Embedding variable length sequences with extra features

nlp
1

Hi!

I am working on RNN model.

I have several sentences of different length-I want to embed them and combine with extra features-binary list that describes each word (not important for the question what is actually means :wink: ). This would serve as an input to LSTM.

I do know how to use padding, pack_padded_sequence etc and that recently a new method pack_sequence was intorduced.
However I do have a problem with the following issue:

  1. How can I embed variable length sequences (no padding)?

  2. If padding in embedding is mandatory, how can I concat extra binary features with my embedding matrix? Do I assume correctly that extra features should be embedded as well with invariant values for 0 and 1?