[Question] Policies with layer normalization

[astrofinix]

Question

Will there be there be future plans to implement layer normalized policies (MlpLnLstmPolicy, CnnLnLstmPolicy)?

Checklist

• [X] I have read the documentation (required)
• [X] I have checked that there is no similar issue in the repo (required)

[araffin]

Hello, Layer normalization was originally made for using parameter noise, so they were removed in SB3 (cf. docs). However, recently, there have some works that do use LN and dropout, so I plan to add those as options in the building blocs: https://github.com/DLR-RM/stable-baselines3/pull/1036

In the meantime, you can always define custom policies (cf. doc again) and for LSTM policies, only RecurrentPPO from our contrib repo support them.