mpcrl.wrappers.agents.Wrapper#

class mpcrl.wrappers.agents.Wrapper(agent)[source]#

Bases: CallbackMixin, Generic[SymType]

Wraps an instance of mpcrl.Agent to allow a modular transformation of its behaviour. This class is the base class for all wrappers. The subclass could override some methods to change the behavior of the original agent without touching the original code.

Parameters:

agentAgent or subclass: The agent to wrap.

Methods

`detach_wrapper`([recursive])	Detaches the wrapper from the agent, returning the unwrapped agent.
`is_wrapped`(wrapper_type)	Gets whether the agent instance is wrapped or not by the wrapper type.

Attributes

unwrapped

Returns the original agent wrapped by this wrapper.

detach_wrapper(recursive=False)[source]#

Detaches the wrapper from the agent, returning the unwrapped agent. De facto, this method detaches all the hooks attached by this wrapper.

Parameters:

recursivebool, optional: If True, detaches all the wrappers around the agent recursively.

Returns:

Agent or LearningAgent or Wrapper: Returns the wrapped agent (or other wrapper) instance. This instance has no more active hooks attached by this wrapper. If recursive=True, all the wrappers around the agent and their hooks are detached.

Return type:

Union[Agent[TypeVar(SymType, SX, MX)], LearningAgent[TypeVar(SymType, SX, MX), TypeVar(ExpType)], Wrapper[TypeVar(SymType, SX, MX)]]

Notes

Detaching a wrapper is useful when you want to make sure that the wrapper’s hooked callback cannot modify the behaviour or data of the agent, for example, after learning is done and you want to save and evaluate your learnt policy.

is_wrapped(wrapper_type)[source]#

Gets whether the agent instance is wrapped or not by the wrapper type.

Parameters: