Learning-from-memory

To reproduce our code you need to modify some of our improvements into stable-baselines3