MoGo: a Grid5000-based software for the Game of Go

MoGo: logiciel de Computer-Go

(Futurs)

$\epsfig{file=line.eps,width=250mm,height=2mm}$

Contributors for TAAI 2011:

European side:
- V. Neri( Lri )
- P. Dague ( Lri )
- T. Herault( Lri )
- M. Sebag( Tao )
- M. Schoenauer( Tao )
- G. Chaslot (Univ. Maastricht)
- H. Doghmen ( Tao )
- S. Gelly( Tao )
- J.B. Hoock( Tao )
- A. Rimmel( Tao )
- O. Teytaud( Tao )
Taiwan side:

MoGo is a computer software dedicated to computer-Go. A brief video , an excellent longer video by F. Donzet and popularization in Plein-Sud special recherche 2008. It is based on

Monte-Carlo evaluation;
Tree Search principles;
High-Performance Clusters for main events.

Our team also works on other applications of Monte-Carlo Tree Search. A technical introduction here (video) .

Last news:

MoGoTW wins some games against a top amateur player in Go from random positions
MoGoTW wins 3 games (out of 4) against pros in blind 9x9 Go
MoGoTW wins against a pro in 13x13 Go with handicap 2.5
MoGoTW won a 9x9 game as black (difficult side) against a top pro
MoGoTW wins with H2 against 6D players in 13x13 Go
MoGo won the ChessBase award 2009 for the biggest contribution to computer games in 2009.
MoGoTW (version including some work by our Taiwanese colleagues) won a game as black against Fan Hui (2P) in 9x9 Go, komi 7.5. MoGo lost also a game as black and another as white - the human player was too tired for the fourth game which was scheduled. MoGo is still, to the best of our knowledge, the only program which won games as black - the most difficult setting in 9x9 Go komi 7.5; MoGo did it several times against pros, including one game against a top professional player.
MoGo's results in Barcelona 2010: MoGo was the only Bot which won his two games in 13x13 against the 6D players with handicap 2 - this is the first time a bot wins against a 6D with handicap 2 in 13x13. ManyFaces won one out of two games in this context, and Fuego lost both games. MoGo also won a game against a 9P player as white.
MoGo's results in Taiwan Open 2009 here ; nouvelle en francais ici.

MoGo sur Europe 1 (interview par Laurent Ruquier) et dans 20 minutes (un peu pas clair...).
MoGo sur Radio Suisse Romande ici . MoGo sur "The Guardian" ici . MoGo sur "Sciences et Vie" ici . MoGo sur "Le Monde" ici . MoGo sur "Micro Hebdo" ici . Dans beaucoup d'autres (notamment en Asie), j'ai perdu le compte...

Main results:

MoGo won:
- the first ever win against a pro 9p in 9x9 Go as black (agasint Zhou-Jun Xhou 9p, in Taipei, 2009) chinese audio report (National Education Radio Report) , the same in PDF , and a HTML doc here .
- the first ever win against a pro in 9x9 Go as black (against Catalin Taranu, 5p, in Rennes 2009) ( Video des champs-libres partiellement diffusée sur France-3)
- a handicap 7 game against a pro 9Dan (9p) player at Taiwan Open 2009; game available here (popularization papers here; radio francaise, interview par Laurent Ruquier ici )
- a handicap 6 game against a pro 1Dan (1p) player at Taiwan Open 2009; game available here" (popularization papers here; radio francaise, interview par Laurent Ruquier ici )
- a 9x9 game as white against a 2p/10Dan player at Taiwan Open 2009 ; game available here
- The Hakone GPW Cup 2008
- The 19x19 Computer Olympiads 2007
- 2008 World 9x9 Computer Go Championship in Taiwan, also opportunity for several games against humans ( popularization papers)
- One of the games against Catalin Taranu
- Two out of four 9x9 games against Motoki Noguchi in Clermont-Ferrand
First ever win against a professional player in even game: game won against Guo Juan, Amsterdam 2007, 9x9 Go. Guo Juan won the two other games. more details
First ever win against a professional player in even non-blitz game: game won against Catalin Taranu, Paris 2008, 9x9 Go. Catalin Taranu won the two other games. more details; some popularization papers
First ever win against a professional player in 19x19 Go: game won against Kim Myungwan, 19x19 Go, Handicap 9 (yes, Kim is stronger than MoGo, by far!). more details here

