not enough look-back data in traning

[baont]

Hi @selimamrouni , when i debugged though your code. I saw that when you trained the model, it receives the state as a combination of a vector of m asset and the previous vector w. Is this a mistake since the paper states that the current state consist of previous 50 vectors of m and a previous w?