Replay Buffer

[theovincent]

There are several ways of doing it:

• Parallel actors collect samples. Each actor collect one sample. Like in the last TP. (Problem colab is not efficient with parallel computing).
• In a class. Like in the TD1. I will go for this