using LSTM nets for classification with multiple outputs

I'm using LSTM nets for classification.

I would like to have 3 outputs of 3 values (-1 0 +1)

Apparently the Matlab framework for that nets accepts only one output. In this case it should have 27 values (3^3), but it adds complications.

Any suggestion?

Giuseppe Menga

Answer 1

[Shreeya]

To build an LSTM based neural netowkr with three prediction classes, create a layer array containing a sequence input layer, an LSTM layer, a fully connected layer, a softmax layer, and a classification output layer. Further, set the size of the sequence input layer to the number of features of the input data and the size of the fully connected layer to the number of prediction classes classes.

Refer to the link below for more details:

https://www.mathworks.com/help/deeplearning/ug/long-short-term-memory-networks.html#mw_9cf7b3dc-bc5e-40f2-a20a-82bdd6df1b03

[Giuseppe Menga]

Thanks for the answer, I saw the link you indicated.

I didn't understain the difference between the two examples of classification:

To create an LSTM network for sequence-to-label classification

lstmLayer(numHiddenUnits,'OutputMode','last')

To create an LSTM network for sequence-to-sequence classification

lstmLayer(numHiddenUnits,'OutputMode','sequence').

I used sequence-to-label classification, but to apply the classification in real time I will test the other.

But you didn't answer to my basic question:

using three outputs, each with three levels of classification, or transforming them in only one output with twentyseven levels of classiicaton.

I suspect that I have to transform the problem with only one output with twentyseven levels, as I haven't found any other example.

Have you some observation?

Giuseppe