DeepESN for multivariate time-series classification problem

[rjpg]

hello,

I would like to use this implementation for a multivariable time series classification.

My dataset is composed of several examples of 9 time-series segments with 128 steps. Each segment belongs to one of 5 classes.

Input shape (2128,128,9) -> (number of examples, time-steps, variables) output shape (2128,1) - > (number of examples, class) (the class is one-hot-encoded so I use categorical-cross-entropy with 5 neurons output layer)

I have it working with a LSTMs based model.

I would like to test with deepESN to see if there is any improvement.

If you can help to provide some lights/base code to do this adaptation it would be nice.

Thanks in advance, Rui