Two players game examples ended with a bug in case the second player is also a learning agent.
Might be ok now.
The good version of piqle-core is hence version2.1.1
Reinforcement learning framework Piqle Version 2_1 is released. Adding support for 'cooperative agents' (i.e. multi-agents awared of their neighbours actions). See an example of use in maabacV2 package, and all information into the pdf doc.
A version of PIQLE to interface with RLAI's RL_GLUE framework is now available.
Note that, as RL-GLUE only runs under Unix, this version of Piqle is hence only meaningful under Unix too.
The new version of Piqle is a major refactoring, together with a complete translation of names and comments in English. Examples have been added. Piqle has now a complete unified and synthetic pdf documentation.
An easy example of implmenting a two players game within Piqle
An easy puzzle and a good example to see how to implment problems within Piqle.
An implementation of the Memory game learning example is now available for Piqle.The game is customizable, and the agent actually learns to win.
An example on how to implement a two players game in Piqle, and how to train it, against a random player, or another learning player.
Implementation of Sutton and Barto example
Added an example of a mobile robot with limited perceptions (quite similar to Khepera or Alice)
An example illustrating the implementation of Sutton and Barto Gambler's Problem, with value iteration algorithm and optimal policy.
Some little examples of the mountain car task are now available for Piqle:
Download the file in your piqle source directory, unompress it, compile examples.
ICMLOctopus contains all the files needed to connect Piqle with the Octopus Arm Simulator developed at McGill, which was used in the competition that hold at ICML2006 workshop