policy

Control policies.

Note

Policies can be either time-invariant or time-varying, and can be either open- or closed-loop. Thus, the arguments to the __call__() method should allow for time and state to be specified (if needed), and should be optional kwargs:

>>> def __call__(self, time=None, state=None):
...     ...

class gym_socks.policies.policy.BasePolicy

Base policy class.

This class is abstract, meaning it is not meant to be instantiated directly. Instead, define a new class that inherits from BasePolicy.

The __call__() method is the main point of entry for the policy classes. All subclasses must implement a __call__() method. This makes the class callable, so that policies can be evaluated as:

action = policy(state)

abstract __call__(*args, **kwargs)

Evaluate the policy.

Returns

An action in the action space.

action_space = None

state_space = None

class gym_socks.policies.policy.ConstantPolicy(action_space=None, constant=0)

Constant policy.

A policy which returns a constant control action.

Parameters

• action_space (gym.Space) – The action space of the system.

• constant – The constant value returned by the policy.

__call__(*args, **kwargs)

Evaluate the policy.

Returns

An action in the action space.

action_space = None

state_space = None

class gym_socks.policies.policy.RandomizedPolicy(action_space=None)

Randomized policy.

A policy which returns a random control action.

Parameters

action_space (gym.Space) – The action space of the system.

__call__(*args, **kwargs)

Evaluate the policy.

Returns

An action in the action space.

action_space = None

state_space = None

class gym_socks.policies.policy.ZeroPolicy(action_space=None)

Zero policy.

A policy which returns a constant (zero) control action.

Parameters

action_space (gym.Space) – The action space of the system.

__call__(*args, **kwargs)

Evaluate the policy.

Returns

An action in the action space.

action_space = None

state_space = None