More references

Famous 19x19 games:
- The game won against Kim MyungWan (8p) on August 7th, 2008 with handicap 9 (in blitz, Kim won with handicap 9 and 11, and mogo won with handicap 12). Using 25\% of the cluster "Huygens"; detailed information here :
  
  Begin of the game; MoGo is black. H9.
  
  O2(white) will die.
  
  The left part will belong to white.
  
  Game almost finished.
  
  MoGo wins by 1.5 points.
  
  all comments welcome! sgf file
- The game organized by the National University of Tainan, won on August 26th, 2008 with handicap 5. Using 5% of the cluster "Huygens". MoGo won, below a middle game position (MoGo is black). download
- The 19x19 game lost against Catalin Taranu in spite of H9. Using a cluster provided by Bull; game organized by Recitsproque and the French Federation of Go. More information here. MoGo is black and plays a big mistake (left, MoGo plays P5 which creates a Ko for white) and looses (final position, right):
Famous 9x9 games
- 9x9 games organized on August 26th, 2008, by the National University of Tainan:
  - The game lost by MoGo as white download
  - The game won by MoGo as black download
  - The game won by MoGo as white download
- The only 9x9 game won against Catalin Taranu (out of three games). Using a cluster provided by Bull; game organized by Recitsproque and the French Federation of Go. More information here.
  - MoGo is white and wins.
Docs
- Patterns in the Monte-Carlo part:technical report
- The RAVE heuristic: Icml paper from S. Gelly and D. Silver
- Parallelization: S. Gelly, J. B. Hoock, A. Rimmel, O. Teytaud and Y. Kalemkarian. THE PARALLELIZATION OF MONTE-CARLO PLANNING. Proceedings of ICINCO 2008 (Insticc Press), pp. 198-203.
- Report from Christophe Fiter (in french): PDF
- Go-expertise: EWRL 2008 paper (including: solving the Nakade problem and other improvements in the Monte-Carlo part)
- See Sylvain Gelly's ph.D.
- IEEE TCAIG paper
Success story
Contacts (all comments welcome!):
- Jean-Yves Audibert
- Guillaume Chaslot
- Louis Chatriot
- Hassen Doghmen (doghmen@lri.fr)
- Christophe Fiter
- Sylvain Gelly (we will try not to disturb him any more with MoGo, as he now works here )
- Jean-Baptiste Hoock
- Remi Munos (the one who had the idea of plugging UCT in Monte-Carlo Go, as far as I remember (?) )
- Julien Perez
- Arpad Rimmel
- Fabien Teytaud
- Olivier Teytaud (olivier.teytaud@inria.fr)
- Yizao Wang (who developped the very successful random player)
- Remi Coulom provided many important informations for the development of MoGo (cf CrazyStone).
- Parallel-people: Thomas Herault, Vincent Danjean, Grid'5000, Ziqin Yu.
- KGS-people, Cgos-people, computer-go-mailing-list-people, Tao, Linux.
- For MoGoTW: people from various Taiwan universities, including Shi-Jim Yen, Chang-Shing Lee, Mei-Hui Wang, and others!
Download:
- Binary only for the moment . Examples of use:
  - mogo --9 --nbTotalSimulations 5000 --pondering 1
  - mogo --13 --nbTotalSimulations 5000 --pondering 0
  - mogo --19 --nbTotalSimulations 5000 --pondering 1 --nbThreads 8
- Contact for a cvs-access: olivier.teytaud@inria.fr