Invalid training data. Predictors and responses must have the same number of observations.

I wan to train a LSTM.

But I get Error:

Error using trainNetwork (line 191)

Invalid training data. Predictors and responses must have the same number of observations.

layers = [ ...
    sequenceInputLayer(6)
    lstmLayer(120,'OutputMode','last')
    fullyConnectedLayer(2)
    softmaxLayer
    classificationLayer];


options = trainingOptions('adam', ...
    'MaxEpochs',20, ...
    'MiniBatchSize',32, ...
    'GradientThreshold',1, ...
    'InitialLearnRate',0.005, ...
    'Shuffle','every-epoch', ...
    'Verbose',0, ...
    'Plots','training-progress');

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

Answer 1

[Matt J]

Your XTrain shouldn't be a 100x6 cell. It should be a 100x1 cell where each XTrain{i} is a matrix with 6 rows. Example,

layers = [ ...
    sequenceInputLayer(6)
    lstmLayer(120,'OutputMode','last')
    fullyConnectedLayer(2)
    softmaxLayer
    classificationLayer];

for i=1:100
 XTrain{i,1} = rand(6,randi(20));
end

YTrain = categorical(randi([0,1],100,1));


whos YTrain

  Name          Size            Bytes  Class          Attributes

  YTrain      100x1               346  categorical              

XTrain,

XTrain = 100×1 cell array
    {6×20 double}
    {6×16 double}
    {6×11 double}
    {6×2  double}
    {6×18 double}
    {6×8  double}
    {6×13 double}
    {6×17 double}
    {6×19 double}
    {6×6  double}
    {6×1  double}
    {6×1  double}
    {6×17 double}
    {6×10 double}
    {6×5  double}
    {6×4  double}

options = trainingOptions('adam', ...
    'MaxEpochs',20, ...
    'MiniBatchSize',32, ...
    'GradientThreshold',1, ...
    'InitialLearnRate',0.005, ...
    'Shuffle','every-epoch', ...
    'Verbose',1, ...
    'Plots','none');

net = trainNetwork(XTrain, YTrain, 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 |       46.88% |       0.7000 |          0.0050 |
|      17 |          50 |       00:00:01 |       71.88% |       0.6504 |          0.0050 |
|      20 |          60 |       00:00:01 |       53.12% |       0.6374 |          0.0050 |
|========================================================================================|
Training finished: Max epochs completed.

net = 
  SeriesNetwork with properties:

         Layers: [5×1 nnet.cnn.layer.Layer]
     InputNames: {'sequenceinput'}
    OutputNames: {'classoutput'}

[Matt J]

Consider also migrating from trainNetwork to trainnet.

[Bahadir]

When ı try trainnet, I get the error.

Caused by:

Layer 'classoutput': Detected output layer. The network must not have output layers.

[Matt J]

The error is complaining that you have not removed the output layer (classificationLayer) from your layers array. Output layers do not belong in the network when training with trainnet, because the loss function is separately specified to trainnet using the lossFcn input parameter.