reagent.model_managers.actor_critic package

Submodules

reagent.model_managers.actor_critic.sac module

class reagent.model_managers.actor_critic.sac.SAC(state_preprocessing_options: Optional[reagent.workflow.types.PreprocessingOptions] = None, action_preprocessing_options: Optional[reagent.workflow.types.PreprocessingOptions] = None, action_feature_override: Optional[str] = None, state_feature_config_provider: reagent.workflow.types.ModelFeatureConfigProvider__Union = <factory>, action_float_features: List[Tuple[int, str]] = <factory>, reader_options: Optional[reagent.workflow.types.ReaderOptions] = None, eval_parameters: reagent.core.parameters.EvaluationParameters = <factory>, save_critic_bool: bool = True, trainer_param: reagent.training.parameters.SACTrainerParameters = <factory>, actor_net_builder: reagent.net_builder.unions.ContinuousActorNetBuilder__Union = <factory>, critic_net_builder: reagent.net_builder.unions.ParametricDQNNetBuilder__Union = <factory>, value_net_builder: Optional[reagent.net_builder.unions.ValueNetBuilder__Union] = <factory>, use_2_q_functions: bool = True, serve_mean_policy: bool = True)

Bases: reagent.model_managers.actor_critic_base.ActorCriticBase

actor_net_builder: reagent.net_builder.unions.ContinuousActorNetBuilder__Union

build_serving_module(trainer_module: reagent.training.reagent_lightning_module.ReAgentLightningModule, normalization_data_map: Dict[str, reagent.core.parameters.NormalizationData]) → torch.nn.modules.module.Module

Optionaly, implement this method if you only have one model for serving

build_trainer(normalization_data_map: Dict[str, reagent.core.parameters.NormalizationData], use_gpu: bool, reward_options: Optional[reagent.workflow.types.RewardOptions] = None) → reagent.training.sac_trainer.SACTrainer

Implement this to build the trainer, given the config

TODO: This function should return ReAgentLightningModule & the dictionary of modules created

critic_net_builder: reagent.net_builder.unions.ParametricDQNNetBuilder__Union

get_reporter()

serve_mean_policy: bool = True

trainer_param: reagent.training.parameters.SACTrainerParameters

use_2_q_functions: bool = True

value_net_builder: Optional[reagent.net_builder.unions.ValueNetBuilder__Union]

reagent.model_managers.actor_critic.td3 module

class reagent.model_managers.actor_critic.td3.TD3(state_preprocessing_options: Optional[reagent.workflow.types.PreprocessingOptions] = None, action_preprocessing_options: Optional[reagent.workflow.types.PreprocessingOptions] = None, action_feature_override: Optional[str] = None, state_feature_config_provider: reagent.workflow.types.ModelFeatureConfigProvider__Union = <factory>, action_float_features: List[Tuple[int, str]] = <factory>, reader_options: Optional[reagent.workflow.types.ReaderOptions] = None, eval_parameters: reagent.core.parameters.EvaluationParameters = <factory>, save_critic_bool: bool = True, trainer_param: reagent.training.parameters.TD3TrainerParameters = <factory>, actor_net_builder: reagent.net_builder.unions.ContinuousActorNetBuilder__Union = <factory>, critic_net_builder: reagent.net_builder.unions.ParametricDQNNetBuilder__Union = <factory>, use_2_q_functions: bool = True)

Bases: reagent.model_managers.actor_critic_base.ActorCriticBase

actor_net_builder: reagent.net_builder.unions.ContinuousActorNetBuilder__Union

build_serving_module(trainer_module: reagent.training.reagent_lightning_module.ReAgentLightningModule, normalization_data_map: Dict[str, reagent.core.parameters.NormalizationData]) → torch.nn.modules.module.Module

Optionaly, implement this method if you only have one model for serving

build_trainer(normalization_data_map: Dict[str, reagent.core.parameters.NormalizationData], use_gpu: bool, reward_options: Optional[reagent.workflow.types.RewardOptions] = None) → reagent.training.td3_trainer.TD3Trainer

