Denys88 / rl_games

RL implementations
MIT License
820 stars 138 forks source link

Is Regularizer option working ? #245

Closed arminvburren closed 1 year ago

arminvburren commented 1 year ago

Hello,

I've been working on OmniIsaacGym from NVIDIA which uses rl_games. In the config files there's an option called "regularizer" in the mlp section:

    mlp:
      units: [512, 256, 128]
      activation: elu
      d2rl: False

      initializer:
        name: default
      regularizer:
        name: None #'l2_regularizer' # None
        #scale: 10

Is this option actually doing something ? I tried activated it and I don't have any visible difference, including with strong scales. Where in the code is this used ?

In general is the library still worked on ? There's no commit since 3 month apparently

Denys88 commented 1 year ago

yes, it doesnt work here. It was old configuration from the TF version because you could add regularizer independently. There is weight_decay paramter.