Expand description
Behavioral Cloning (BC) configuration and hyperparameters.
BcConfig mirrors the builder + validate() style of
crate::train::a2c::A2cConfig and crate::train::sac::SacConfig.
Behavioral cloning trains a policy to reproduce expert actions from a
fixed Demonstrations dataset via
supervised cross-entropy. Defaults target classic discrete-control
tasks like CartPole: a 1e-3 learning rate over 64-sample minibatches
for 10 epochs.
This is the config half of PR A of the BC decomposition (#164); the BC trainer (#167) consumes it.
Structsยง
- BcConfig
- Behavioral Cloning configuration parameters.