Implement this to build the trainer, given the config

TODO: This function should return ReAgentLightningModule & the dictionary of modules created

critic_net_builder: reagent.net_builder.unions.ParametricDQNNetBuilder__Union

eval_parameters: reagent.core.parameters.EvaluationParameters

get_reporter()

trainer_param: reagent.training.parameters.TD3TrainerParameters

use_2_q_functions: bool = True

Module contents

class reagent.model_managers.actor_critic.SAC(state_preprocessing_options: Optional[reagent.workflow.types.PreprocessingOptions] = None, action_preprocessing_options: Optional[reagent.workflow.types.PreprocessingOptions] = None, action_feature_override: Optional[str] = None, state_feature_config_provider: reagent.workflow.types.ModelFeatureConfigProvider__Union = <factory>, action_float_features: List[Tuple[int, str]] = <factory>, reader_options: Optional[reagent.workflow.types.ReaderOptions] = None, eval_parameters: reagent.core.parameters.EvaluationParameters = <factory>, save_critic_bool: bool = True, trainer_param: reagent.training.parameters.SACTrainerParameters = <factory>, actor_net_builder: reagent.net_builder.unions.ContinuousActorNetBuilder__Union = <factory>, critic_net_builder: reagent.net_builder.unions.ParametricDQNNetBuilder__Union = <factory>, value_net_builder: Optional[reagent.net_builder.unions.ValueNetBuilder__Union] = <factory>, use_2_q_functions: bool = True, serve_mean_policy: bool = True)

Bases: reagent.model_managers.actor_critic_base.ActorCriticBase

action_float_features: List[Tuple[int, str]]

actor_net_builder: reagent.net_builder.unions.ContinuousActorNetBuilder__Union

build_serving_module(trainer_module: reagent.training.reagent_lightning_module.ReAgentLightningModule, normalization_data_map: Dict[str, reagent.core.parameters.NormalizationData]) → torch.nn.modules.module.Module

Optionaly, implement this method if you only have one model for serving

build_trainer(normalization_data_map: Dict[str, reagent.core.parameters.NormalizationData], use_gpu: bool, reward_options: Optional[reagent.workflow.types.RewardOptions] = None) → reagent.training.sac_trainer.SACTrainer

Implement this to build the trainer, given the config

TODO: This function should return ReAgentLightningModule & the dictionary of modules created

critic_net_builder: reagent.net_builder.unions.ParametricDQNNetBuilder__Union

eval_parameters: EvaluationParameters

get_reporter()

serve_mean_policy: bool = True

state_feature_config_provider: ModelFeatureConfigProvider__Union

trainer_param: reagent.training.parameters.SACTrainerParameters

use_2_q_functions: bool = True

value_net_builder: Optional[reagent.net_builder.unions.ValueNetBuilder__Union]

class reagent.model_managers.actor_critic.TD3(state_preprocessing_options: Optional[reagent.workflow.types.PreprocessingOptions] = None, action_preprocessing_options: Optional[reagent.workflow.types.PreprocessingOptions] = None, action_feature_override: Optional[str] = None, state_feature_config_provider: reagent.workflow.types.ModelFeatureConfigProvider__Union = <factory>, action_float_features: List[Tuple[int, str]] = <factory>, reader_options: Optional[reagent.workflow.types.ReaderOptions] = None, eval_parameters: reagent.core.parameters.EvaluationParameters = <factory>, save_critic_bool: bool = True, trainer_param: reagent.training.parameters.TD3TrainerParameters = <factory>, actor_net_builder: reagent.net_builder.unions.ContinuousActorNetBuilder__Union = <factory>, critic_net_builder: reagent.net_builder.unions.ParametricDQNNetBuilder__Union = <factory>, use_2_q_functions: bool = True)

Bases: reagent.model_managers.actor_critic_base.ActorCriticBase

action_float_features: List[Tuple[int, str]]

actor_net_builder: reagent.net_builder.unions.ContinuousActorNetBuilder__Union

build_serving_module(trainer_module: reagent.training.reagent_lightning_module.ReAgentLightningModule, normalization_data_map: Dict[str, reagent.core.parameters.NormalizationData]) → torch.nn.modules.module.Module

Optionaly, implement this method if you only have one model for serving

build_trainer(normalization_data_map: Dict[str, reagent.core.parameters.NormalizationData], use_gpu: bool, reward_options: Optional[reagent.workflow.types.RewardOptions] = None) → reagent.training.td3_trainer.TD3Trainer

Implement this to build the trainer, given the config

TODO: This function should return ReAgentLightningModule & the dictionary of modules created

critic_net_builder: reagent.net_builder.unions.ParametricDQNNetBuilder__Union

eval_parameters: reagent.core.parameters.EvaluationParameters

get_reporter()

state_feature_config_provider: ModelFeatureConfigProvider__Union

trainer_param: reagent.training.parameters.TD3TrainerParameters

use_2_q_functions: bool = True