sequence to sequence classification - invalid training data

Hi, I'm training a classification network:

%% def

inputSize = 1;

numHiddenUnits = 200;

numClasses = 2;

layers = [ ...

sequenceInputLayer(inputSize)

bilstmLayer(numHiddenUnits,'OutputMode','last')

dropoutLayer(0.5,"Name","dropout") %

fullyConnectedLayer(numClasses)

softmaxLayer

classificationLayer]

I generated my data, which are:

• the labels, that are arrays long 4000 elements (which are categorical), memorized in a cell array like this:

• the signals, each long 4000 samples, memorized in a cell array, like this:

I tried to ''stick'' to this example (https://it.mathworks.com/help/deeplearning/ug/sequence-to-sequence-classification-using-deep-learning.html) to understand how to prepare the cell array for the labels, but when I use trainNetwork it gives me this error:

Error using trainNetwork (line 184)

Invalid training data. For image, sequence-to-label, and feature classification tasks, responses must be categorical.

I can't understand what's wrong, I tried to prepare the data in a different way, but it didn't work anyway, can you suggest what to try?

Answer 1

[Srivardhan Gadila]

The issue here is that you are using the bilstmLayer with 'OutputMode' set to 'last', which means that it would only output the last time step of the sequence and hence your network output would not be a 1x4000 categorical but instead 1x1 categorical.

Hence set the 'OutputMode' to 'sequence' and execute the code, it should work fine.

Refer to the documentation of bilstmLayer & description of OutputMode property for more information.

[Floriana Vasile]

Thank you very much!