Skip to content

Sb3EvalScriptSettings

Full path: schola.scripts.sb3.eval.settings.Sb3EvalScriptSettings

Top-level settings for evaluating a saved Stable-Baselines3 policy with Schola.

Sb3EvalScriptSettings(checkpoint = None, n_eval_episodes = 10,
                      deterministic = True, vecnormalize = None,
                      logging_settings = <factory>,
                      environment_settings = <factory>,
                      algorithm_settings = <factory>)

Parameters

  • checkpoint (Annotated)

  • n_eval_episodes (Annotated)

  • deterministic (bool)

  • vecnormalize (Annotated)

  • logging_settings (Annotated)

  • environment_settings (Annotated)

  • algorithm_settings (Annotated)

Methods

init

__init__(checkpoint = None, n_eval_episodes = 10, deterministic = True,
         vecnormalize = None, logging_settings = <factory>,
         environment_settings = <factory>, algorithm_settings = <factory>)

Parameters

  • checkpoint (Annotated)

  • n_eval_episodes (Annotated)

  • deterministic (bool)

  • vecnormalize (Annotated)

  • logging_settings (Annotated)

  • environment_settings (Annotated)

  • algorithm_settings (Annotated)

Attributes

algorithm_settings

algorithm_settings

The algorithm used to train the checkpoint that is being evaluated.

checkpoint

checkpoint

Path to a trained policy checkpoint (.zip) produced by SB3 model.save() (required via —checkpoint).

deterministic

deterministic

If True, the policy mean (or mode) is used; if False, stochastic actions are sampled.

environment_settings

environment_settings

How to launch or attach to the Unreal simulator and gRPC protocol for the environment.

logging_settings

logging_settings

Logging verbosity for Schola and SB3 components.

n_eval_episodes

n_eval_episodes

Number of evaluation episodes for stable_baselines3.common.evaluation.evaluate_policy.

vecnormalize

vecnormalize

Optional VecNormalize statistics file (.pkl) saved alongside the policy.