[computer-go] GPW08 paper: Parallel MCTS

2008-11-11 Thread Hideki Kato
Hi,

This is my GPW08 paper.

Parallel Monte-Carlo Tree Search with Simulation Servers
by Hideki Kato and Ikuo Takeuchi

Abstract: Recently Monte-Carlo tree search is boosting the performance 
of computer Go playing programs. A novel parallel Monte-Carlo tree 
search algorithm is proposed. A tree searcher runs on a client 
computer and multiple Monte-Carlo simulation servers run on other 
computers on a network. The client broadcasts a position to be 
simulated to every server, which then simulates a game from the 
position to the end and sends the result (win or loss) back to the 
client. The statistical information in the search tree is updated by 
the client according to the result. This architecture allows servers 
on-the-fly connection or disconnection. Experimental results using 
four quad-core Linux computers on a private Gigabit Ethernet LAN show 
its performance scales well.

http://www.gggo.jp/publications/gpw08-private.pdf
http://www.gggo.jp/publications/gpw08.ppt

If you read the paper, please check gpw08.ppt together as it contains 
revised charts which will help avoid confusion.

Hideki
--
[EMAIL PROTECTED] (Kato)
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/


[computer-go] Result of 2nd GPW Cup Computer Go Tournament in Hakone, Japan

2008-11-11 Thread Hideki Kato
Japanese version of the result of 2nd GPW Cup Computer Go Tournament 
is available at 
http://sig-gi.c.u-tokyo.ac.jp/gpw/2008/night.html#go-results

English version follows.

Ten programs from three countries were participated and we changed 
planed alternating BW round-robin to simple one.

HappyGo was resigned after 5th round due to critical bugs.
MC_arc had troubles in time control.

Date: November 7 and 8, 2008.
Place: Surugadai-gakuen Hakone seminor house Kounso, Sengokuhara 
845, Hakone, Kanagawa, Japan.
Format: round-robin
Rules: Chinese
Komi: 6.5
Time: 15 minutes

[Result]
Pos Program  Operator  Country Hardware Win Type
  1 MoGo Arpad Rimmel  France  4x3 GHz9 MCTS
  2 Aya  Hiroshi Yamashita Japan   8x2.6 GHz  8 MC/UCT
  3 Fudo Go  Hideki Kato   Japan   4x3.4 GHz  7 MC/UCT
  4 JimmyShun-Chin Hsu Taiwan  laptop 6 MC/UCT
  5 DMC  Matsui & Noguchi  Japan   laptop 5 MC/UCT
  6 KK   Satohiro Kadowaki Japan   laptop 4 MC/UCT
  7 KatsunariShin'ich Sei  Japan   laptop 3 Classic
  8 KIDS For IGo Nozomu IkehataJapan   laptop 2 Pattern
  9 MC_arc   Nobuo Araki   Japan   4x2.4 GHz  1 MC/UCT
 10 HappyGo  Yong-Ler Wang Taiwan  laptop 0 MCTS?

[Cross table]
No Program  1  2  3  4  5  6  7  8  9 10 Point
 1 MoGo -  1  1  1  1  1  1  1  1  1 9
 2 Aya  0  -  1  1  1  1  1  1  1  1 8
 3 Fudo Go  0  0  -  1  1  1  1  1  1  1 7
 4 Jimmy0  0  0  -  1  1  1  1  1  1 6
 5 DMC  0  0  0  0  -  1  1  1  1  1 5
 6 KK   0  0  0  0  0  -  1  1  1  1 4
 7 Katsunari0  0  0  0  0  0  -  1  1  1 3
 8 KIDS For IGo 0  0  0  0  0  0  0  -  1  1 2
 9 MC_arc   0  0  0  0  0  0  0  0  -  1 1
10 HappyGo  0  0  0  0  0  *  *  *  *  - 0

Director of the tournament,
Hideki Kato

Hideki Kato: <[EMAIL PROTECTED]>:
>The 2nd GPW Cup Gomputer Go Tournament will be held as a night event 
>of Game Programming Workshop 2008.
>GPW2008:  http://sig-gi.c.u-tokyo.ac.jp/gpw/2008/ (in Japanese)
>#Those who can't read Japanese and have interests in paticipating GPW 
>Cup Computer Go Tournament, please mail me.
>
>The 13th Game Programming Workshop will be held November 7 - 9 at 
>Hakone Seminor House, Hakone, Japan.  GPW Cup Computer Go 
>Tournament will be held at the first and second nights, together with 
>the other night events, such as GPW Cup Computer Shogi Tournament, 
>using Chinese rules, round robin alternating black and white, 15 
>minutes SD, and 6.5 points komi.  Participants have to prepare 
>necessary equipments by themselves (no Internet connection is 
>available at the Seminor House, no loan PCs and/or no agents are 
>provided by the organizer).  The programs should be original.  Human 
>players can participate as well.
>
>The rules of the tournament may be changed on particular conditions if 
>necessary, ex., if there will be too many participants to keep the 
>schedule.
>
>Hideki
>--
>[EMAIL PROTECTED] (Kato)
>___
>computer-go mailing list
>computer-go@computer-go.org
>http://www.computer-go.org/mailman/listinfo/computer-go/
--
[EMAIL PROTECTED] (Kato)
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/


[computer-go] Monte carlo play

2008-11-11 Thread tony tang

Greetings all go gurus,
 
I am new to programming go, could some one explain to me how a monte carlo 
based evalution manages to play random games by itself?
ie: who/what is the oppoent which supplies the opposing moves which allows 
another move to be randomly played after making the initial move at the root.
I am implementing in java, is there a package/framework which allows me to 
train my software.
 
Thank you
 
Kind regards
 
Tony
_
See the most popular videos on the web 
http://clk.atdmt.com/GBL/go/115454061/direct/01/___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/

Re: [computer-go] Monte carlo play

2008-11-11 Thread Darren Cook
> I am new to programming go, could some one explain to me how a monte
> carlo based evalution manages to play random games by itself? ie:
> who/what is the oppoent which supplies the opposing moves which
> allows another move to be randomly played after making the initial

It is self-play, so both players are random.

I'm not aware of any programs that use different playout strategies for
black/white, though it has been briefly mentioned before on this list,
and I personally think it might be interesting as a way of discovering
and more accurately evaluating the imbalances present in most go
positions (i.e. usually one side has more strength and one side has more
territory, and so to correctly understand the position one side has to
play aggressive moves and the other has to play defensive moves).

> move at the root. I am implementing in java, is there a
> package/framework which allows me to train my software.

This page, recently started by Eric Marchand, might be a good starting
place:
  http://ricoh51.free.fr/go/engineeng.htm

Darren


-- 
Darren Cook, Software Researcher/Developer
http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic
open source dictionary/semantic network)
http://dcook.org/work/ (About me and my work)
http://dcook.org/blogs.html (My blogs and articles)
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/