issue about RPPO

[tingtingLiuLiu]

Hello, when the LSTM network is used as the basic unit, the environmental state transition model is used for sampling, and the state sequence of t steps consecutive moments is used as the input of the network (s1,..., st).
How does Recurrent ppo set this value? Can you tell me the code location?
Thanks

[araffin]

the environmental state transition model is used for sampling, and the state sequence of t steps consecutive moments is used as the input of the network (s1,..., st).

you mean where do we create sequences of observations/actions to update the network?
this is done in the buffer: https://github.com/Stable-Baselines-Team/stable-baselines3-contrib/blob/master/sb3_contrib/common/recurrent/buffers.py#L25 (the code is complex to have a fixed mini-batch size without constrain on the number of envs/steps)

there is a simpler version available in #118 but with variable batch size.