Module: tf_agents.policies

Policies Module.

Modules

actor_policy module: Actor Policy based on an actor network.

async_policy_saver module: Async helper for the policy saver.

boltzmann_policy module: Policy implementation that applies temperature to a distribution.

categorical_q_policy module: Simple Categorical Q-Policy for Q-Learning with Categorical DQN.

epsilon_greedy_policy module: Policy implementation that generates epsilon-greedy actions from a policy.

fixed_policy module: A policy which always returns a fixed action.

gaussian_policy module: A policy that wraps a given policy and adds Gaussian noise.

greedy_policy module: Policy implementation that generates greedy actions from another policy.

ou_noise_policy module: A policy that wraps a given policy and adds Ornstein Uhlenbeck (OU) noise.

policy_saver module: TF-Agents SavedModel API.

py_policy module: Python Policies API.

py_tf_eager_policy module: Converts tf_policies when working in eager mode to py_policies.

py_tf_policy module: Converts TensorFlow Policies into Python Policies.

q_policy module: Simple Policy for DQN.

random_py_policy module: Policy implementation that generates random actions.

random_tf_policy module: Policy implementation that generates random actions.

scripted_py_policy module: Policy implementation that steps over a given configuration.

tf_policy module: TensorFlow Policies API.

tf_py_policy module: Exposes a python policy as an in-graph TensorFlow policy.

utils module: Utilities for policies.

Classes

class ActorPolicy: Class to build Actor Policies.

class EpsilonGreedyPolicy: Returns epsilon-greedy samples of a given policy.

class GreedyPolicy: Returns greedy samples of a given policy.

class PolicySaver: A PolicySaver allows you to save a tf_policy.Policy to SavedModel.

class PyTFEagerPolicy: Exposes a numpy API for TF policies in Eager mode.

class SavedModelPyTFEagerPolicy: Exposes a numpy API for saved_model policies in Eager mode.

class TFPolicy: Abstract base class for TF Policies.