QLearningAgent Class Reference

#include <QLearningAgent.h>

Inheritance diagram for QLearningAgent:

Collaboration diagram for QLearningAgent:

Public Member Functions
	QLearningAgent (QFunction *qFunction, Policy *policy, unsigned int observationDim, ActionProperties *actionProperties, float lambda, float gamma, bool offPolicy=false)
virtual	~QLearningAgent ()
virtual void	init ()
	Initializes the agent. More...
virtual Action *	start (const Observation *observation)
	Chooses the first action in the episode. More...
virtual Action *	step (const Observation *observation)
	Performs an episode step. More...
virtual void	end (const Observation *observation)
	Performs the last episode step. More...
virtual void	save (XFile *file)
	Saves the agent. More...
virtual void	load (XFile *file)
	Loads an agent. More...
Public Member Functions inherited from Agent
virtual	~Agent ()

Public Attributes
bool	isLearning
	Toggle learning. More...
Policy *	policy
	The policy used by the agent. More...
QFunction *	qFunction
	The state-action value approximator function. More...
TDTrainer	trainer
	The trainer that we use to train the function. More...
Action	lastAction
Action	currentAction
RLObservation	lastObservation
unsigned int	observationDim

Detailed Description

An agent that behaves according to an action-value, TD-lambda reinforcement learning algorithm. The model allows for both on-policy (SARSA) and off-policy (Q-learning) learning.

Constructor & Destructor Documentation

QLearningAgent::QLearningAgent	(	QFunction *	qFunction,
		Policy *	policy,
		unsigned int	observationDim,
		ActionProperties *	actionProperties,
		float	lambda,
		float	gamma,
		bool	offPolicy = false
	)

QLearningAgent::~QLearningAgent ( )

virtual

Member Function Documentation

void QLearningAgent::end ( const Observation *  observation )

virtual

Performs the last episode step.

Reimplemented from Agent.

void QLearningAgent::init ( )

virtual

Initializes the agent.

Reimplemented from Agent.

void QLearningAgent::load ( XFile *  file )

virtual

Loads an agent.

Reimplemented from Agent.

void QLearningAgent::save ( XFile *  file )

virtual

Saves the agent.

Reimplemented from Agent.

Action * QLearningAgent::start ( const Observation *  observation )

virtual

Chooses the first action in the episode.

Implements Agent.

Action * QLearningAgent::step ( const Observation *  observation )

virtual

Performs an episode step.

Implements Agent.

Member Data Documentation

Action QLearningAgent::currentAction

bool QLearningAgent::isLearning

Toggle learning.

Action QLearningAgent::lastAction

RLObservation QLearningAgent::lastObservation

unsigned int QLearningAgent::observationDim

Policy* QLearningAgent::policy

The policy used by the agent.

QFunction* QLearningAgent::qFunction

The state-action value approximator function.

TDTrainer QLearningAgent::trainer

The trainer that we use to train the function.

---

The documentation for this class was generated from the following files:

• src/qualia/rl/QLearningAgent.h
• src/qualia/rl/QLearningAgent.cpp