tf_agents.policies.SavedModelPyTFEagerPolicy

View source on GitHub

Exposes a numpy API for saved_model policies in Eager mode.

Inherits From: PyTFEagerPolicyBase, PyPolicy

tf_agents.policies.SavedModelPyTFEagerPolicy(
    model_path: Text,
    time_step_spec: Optional[tf_agents.trajectories.TimeStep] = None,
    action_spec: Optional[tf_agents.typing.types.DistributionSpecV2] = None,
    policy_state_spec: tf_agents.typing.types.NestedTensorSpec = (),
    info_spec: tf_agents.typing.types.NestedTensorSpec = (),
    load_specs_from_pbtxt: bool = False,
    use_tf_function: bool = False,
    batch_time_steps=True
)

Used in the notebooks

model_path Path to a saved_model generated by the policy_saver. time_step_spec Optional nested structure of ArraySpecs describing the policy's time_step_spec. This is not used by the SavedModelPyTFEagerPolicy, but may be accessed by other objects as it is part of the public policy API. action_spec Optional nested structure of ArraySpecs describing the policy's action_spec. This is not used by the SavedModelPyTFEagerPolicy, but may be accessed by other objects as it is part of the public policy API. policy_state_spec Optional nested structure of ArraySpecs describing the policy's policy_state_spec. This is not used by the SavedModelPyTFEagerPolicy, but may be accessed by other objects as it is part of the public policy API. info_spec Optional nested structure of ArraySpecs describing the policy's info_spec. This is not used by the SavedModelPyTFEagerPolicy, but may be accessed by other objects as it is part of the public policy API. load_specs_from_pbtxt If True the specs will be loaded from the proto file generated by the policy_saver. use_tf_function See PyTFEagerPolicyBase. batch_time_steps See PyTFEagerPolicyBase.

action_spec Describes the ArraySpecs of the np.Array returned by action().

action can be a single np.Array, or a nested dict, list or tuple of np.Array. collect_data_spec Describes the data collected when using this policy with an environment. info_spec Describes the Arrays emitted as info by action(). observation_and_action_constraint_splitter

policy_state_spec Describes the arrays expected by functions with policy_state as input. policy_step_spec Describes the output of action(). time_step_spec Describes the TimeStep np.Arrays expected by action(time_step). trajectory_spec Describes the data collected when using this policy with an environment.

Methods

action

View source

action(
    time_step: tf_agents.trajectories.TimeStep,
    policy_state: tf_agents.typing.types.NestedArray = (),
    seed: Optional[types.Seed] = None
) -> tf_agents.trajectories.PolicyStep

Generates next action given the time_step and policy_state.

time_step A TimeStep tuple corresponding to time_step_spec(). policy_state An optional previous policy_state. seed Seed to use if action uses sampling (optional).

get_initial_state

View source

get_initial_state(
    batch_size: Optional[int] = None
) -> tf_agents.typing.types.NestedArray

Returns an initial state usable by the policy.

batch_size An optional batch size.

get_metadata

View source

get_metadata()

Returns the metadata of the saved model.

get_train_step

View source

get_train_step() -> tf_agents.typing.types.Int

Returns the training global step of the saved model.

get_train_step_from_last_restored_checkpoint_path

View source

get_train_step_from_last_restored_checkpoint_path() -> Optional[int]

Returns the training step of the restored checkpoint.

update_from_checkpoint

View source

update_from_checkpoint(
    checkpoint_path: Text
)

Allows users to update saved_model variables directly from a checkpoint.

checkpoint_path is a path that was passed to either PolicySaver.save() or PolicySaver.save_checkpoint(). The policy looks for set of checkpoint files with the file prefix `/variables/variables'

checkpoint_path Path to the checkpoint to restore and use to udpate this policy.

variables

View source

variables()