Freezing Hidden layer Parameters in RNN

1

Hi,

I have trained encoder-decoder for some data. Now I want to use same weights in the hidden layers of encoder and decoder for different data. Only change I need in embedding layer of encoder & decoder and output layer of decoder.
Network is same as this
Any pointers, how to do this?

Thanks.

2

you can copy the .weight.data over from the Embedding layers of the trained encoder-decoder into the new one.