[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gnugo-devel] MonteGNU

From: alain Baeckeroot
Subject: Re: [gnugo-devel] MonteGNU
Date: Wed, 26 Apr 2006 01:39:18 +0200
User-agent: KMail/1.9.1


For fun and curiosity i have put a simplified and slightly modified MonteGNU
 on CGOS: MC379_3500.
- No more alpha/beta currently.
- it does 3500 simulations for each of top_moves better than 1/3 of best move.
- it uses the mean probability of victory instead of mean score. Maybe this
 need tuning before it shows improvement.
- if the best move has a SEMEAI move reason, keep it and desactivate MC. This
 seems to weaken it (at least on 9x9.tst, run in 1/4 hour)

It is fast enough to stay at level 10 and play in less than 10 min.

I tried to make it run on all boardsize, but i have trouble with
 this_score = play_random_game(forbidden_move)
i'm not sure how to compute the score! so even if it runs it does not work yet.

MC seems also very interesting for analysing the regressions tests, as it is
easy to change the aim (score or winning_probability).


reply via email to

[Prev in Thread] Current Thread [Next in Thread]