I'm trying to implement your approach and I have a question about the estimation procedure. Is it correct that you don't use batches and always feed in the whole training set as a whole, compute the loss at t = T and backpropagate once every epoch? Thus, for the LSTMs you feed the macro data from t=1 until t=T and then repeat the LSTM output N times? Thanks & regards
I'm trying to implement your approach and I have a question about the estimation procedure. Is it correct that you don't use batches and always feed in the whole training set as a whole, compute the loss at t = T and backpropagate once every epoch? Thus, for the LSTMs you feed the macro data from t=1 until t=T and then repeat the LSTM output N times? Thanks & regards