[Welcome to Sensei's Library!]

StartingPoints
ReferenceSection
About


Languages
Français Français

Paths
GoHistory
BeginnerGameSampl...
BeginnerGameSample3

Referenced by
GoClient
GoEditingPrograms
GGo
GoTextProtocol
PalmPilot
GoPlayingPrograms
Drago
MultiGo
EndUserLicenseAgr...
TeachYourselfGo
ComputerGo
TanGo
GoKnot
TorazuSanmoku
GoSuite
MathematicalBound...
KGSWishlist/GameH...
TartrateGossip
MultiGo/Wishlist
GoPlayingPrograms...

Homepages
Bisqwit
David
Dnerra
Evand
EwoutStam
Frs
Fwiffo

 

GNU Go
    Keywords: Software

GNU Go is a [ext] free (under the [ext] GNU GPL) go-playing program. Maintained ports exist for (at least) GNU/Linux, Unix, Windows and Mac OS (both classic and Mac OS X).

A graphical go client is not included, but GNU Go supports the Go Text Protocol and the Go Modem Protocol which can be used to connect with third-party clients as well as text-based ASCII and [ext] GNU Emacs user interfaces.

Table of contents


URLs


GNU Go at go servers


Tips

  • The --replay option is very useful to teach yourself go. It adds weighted move alternatives as comments to SGF files. (But does not keep previous comments.) The command line is as follow: gnugo -l game.sgf -o replay.sgf --replay colour where colour is the colour to replay (can be black, white or both).

ChipUni: The --replay option is useful, but it is only as good as GnuGo itself. For example, I ran the 18th Meijin title game ([ext] http://gobase.org/games/japan/titles/meijin/18/game-l21.sgf) between Rin Kaiho (9p) and Otake Hideo (9d) through GnuGo 3.5.1. It reported a global score of 7554.17 out of 12987.95 possible points. 69 of the 267 moves did not receive any scores; only 45 moves in the game were as the program predicted.

Eratos: What does it mean for GnuGo to not give a score to a move? Does that mean GnuGo didn't consider that move or something? Also, as of version 3.4 it *does* retain the original comments.

Wanted tips

  • Malweth I see that the NNGS rating of GnuGo is about 7-8k. Is it playing at level 10, or limiting itself based on the time available in the game? I've been getting stronger, and I beat it by about 6.5 on level 3 (the default in TanGo), but I'm not 8k KGS! Is there that big a difference between L3 and L10?
Evand: On nngs, it varies the level as needed to keep under time; however, it plays with long enough time settings (1/10 minimum) on a reasonably fast computer, so it rarely goes below level 10 (unless said computer is in use for other things). And yes, the difference between level 10 and level 3 is that large. I don't know why level 3 is the default, it will be very weak at that setting. Level 10 is the default level and the strongest of the "normal" levels; however, increasing the level will provide continued small increases in strength, but it will get much slower.
  • [email] Hans-Georg: I have used GoKnot as a Windows GUI for GnuGo 3.3, but it doesn't work with GnuGo 3.4. Which Windows client works with 3.4 or is there any way to adapt GoKnot?
Evand: Can you give a more complete description of what doesn't work? This is probably a result of GoKnot not suporting the newer GTP verion 2; also, some of the commands (eg the dragon_status command that GoKnot seems to make use of) have changed their output format. So, a feature request to the GoKnot author is probably in order, as the changes are probably fairly small. I can't get GoKnot to run under wine, so I can't really investigate the issue directly.
  • Frs: The --replay option (in version 3.2) is very talkative. Is there a Go program or Go client or software tool to cut these comments to a readable limit?
Nico: I have tried the latest developement version (3.3.16), and it is not verbose anymore. Only the best move (according to Gnu Go) is marked. I cannot tell wether this will still be the case for the upcoming final 3.4 version.
  • Frs: GNU Go alway plays its best move, frustrating deshis. Is there an option, so that it will play an occasional "weak" move by random? I do not want to save, reload and reset GNU Go --level's.
dnerra: If you are using a client such as gGo that communicates to GNU Go via GTP, and allows you to intercept the GTP stream in a shell-like window, you can just type "level 2" in the GTP window to change the level during play.
Toey: I am playing GNU Go 3.2 on gGo 0.3.7. Is the replay mode the one giving comments on moves? How can I make it work? I tried to put the command into the GNU Go arguments, in advance page of the preference. After doing so, there's no comment and the programme does not play at all. Probably I put it in the wrong place or I used the wrong format. Could anyone please let me know?
Nico: The replay mode does not give comments, but only the move(s) GnuGo would have played if it was its turn (actually, depending of the version, it proposes several weighted moves; the higher the better). This is a command line only option. You provide your sgf file and the colour you were playing, and GnuGo generates another sgf file with all the moves (see the tips section above).
dnerra: If you want "online" commentary, you can add "-t -w" to the command line options. This might give too much output (in the GTP window), but it shows which dragons are (according to GNU Go's calculations) alive/dead/critical (including the moves to attack/defend them), and how it valued its own possible moves.
  • mAsterdam: How to talk to GnuGo from a perlscript on Windows (XP)? A search at the mailinglist archive did not help. I looked at 2ptkgo.pl and twogtp.pl, but they don't do Windows. At the GnuGo homepage [ext] ftp://outgrabe.netchip.com/pub/go/ is mentioned, but I can't reach the site. Any information on this is welcome.
Evand: I don't use WinXP, so I can't help you with specifics, I'm afraid. However, I don't see any reasons it would be fundamentally different. Last I checked, perl on windows was fairly similar, though perhaps the calls to open the gnugo process would be different. The general idea is to spawn a gnugo process in GTP mode ('gnugo --mode gtp --quiet') and send it GTP commands, and parse the responses. I would suggest twogtp-a as the easiest of the perl scripts we ship to read, and it should provide enough to get you started. The texinfo docs that ship with gnugo should cover most of the rest, but I don't know how easy those are to read in windows. If you need more help, feel free to post to the gnugo-devel mailing list with questions.
mAsterdam: Great, Evand, thank you. twogtp-a (in the GNUGo source distribution) works exactly like I hoped it would.

