callback

class xuance.common.callback.BaseCallback(*args, **kwargs)

基类：ABC

Base class for callback hooks in reinforcement learning training and testing.

Users can inherit this class to implement custom logic during different stages of training and evaluation.

on_test_end(*args, **kwargs)

Called at the end of the testing loop.

参数:

• *args – Optional positional arguments.

• **kwargs – Optional keyword arguments.

on_test_step(*args, **kwargs)

Called during each step in the testing loop.

参数:

• *args – Optional positional arguments.

• **kwargs – Optional keyword arguments.

on_train_episode_info(**kwargs)

Called at the termination or truncation of one episode for an environment.

on_train_epochs_end(current_step, **kwargs)

Called after each training epoch (i.e., after collecting one transition). :param current_step: The current global training step. :type current_step: int :param **kwargs: Additional optional information.

on_train_step(current_step, **kwargs)

Called after each training step (i.e., after collecting one transition).

参数:

• current_step (int) – The current global training step.

• **kwargs – Additional optional information.

on_train_step_end(current_step, **kwargs)

Called after a training step is completed (includes update, logging, etc.).

参数:

• current_step (int) – The current global training step.

• envs_info – Environment information.

• train_info – Training information.

on_update_end(iterations, **kwargs)

Called after the policy update is completed.

参数:

• iterations (int) – Number of update iterations that have performed.

• **kwargs – Optional keyword arguments.

on_update_start(iterations, **kwargs)

Called before the policy update begins.

参数:

• iterations (int) – Number of update iterations that have performed.

• **kwargs – Additional optional keyword arguments.

class xuance.common.callback.MultiAgentBaseCallback(*args, **kwargs)

基类：BaseCallback

on_update_agent_wise(iterations, agent_key, **kwargs) → dict

Called when updating an agent’s policy.

参数:

• iterations (int) – Number of update iterations that have performed.

• agent_key (str) – The key of the agent to update.

• **kwargs – Optional keyword arguments.