Hi I have a question about how to collect the correct result from a BI-LSTM module’s output.
Suppose I have a 10-length sequence feeding into a single-layer LSTM module with 100 hidden states:
lstm = nn.LSTM(5, 100, 1, bidirectional=True)
The output will be have dimension:
[10 (seq_length), 1 (batch), 200 (num_directions * hidden_size)]
# or according to the doc, can be viewed as
[10 (seq_length), 1 (batch), 2 (num_directions), 100 (hidden_size)]
If I want to get the 3rd (1-index) input’s output at both directions (two 100-dim vectors), how can I do it correctly?
I know output[2, 0] will give me a 200-dim vector. Does this 200 dim vector represent the output of 3rd input at both directions?
A thing bothering me is that when do reverse feeding, the 3rd (1-index) output vector is calculated from the 8th(1-index) input, right?
Will pytorch automatically take care of this and group output considering direction?
Thanks!