rikdz
commented
3 years ago sorry this code is so messy. I think the point of these two lines is to get the final hidden state from the top layer of the encoder BiLSTM and cat both directions together into a single vector. What they seem to actually do is concatenate the hidden states for all but the first layer of the BiLSTM. This results in the same behavior when layers==2, as is used in this codebase.
hustcxx
def forward(self,batch,pad=True): batch,phlens,batch_lens = batch batch_lens = tuple(batchlens.tolist())#list-to-tuple ,enc = self.seqenc((batch,phlens)) enc = enc[:,2:] enc = torch.cat([enc[:,i] for i in range(enc.size(1))],1) m = max(batch_lens) encs = [self.pad(x,m) for x in enc.split(batch_lens)] out = torch.stack(encs,0) return out i don't understand why use enc = enc[:,2:] and enc = torch.cat([enc[:,i] for i in range(enc.size(1))],1)??? anybody can explain to me?thanks .