Index of /~willmlam/teaching/cs175_w12/Mancala

	Name	Last modified	Size	Description
	Parent Directory		-
	AILearning.jar	2012-02-29 17:48	7.7K
	Group9RL.jar	2012-02-29 17:48	6.8K
	KalahGameState.java	2012-02-29 17:52	3.8K
	KalahPieGameState.java	2012-02-29 17:52	4.8K
	MancalaFramework.zip	2012-02-29 17:52	11K
	MancalaGameState.java	2012-02-29 17:52	2.1K
	MancalaGameState.txt	2012-01-18 18:24	1.5K
	MancalaPlayer.java	2012-02-29 17:52	665
	OwareGameState.java	2012-02-29 17:52	5.0K
	README.html	2012-02-29 17:52	3.8K
	README.md	2012-01-18 18:24	1.5K
	Team8.jar	2012-02-29 17:48	5.2K
	TermGame.java	2012-02-29 17:52	4.5K
	group1.jar	2012-02-29 17:48	5.3K
	group10_data.txt	2012-02-27 12:04	161
	interactive_Player.java	2012-02-29 17:52	1.0K
	random_Player.java	2012-02-29 17:52	1.7K

Changelog

• 02/29/2012: Tiny change to TermGame so that a gs.copy() is passed into the getMove() method of a player for safety reasons.
• 02/24/2012: Added an actionsBeforeDeletion() method for use when a player class is no longer going to be used. More details on its usage and reason in TermGame.java
• 02/08/2012: Fixed a bug where the second player could keep making pie moves as long as it never makes a different move. (Thanks to a student for pointing this out!)
• 01/26/2012: Changed to feature added yesterday to have the method take in the game state instead. This should allow for more flexibility by providing the entire final state as information.
• 01/25/2012: Added a method to MancalaPlayer which allows it to do something after the game ends. This can be used to provide feedback to the player for learning purposes.
• 01/23/2012: Fixed a bug in KalahGameState where a grab was not occuring when it lands on the first bucket after the opponent's.
• 01/21/2012: Modified the interactive_Player class so it works in Eclipse.
• 01/18/2012: Added a version of Kalah using the pie rule (KalahPieGameState). Removed "winner" fields from GameState classes.
• 01/14/2012: Initial release.

About the Mancala Game Framework

This framework was adapted for Mancala by William Lam for the Winter 2012 offering of CS 175, taught by Max Welling. The code was originally designed for Connect Four, provided by Alex Ihler.

Classes/Interfaces

MancalaGameState (abstract)
|- KalahGameState
|- OwareGameState
|- KalahPieGameState

Use one of the subclasses to decide on the Mancala variant.

MancalaPlayer (interface)
|- random_Player
|- interactive_Player

Implement the MancalaPlayer interface to create your AI routine. The name of your class should be formatted as YourClassName_Player.

TermGame

TermGame provides an example of how to use MancalaGameState and MancalaPlayer. It uses reflection to instantiate the state and player classes based on command line parameters. We will use a similar main class to this to run the tournament.

The command-line parameter format is: 
java TermGame <GameType> <StartingStones> <Player0Class> <Player1Class>

Ex: java TermGame Kalah 4 interactive random
Set player 0 as the interactive player and player 1 as the random player, 
playing the Kalah variant with 4 starting stones in each bucket.

Note that for Oware, repeated states are possible in the game tree. Since we cannot assume that two players will make a consensus to end the game, we will leave this decision up to the user interface running the game. (It should detect a sequence of repeated states and end the main execution loop, if so.)

About the competition

For the competition, we will run a program using the KalahPieGameState class that uses your AI player classes (and any of their dependencies) only. It will stop the execution of your player class without warning (and thus count as a forfeit) if it does not make a move within the time limit. You will be responsible of keeping track of the time within your implementation.

We will use a machine running Java 6 and your agent should use no more than 1GB memory and a single thread.

Time limit: 5 seconds per move

Tournament format: Double round-robin (allowing players to take turns going first)

We will then run the tournament as many times as we have time/resources for. The hope is that learning agents will play differently as we run more tournaments. The rankings will then be determined by the number of wins across all tournaments.