Pytorch lstmcell issue regarding usage of lstmcell instead of into lstm

nlp
1

Hi.
I’m trying to refactor a code that uses lstm into lstmcell
it’s sniippet is:

def forward(self, sentence):

    embeds = self.embeddings(sentence)
    #
    out = out.view(self.sequence_len, embeds.size(0), -1)
    lstm_out, _ = self.lstm(sentence)
    None
    #
    # Run through the lstm.
    lstm_out, _ = self.lstm(embeds.view(len(sentence), 1, -1))

    # Activate the ReLU on the linear layer's output.
    x = self.activation(self.linear1(lstm_out[-1].view(1, -1)))

    # For the second linear layer.
    x = self.linear2(x)

    return F.log_softmax(x, dim=-1)

can u assist?
thanks.

2

Well, nn.LSTM processes the whole input sequence, whole nn.LSTMCell only a single time step. So you will need a loop over your sequence. The docs actually provide an example.

Apart from that, I’m rather sure that your use of view() messes up your data; have a look here.