benchmarl.algorithms.IsacConfig

class IsacConfig(share_param_critic: bool, num_qvalue_nets: int, loss_function: str, delay_qvalue: bool, target_entropy: float | str, discrete_target_entropy_weight: float, alpha_init: float, min_alpha: float | None, max_alpha: float | None, fixed_alpha: bool, scale_mapping: str, use_tanh_normal: bool)[source]

Bases: AlgorithmConfig

Configuration dataclass for Isac.

share_param_critic: bool = <dataclasses._MISSING_TYPE object>

num_qvalue_nets: int = <dataclasses._MISSING_TYPE object>

loss_function: str = <dataclasses._MISSING_TYPE object>

delay_qvalue: bool = <dataclasses._MISSING_TYPE object>

target_entropy: float | str = <dataclasses._MISSING_TYPE object>

discrete_target_entropy_weight: float = <dataclasses._MISSING_TYPE object>

alpha_init: float = <dataclasses._MISSING_TYPE object>

min_alpha: float | None = <dataclasses._MISSING_TYPE object>

max_alpha: float | None = <dataclasses._MISSING_TYPE object>

fixed_alpha: bool = <dataclasses._MISSING_TYPE object>

scale_mapping: str = <dataclasses._MISSING_TYPE object>

use_tanh_normal: bool = <dataclasses._MISSING_TYPE object>

static associated_class() → Type[Algorithm][source]: The algorithm class associated to the config

static supports_continuous_actions() → bool[source]: If the algorithm supports continuous actions

static supports_discrete_actions() → bool[source]: If the algorithm supports discrete actions

static on_policy() → bool[source]: If the algorithm has to be run on policy or off policy

static has_independent_critic() → bool[source]: If the algorithm uses an independent critic