A problem about implementation of the model

[halehdizaji]

Dear Shaw Liu,

Thanks for the code and for making it available.
I wanted to ask you about a problem I noticed in the model implementation, in the CTLSTM class, in the forward function you have written self.init_states(batch_size) at the start of every time step. I think it should be written out of this loop (before it). Because the states should be initialized once at the start of feeding the sequence. Thanks to let me know if I am wrong.