Open thombrey-vinil opened 6 years ago
I think it should work in general. But you might need to adjust the way you access and slice the dimensions, because the output of a bidirectional LSTM will have a different dimension than the output of a unidirectional LSTM, I suppose. See the PyTorch documentation for details. But I never tried it... Also, using a bidirectional LSTM might lead to even more overfitting.
Is it possible to implement a bidirectional LSTM just by changing parameter 'bidirectional = True' in the code?
Please let me know if this works.