[computer-go] Monte Carlo and transposition tables

Chrilly c.donninger at wavenet.at
Thu Jul 27 11:22:04 PDT 2006


It depends on whether you want to write a heuristic Go program or an exact solver. For an exact solver, you have to deal with GHI. For a heuristic program, forget about it.


Martin

I think this is exactly the point. In a chess-programm it does not matter if sometimes the mate-distance is wrong by 1 move (Actually some users complain, but it does not matter in respect to Elo-performance). In a perfect endgame database a wrong mate-distance is a serious bug.
Erik is right, that under other rules GHI is a problem. But as a (professional) game programmer one does not care about this. The job is to write the best possible programm for the existing way a game is played. But as an academic researcher one has to ask the question: Under what conditions is GHI a problem and how can I solve it. Thats the difference between science and engineering.

Chrilly
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://computer-go.org/pipermail/computer-go/attachments/20060727/c81b6fbd/attachment.htm


More information about the computer-go mailing list