/Efficient-Unitary-RNN-Keras-Implementation

Keras Implementation for models of the EURNN

Primary LanguagePython

Efficient-Unitary-RNN Keras Implementation

This is a keras version of the EURNN that allows new models to be constructed by the line:

  • model.add(EURNNCell(80,input_shape=(history,d),return_sequences='true'))

##Setup: Note that the keras setup file must be switched to use tensorflow as the backend.

##Notes:

The output dimension is the same as the number of units. The architecture is simply: hidden = activation(Ux+Wh,bias) where the hidden is thus the output.

For original EUNN and EURNN code, see EUNN-tensorflow

For more detail, see paper