[x] instead of arrays, save the tensors in the memory experience buffer so that the retrieval be faster (will it improve or worsen??) -> we have already paid the cost of transferring to tensor when choosing action!
[x] avoids the copy from CPU to GPU which improves performance hugely
in ----> def separate_out_data_types(self, experiences):
[x] refactor time-to-learn for Multi_Agent for performance improvement.