1:split_index,:);

X_test=Xdata(split_index+1:end,:);
y_test=ydata(1:split_index+1:end,:);


layers = [
    sequenceInputLayer(inputSize)
    lstmLayer(100, 'OutputMode', 'last')
    fullyConnectedLayer(numClasses)
    softmaxLayer
    classificationLayer
];

options = trainingOptions('adam', 'MaxEpochs', 100);
net = trainNetwork(X_train(:), categorical(y_train(:)), layers, options);

Training on single CPU.
|========================================================================================|
|  Epoch  |  Iteration  |  Time Elapsed  |  Mini-batch  |  Mini-batch  |  Base Learning  |
|         |             |   (hh:mm:ss)   |   Accuracy   |     Loss     |      Rate       |
|========================================================================================|
|       1 |           1 |       00:00:00 |       20.31% |       1.6082 |          0.0010 |
|       9 |          50 |       00:00:00 |       79.69% |       0.4997 |          0.0010 |
|      17 |         100 |       00:00:00 |       82.81% |       0.2851 |          0.0010 |
|      25 |         150 |       00:00:01 |       76.56% |       0.3004 |          0.0010 |
|      34 |         200 |       00:00:01 |       79.69% |       0.2844 |          0.0010 |
|      42 |         250 |       00:00:01 |       82.81% |       0.2591 |          0.0010 |
|      50 |         300 |       00:00:01 |       76.56% |       0.2918 |          0.0010 |
|      59 |         350 |       00:00:02 |       79.69% |       0.2794 |          0.0010 |
|      67 |         400 |       00:00:02 |       82.81% |       0.2565 |          0.0010 |
|      75 |         450 |       00:00:02 |       76.56% |       0.2902 |          0.0010 |
|      84 |         500 |       00:00:03 |       79.69% |       0.2782 |          0.0010 |
|      92 |         550 |       00:00:03 |       82.81% |       0.2557 |          0.0010 |
|     100 |         600 |       00:00:03 |       76.56% |       0.2895 |          0.0010 |
|========================================================================================|
Training finished: Max epochs completed.

[Matt J]

Also consider migrating to trainnet() instead of tranNetwork(), since the latter seems ot be deprecated.

[Ernest Modise - Kgamane]

Hi Mat, You have created an interesting data structure for this purpose. I would like to spend time on learning how to configure the data structure. Please send me tops to look at.

[Matt J]

It's just a cell array of numeric data. You had tables nested inside cells, I think.