I notice that Model and Network both inherit nn.Module, so if I want to custom my network, which one should I use to overwrite, and what are their differences? By the way, is it possible to pre-train the actor network use a pid controller, then train it like normal ways in rl_games?
Hi, @HGGshiwo
network represents exact NN: MLP or CNN or RNN or etc.
Model has distribution around it. For example it could be continuous or discrete. Plus normalization is applied inside of the model.
I notice that Model and Network both inherit
nn.Module
, so if I want to custom my network, which one should I use to overwrite, and what are their differences? By the way, is it possible to pre-train the actor network use a pid controller, then train it like normal ways in rl_games?Looking forward your quick reply, thanks.