Modulesยง
- agents
- buffers
- distributions
- gym
- init
- Orthogonal weight initialization (PyTorch
nn.init.orthogonal_semantics), as used by reference PPO implementations: sqrt(2) gain for hidden layers, 0.01 for policy heads, 1.0 for value heads, zero biases. - models
- objectives
- parameter_
schedule - prelude
- Common imports for building ModuRL training programs.
- sampling
- spaces
- wrappers
- Environment-independent wrappers for
crate::gym::Gym.