Implementing many-to-one RNN about rnn HOT 3 CLOSED

Yeah it's actually pretty easy to do.

mlp = nn.Sequential()
mlp:add(nn.Sequencer(nn.LSTM(100,100)))
mlp:add(nn.SelectTable(-1))
mlp:add(nn.Linear(100,10)
mlp:add(nn.LogSoftMax())

So the key is to use SelectTable(-1) to retrieve the last element of the sequence and feed into a classifier (or whatever you want).

from rnn.

Thanks! That's exactly what I was looking for - the SelectTable() is the key here.

from rnn.

@nvram83 Can you pl upload the complete code?

from rnn.