reagent.net_builder package

Subpackages

Submodules

reagent.net_builder.categorical_dqn_net_builder module

class reagent.net_builder.categorical_dqn_net_builder.CategoricalDQNNetBuilder

Bases: object

Base class for categorical DQN net builder.

abstract build_q_network(state_normalization_data: reagent.core.parameters.NormalizationData, output_dim: int, num_atoms: int, qmin: int, qmax: int) → reagent.models.base.ModelBase

build_serving_module(q_network: reagent.models.base.ModelBase, state_normalization_data: reagent.core.parameters.NormalizationData, action_names: List[str], state_feature_config: reagent.core.types.ModelFeatureConfig) → torch.nn.modules.module.Module: Returns a TorchScript predictor module

reagent.net_builder.continuous_actor_net_builder module

class reagent.net_builder.continuous_actor_net_builder.ContinuousActorNetBuilder

Bases: object

Base class for continuous actor net builder.

abstract build_actor(state_feature_config: reagent.core.types.ModelFeatureConfig, state_normalization_data: reagent.core.parameters.NormalizationData, action_normalization_data: reagent.core.parameters.NormalizationData) → reagent.models.base.ModelBase

build_ranking_serving_module(actor: reagent.models.base.ModelBase, state_normalization_data: reagent.core.parameters.NormalizationData, candidate_normalization_data: reagent.core.parameters.NormalizationData, num_candidates: int, action_normalization_data: reagent.core.parameters.NormalizationData) → torch.nn.modules.module.Module

build_serving_module(actor: reagent.models.base.ModelBase, state_feature_config: reagent.core.types.ModelFeatureConfig, state_normalization_data: reagent.core.parameters.NormalizationData, action_normalization_data: reagent.core.parameters.NormalizationData, serve_mean_policy: bool = False) → torch.nn.modules.module.Module: Returns a TorchScript predictor module

abstract property default_action_preprocessing: str

reagent.net_builder.discrete_actor_net_builder module

class reagent.net_builder.discrete_actor_net_builder.DiscreteActorNetBuilder

Bases: object

Base class for discrete actor net builder.

abstract build_actor(state_normalization_data: reagent.core.parameters.NormalizationData, num_actions: int) → reagent.models.base.ModelBase

build_serving_module(actor: reagent.models.base.ModelBase, state_feature_config: reagent.core.types.ModelFeatureConfig, state_normalization_data: reagent.core.parameters.NormalizationData, action_feature_ids: List[int]) → torch.nn.modules.module.Module: Returns a TorchScript predictor module

reagent.net_builder.discrete_dqn_net_builder module

class reagent.net_builder.discrete_dqn_net_builder.DiscreteDQNNetBuilder

Bases: object

Base class for discrete DQN net builder.

build_binary_difference_scorer(q_network: reagent.models.base.ModelBase, state_normalization_data: reagent.core.parameters.NormalizationData, action_names: List[str], state_feature_config: reagent.core.types.ModelFeatureConfig) → torch.nn.modules.module.Module: Returns softmax(1) - softmax(0)

abstract build_q_network(state_feature_config: reagent.core.types.ModelFeatureConfig, state_normalization_data: reagent.core.parameters.NormalizationData, output_dim: int) → reagent.models.base.ModelBase

build_serving_module(q_network: reagent.models.base.ModelBase, state_normalization_data: reagent.core.parameters.NormalizationData, action_names: List[str], state_feature_config: reagent.core.types.ModelFeatureConfig, predictor_wrapper_type=None) → torch.nn.modules.module.Module: Returns a TorchScript predictor module

reagent.net_builder.parametric_dqn_net_builder module

class reagent.net_builder.parametric_dqn_net_builder.ParametricDQNNetBuilder

Bases: object

Base class for parametric DQN net builder.

abstract build_q_network(state_normalization_data: reagent.core.parameters.NormalizationData, action_normalization_data: reagent.core.parameters.NormalizationData, output_dim: int = 1) → reagent.models.base.ModelBase

build_serving_module(q_network: reagent.models.base.ModelBase, state_normalization_data: reagent.core.parameters.NormalizationData, action_normalization_data: reagent.core.parameters.NormalizationData) → torch.nn.modules.module.Module: Returns a TorchScript predictor module

reagent.net_builder.quantile_dqn_net_builder module

class reagent.net_builder.quantile_dqn_net_builder.QRDQNNetBuilder

Bases: object

Base class for QRDQN net builder.

abstract build_q_network(state_normalization_data: reagent.core.parameters.NormalizationData, output_dim: int, num_atoms: int) → reagent.models.base.ModelBase

build_serving_module(q_network: reagent.models.base.ModelBase, state_normalization_data: reagent.core.parameters.NormalizationData, action_names: List[str], state_feature_config: reagent.core.types.ModelFeatureConfig) → torch.nn.modules.module.Module: Returns a TorchScript predictor module

reagent.net_builder.slate_ranking_net_builder module

class reagent.net_builder.slate_ranking_net_builder.SlateRankingNetBuilder

Bases: object

Base class for slate ranking network builder.

abstract build_slate_ranking_network(state_dim, candidate_dim, candidate_size, slate_size) → torch.nn.modules.module.Module

reagent.net_builder.slate_reward_net_builder module

class reagent.net_builder.slate_reward_net_builder.SlateRewardNetBuilder

Bases: object

Base class for slate reward network builder.

abstract build_slate_reward_network(state_dim, candidate_dim, candidate_size, slate_size) → torch.nn.modules.module.Module

abstract property expect_slate_wise_reward: bool

reagent.net_builder.synthetic_reward_net_builder module

class reagent.net_builder.synthetic_reward_net_builder.SyntheticRewardNetBuilder

Bases: object

Base class for Synthetic Reward net builder.

build_serving_module(seq_len: int, synthetic_reward_network: reagent.models.base.ModelBase, state_normalization_data: reagent.core.parameters.NormalizationData, action_normalization_data: Optional[reagent.core.parameters.NormalizationData] = None, discrete_action_names: Optional[List[str]] = None) → torch.nn.modules.module.Module: Returns a TorchScript predictor module

abstract build_synthetic_reward_network(state_normalization_data: reagent.core.parameters.NormalizationData, action_normalization_data: Optional[reagent.core.parameters.NormalizationData] = None, discrete_action_names: Optional[List[str]] = None) → reagent.models.base.ModelBase

reagent.net_builder.unions module

class reagent.net_builder.unions.CategoricalDQNNetBuilder__Union(Categorical: Optional[reagent.net_builder.categorical_dqn.categorical.Categorical] = None)