Reinforcement Learning Framework in Java

See:
Description

Packages

net.pakl.rl	These are the basic reinforcement learning model classes -- every reinforcement learning problem can be described (minimally) as a World (collection of states), Policy, ValueFunction, and Actions.
net.pakl.rl.maze	World and policy for a 2D problem with impassable obstacles.
org.eyelanguage.rl.analysis	Tools to extract data from the adaptive reading agent's logfile for later processing.
org.eyelanguage.rl.reading	Code for the Adaptive Reading Agent; see ReadingMain for parameters and default values.
org.eyelanguage.rl.reading.ui	New simplified graphical user interface for adaptive reading agent.
tools

Reinforcement Learning Framework in Java

This is API documentation for the Reinforcement Learning simulation package developed in the Eye-Language Lab at the University of Pittsburgh. Most of the code here was written initially by Patryk Laurent. To see a list of the basic classes, click on the "net.pakl.rl" package in the upper right.

Note: The adaptive reading agent is implemented in org.eyelanguage.*

General Structure

The basic classes to support reinforcement learning are found in net.pakl.rl.* and subclasses for specific RL problems are found in subpackages. For example, the Adaptive Reading Agent (Reichle & Laurent, 2006) is found in org.eyelanguage.rl.reading.

Examples

Some of the present code demonstrates the reinforcement learning framework, in particular the maze subpackage. To a limited extent there are also unit tests to verify that various components of the learning algorithm and procided environments are functioning correctly.

Copyright (C) 2006 The Eye-Language Lab at the University of Pittsburgh. You are allowed to use this code in your projects, but you must acknowledge its source. You are allowed to extend any of these classes and distribute this source code along with your own new classes.