I am trying to perform a sequence labelling task where I have initialized embedding weights using Glove for a word. I also want to incorporate character level features into the model. I am able to train an LSTM over the character vector and save its final state. Should I use this as the character representation instead?
How can I append a character vector of dim say 5, corresponding to each word in the sentence dynamically (i.e, during run time) using Pytorch? So final input will be of dim = glove_dim + character_dim.
P.S: I am trying to recreate code exercise as suggested here.