PIQLE : a Platform Implementing Q-LEarning algorithms in JAVA

This page is no longer maintained (except the news) and stays alive for History. Up-to-date versions of Piqle and its documentation can be found on sourceforge

December 2006 : Some new classes to interface Piqle with RLAI's RL-GLue : download it at sourceforge !
November 2006 : Piqle version 2 released : major refactoring, translation of names and comments in english, synthetic documentation. Everything is now on sourceforge
September 2006 : A slight alteration of PIQLE can communicate with the Octopus Arm Simulator by mean of sockets : Read the entire story and try it !
August 2006 : Piqle is on Sourceforge, thanks to Martin Mauch !
June 2006 : PIQLE can handle multi-agents learning :
- The multi-agent version of PIQLE, still compatible with previous problems, together with an example multi-agent problem implementation. Old problems are not in this distribution, but you can copy them from a previous version of PIQLE without having to change anything in their code.
- A description of the new classes and architecture.
January 2006 : Tile Coding (CMAC) is implemented in PIQLE. See MountainCarTiling.java and AcrobotCLS2Tiling.java. here is a little documentation on how to use it.
December 9th 2005 : PIQLE will be presented at the NIPS 2005 workshop Reinforcement Learning Benchmarks and Bake-offs II

November 24th 2005 : PIQLE is now licensed under the GNU Lesser General Public License.
October 20th 2005 : Adding the Acrobot environment... but learning seems difficult (for the moment, there is no tiling in PIQLE !)
October 4th 2005 :
- Graphical possibilities for one player games (see Course.java for car races, and VisualMaze.java,CentralCrossMaze.java,ChristianCrossMaze.java for visual mazes)
- New mazes and new perceptions of environment
- Package voitures : simulation of mobile robot in closed arena.

Finding your way among obstacles.

Alice's adventures in Mazeland.

Acrobot.

PIQLE is a set of Java classes for quickly experimenting reinforcement learning schemes : new problems or new algorithms. It comes in two flavors :

Uncompact the tgz file : tar -xvzf piqle.tgz or tar -xvzf piqleWeka.tgz In both cases, you will obtain two directories :

The java sources directory, with all classes and some examples.
The java doc directory. You will perhaps have to recompile it with the good links : have a look at the file named makedoc, where you can make some setups.
For piqleWeka, the file named COMPILATION is a good remainder of how to invoke weka while compiling.