StateCritic

class maze.core.agent.state_critic.StateCritic

Structured state critic class designed to work with structured environments. (see StructuredEnv).

It encapsulates state critic and queries them for values according to the provided policy ID.

abstract predict_value(observation: Dict[str, numpy.ndarray], critic_id: int | str) → StateCriticStepOutput

Query a critic that corresponds to the given ID for the state value.

Parameters:

• observation – Current observation of the environment

• critic_id – The critic id to query

Returns:

The value for this observation

abstract predict_values(critic_input: StateCriticInput) → StateCriticOutput

Query a critic that corresponds to the given ID for the state value.

Parameters:

critic_input – The critic input for predicting the values of all sub-steps of the env.

Returns:

Critic output holding the values, detached values and actor_id