reinforcement

BaseReinforcementSpace.

class common.optim.ne.space.reinforcement.BaseReinforcementSpace(config, env)

Bases: BaseSpace, ABC

Reinforcement Base Space class.

Parameters:

• env (EnvBase) – The torchrl environment to run the evaluation on.

• config (BaseSpaceConfig)

final run_pre_eval(agent, curr_gen)

Resets/loads the environment before evaluation begins.

Parameters:

• agent (BaseAgent) – The agent being evaluated.

• curr_gen (int) – See curr_gen.

Return type:

TensorDict

Returns:

See run_post_eval.out.

env_done_reset(agent, out, curr_gen)

Resets the agent/environment when the environment terminates.

Parameters:

• agent (BaseAgent) – See pre_eval_reset.agent.

• out (TensorDict) – The latest environment output.

• curr_gen (int) – See curr_gen.

Return type:

TensorDict | dict[str, bool]

Returns:

See run_post_eval.out.

final run_post_eval(agent, out, curr_gen)

Resets the agent & saves the environment post-evaluation.

Parameters:

• agent (BaseAgent) – See pre_eval_reset.agent.

• out (TensorDict) – The latest environment output.

• curr_gen (int) – See curr_gen.

Return type:

None

final evaluate(agents, curr_gen)

Evaluation function called once per generation.

Parameters:

• agents (list[list[BaseAgent]]) – A 2D list containing the agent to evaluate.

• curr_gen (int) – See curr_gen.

Return type:

ndarray[float32, Any]