rofunc.learning.RofuncRL.trainers.base_trainer#

1.  Module Contents#

1.1.  Classes#

BaseTrainer

1.2.  API#

class rofunc.learning.RofuncRL.trainers.base_trainer.BaseTrainer(cfg: omegaconf.DictConfig, env: Union[gym.Env, gymnasium.Env], device: Optional[Union[str, torch.device]] = None, env_name: Optional[str] = None, inference: bool = False)[source]#

Initialization

inference_flag = None#

Experiment log directory

start_time = None#

Evaluation and inference configurations

eval_rew_mean = 0#

Environment

setup_wandb()[source]#
get_action(states)[source]#
train()[source]#

Main training loop.

  • Reset the environment

  • For each step:

    • Pre-interaction

    • Obtain action from agent

    • Interact with environment

    • Store transition

    • Reset the environment

    • Post-interaction

  • Close the environment

pre_interaction()[source]#
post_interaction()[source]#

Base post-interaction function - Write to tensorboard - Save checkpoints

write_tensorboard()[source]#
eval()[source]#
inference()[source]#