Hello guys, I have a project related to the time series prediction, I currently use a many2many model, and I want to replace it using a seq2seq model. Since pretty much all the models I can find are NLPs. I manage to build an simple seq2seq model myself, but I not sure if it is the correct way of doing it.
Here is an sample of what I am doing for my project.
Also question about “start” and “end” token added in the NLP model, I used the last element in the input dataset as the input for the first time step in the decoder to replace the “start” token, is this a correct way of doing it, and what about the ''end" token, should I just ignore it?