[Projects] [Publications] [Contact]


Vegos is an attempt at an artificial intelligence that uses simulated annealing to play Go.

A short, informal paper on Vegos.  Also, Bernd Brügmann's original Monte Carlo Go paper (reformatted) that gave me the original idea.

Full source code of Vegos.  To run, also requires the Ideanest library and a modified GoGui library.  You'll need to compile everything with JDK 1.4 first.  No nice user-friendly GUI wrapper!  There are two entry points:

  • com.ideanest.vegos.compete.Runner to run tournaments; the parameters must be edited directly in the main method
  • gui.GoGui to run in interactive mode; the Vegos library must be on the classpath, and GoGui must be set to use the 'java com.ideanest.vegos.ClientProcess' external command; to activate analysis functions, you must place a modified analyze-commands file in the .gogui configuration directory

If you're only interested in the GTP client framework, look at the com.ideanest.vegos.gtp package, especially the GTPClient class.  The com.ideanest.vegos.RandomGTPEngine class gives a basic example of how to use the framework.  Note that the GTP client framework has some dependencies on, but you don't have to use its implementation of the board or rules.

Tournament Results:

Sample SemiPrimitive (9x9; komi: 0; handicap: 0) game records in SGF format:

two Stanley (1x500 steps; count: Null; mix: Far Swap, 0.99, 10):  Game A, Game B

two Stanley (1x1000 steps; count: Null; mix: Far Swap, 0.99, 10):  Game A, Game B