Could someone please explain me. Why the shape of hidden in GRU or LSTM is not same output shape? When we set batch_first is True

1

I am so confusing. I try to put my data to GRU/LSTM model with batch_first is true.
I reshape my data to be of shape (batch,seq, features). I expect the model will give the output have the same shape. Yes, the output of the model has the same shape (batch,seq, features), but I have no idea. Why last hidden output still to be of shape (batch,seq, features). Could you please someone explain to me?
This is my code.

2019-11-03_16-19-37
2019-11-03_16-19-37790×775 106 KB

2

The first dimension isn’t sequence length but num layers * num directions (2 for bidirectional else 1). The documentation for nn.GRU is a bit long and the output and shape subsections which both describe the output shape are towards the end.

Best regards

Thomas

P.S.: Don’t use Variable, sources that still use that have not been updated for a year.
P.P.S.: If you use triple backticks (```) before and after your code, it formats nicely as text, too. That makes it much easier to read than images are.