History

  • 2003-Jul: GNU Go 3.4 released. A bit more than two handicap stones stronger than 3.2 and is supposed to be of about the same speed.
    • mgoetze: Yes, this is what everyone was dreaming of. In truth, however, in truth, it's not quite two stones stronger (perhaps one and a half), and actually a bit slower.
    • evand: Well, judging the speed is a bit difficult. Suffice to say that it depends heavily on the type of game, and also on the computer (performance characterists change differently on different computers). Also, if you want it faster, you should try level 9. Level 9 should be stronger than 3.2 by some amount and faster than 3.2 or of comparable speed. The other thing worth noting is that the strength differences depend strongly on your measurement techniques.
    • Concerning strength: GNU Go 3.4 wins over a half games against 3.2 with two handicap stones and komi 5.5, but less than a half if komi is 0.5. So one can say it is somewhere between 1 and 1.5 ranks stronger than 3.2. Komi matters.
    • Scryer: When I was a chess programmer I found that playing my program against itself did tell which version was better, but it would not tell accurately how much better. This is because both programs shared weaknesses: for example, the better program would believe the worse program's threats, because they were just the same threat it would make itself. To get an objective measure of the actual strength difference I needed to play humans over the board in tournament conditions. Does this factor carry over into GNU Go? I assume you've decided the answer is "no", since you feel it's accurate enough even to make the komi relevant.
    • dnerra: No, the answer is a definite "yes"! :-) Some aspects hardly appear in GNU Go vs GNU Go games, because GNU Go's style makes these games somewhat No-Contact-Games. But it's really the only thing you can measure reliably. GNU Go is playing hundreds of games on NNGS per week, but its rating can still vary highly e.g. due to some 2k* learning its weaknesses very well, and thus being able to beat it at 9 stones or so (and then playing tens of games against GNU Go...)
  • 2002-Apr: GNU Go 3.2 released. About three handicap stones stronger than 3.0, uses less RAM, and about as fast.
  • 2001-Aug: GNU Go 3.0 released. About four handicap stones stronger than 2.6.



This is a copy of the living page "GNU Go" at Sensei's Library.
(OC) 2004 the Authors, published under the OpenContent License V1.0.