[computer-go] Re: CGOS server boardsize
Don, Why we have to have three servers for three boardsizes? Isn't it possoble to build a server that handle any boarsize? -Hideki Don Dailey: [EMAIL PROTECTED]: There has been some discussion about which additional board sizes to use for the server once it is running. Of course running all 3 board sizes is a possibility now that we will have server space, but my fear all along has been that they will kill each other. There is something to be said about numbers and you want as many programs as possible playing on the server you want to test on. Instead of asking for a lot of opinions however, I think it makes sense to put all 3 servers up and see what happens for a while. In other words you will vote with your participation. I think we will see that programs will gravitate more towards one server than another and I don't know which one that will be. If they all get reasonable usage I will leave them all up, but if one tends to get very little usage, I will bring it down later. I'll let them all stay up for a reasonable length of time. So there will be 9x9, 13x13 and 19x19, at least for the first month or so, depending on usage. For time controls, I have changed my previous position, I think I prefer somewhat faster time controls. There are disadvantages but almost many advantages. The foremost advantages is that I believe it encourages participation, more programs are likely to test on the server if they do not have to wait unduly long for solid results. Another advantage is that the games are more fun to watch. Right now, the time control for 9x9 assuming the average number of moves is roughly equivalent to the number of points on the board is about 3.7 seconds per move or 5 minutes. Using this same exact reasoning if we try to match the same rate of play per move we have this table: 9x9 - 300 seconds or 5 minutes 13x13 - 625 seconds or 10 minutes, 18 seconds. 19x19 - 1336 seconds or 22 minutes, 16 seconds per move. There is no particular reason that the time control has to be in multiples of 5 minutes except that we humans seems to be offended if things are rounded nicely for us. So we could accept those values as is, or we could round it to what to our sensibilities seems somehow more normal and use 5 minutes 10 minutes and 20 minutes for 9x9, 13x13 and 19x19 respectively. If we want to speed things up a bit, we might consider going from 3.7 seconds per move to 2.5 seconds per move. This gives the following approximate table: 9x9 - 202.5 seconds or 3 minutes, 22 seconds 13x13 - 422.5 seconds or 7 minutes 2 seconds 19x19 - 902.5 seconds or 15 minutes 2 seconds These could be rounded to 3 minutes, 7 minutes and 15 minutes or kept as is. There is some argument for making the bigger boards play faster based on the notion that you SHOULD play faster since the game will have a lot more moves in it. In this case, the time control could be set the same for all board sizes, perhaps 15 minutes per game or even 10 minutes per game. There is some appeal to having this kind of consistency, but of course the quality of the games on the big boards would suffer accordingly. Of course we don't care about absolute quality since we are testing programs against each other and we accept that they play much better at longer time controls. But we could set the average time per move faster if we were not comfortable with just making them all the same. We could do something like 5, 10, 15 or something like that. In addition to the time control, there is currently a 0.75 second gift which is configurable. The gift makes it possible for some programs with high latency connection issues to finish ridiculously long games without defaulting on time despite the fact that they are playing instantly. So fast time controls shouldn't be dominated by network speed considerations. My current default choice is: 9x9 - 5 minutes. (to keep it the same as it is.) 13x13 - 10 minutes. 19x19 - 15 or 20 minutes. Feedback? - Don - Don ___ 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/
Re: [computer-go] Ladders and UCT again
Peter Drake wrote: On Aug 1, 2008, at 8:08 AM, Mark Boon wrote: The neighbours of the last move come in the picture because usually it's only the last stone played that can be escaping a ladder and it's the neighbours of the last move that could have been put into atari. Nothing to do with the additional complexities Don mentioned. Let me give a specific example. Suppose that, during a playout, the tree leads us to this position, with O to play: *.* *...OO* *..O##a...* *...Ob* *c* *.* *.* *.* *.* Having reached the frontier of the tree, we now finish the game using Monte Carlo with a ladder reader. The last stone played, to the left of a, is trapped in a ladder, but can escape if not chased. Our ladder reader therefore suggests O play at a. For the next move in the playout, if # only reads ladders from the last move played, it will see that the O stone at a is not in a ladder, so move is suggested. The playout now turns completely random, and it's a coin toss as to whether the group will escape. It's often a good idea to bias capturing moves in the playouts, regardless whether it's a ladder or not. This would result in those stones being captured in most simulations. If we also search stones next to the last stone played, things only get slightly better. # sees that its stones are in a ladder from which they cannot escape, so it doesn't suggest b. If we tell it to play a ladder breaker in such situations, it might play c, which is fine. However, on O's next turn, c is not in a ladder, nor is any stone next to c, so no move is suggested. Specifically, O does not make the vital capture at b. It seems too expensive to search every point on the board for ladders. What to do? Don't bother with ladder breakers in playouts. If the tree part of the search has started to run a ladder, the best thing the playout can do is to keep running it until the bitter end. This way the losing player in the ladder will be rightfully punished and discouraged from trying that line of play. /Gunnar ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: CGOS server boardsize
On Sat, 2008-08-02 at 15:35 +0900, Hideki Kato wrote: Don, Why we have to have three servers for three boardsizes? Isn't it possoble to build a server that handle any boarsize? Of course it is, it's just a lot more work. I had my original prototype up and running in about 1 long day of work (although I have spent a substantial amount of time on it since then.) I designed CGOS to handle any board size but only one at a time.I could have also designed it to handle multiple time controls, rule-sets, etc but I wanted to keep it real simple and it is. Also, I have never done anything quite like a server before, so I didn't want to byte off more than I could chew. - Don -Hideki Don Dailey: [EMAIL PROTECTED]: There has been some discussion about which additional board sizes to use for the server once it is running. Of course running all 3 board sizes is a possibility now that we will have server space, but my fear all along has been that they will kill each other. There is something to be said about numbers and you want as many programs as possible playing on the server you want to test on. Instead of asking for a lot of opinions however, I think it makes sense to put all 3 servers up and see what happens for a while. In other words you will vote with your participation. I think we will see that programs will gravitate more towards one server than another and I don't know which one that will be. If they all get reasonable usage I will leave them all up, but if one tends to get very little usage, I will bring it down later. I'll let them all stay up for a reasonable length of time. So there will be 9x9, 13x13 and 19x19, at least for the first month or so, depending on usage. For time controls, I have changed my previous position, I think I prefer somewhat faster time controls. There are disadvantages but almost many advantages. The foremost advantages is that I believe it encourages participation, more programs are likely to test on the server if they do not have to wait unduly long for solid results. Another advantage is that the games are more fun to watch. Right now, the time control for 9x9 assuming the average number of moves is roughly equivalent to the number of points on the board is about 3.7 seconds per move or 5 minutes. Using this same exact reasoning if we try to match the same rate of play per move we have this table: 9x9 - 300 seconds or 5 minutes 13x13 - 625 seconds or 10 minutes, 18 seconds. 19x19 - 1336 seconds or 22 minutes, 16 seconds per move. There is no particular reason that the time control has to be in multiples of 5 minutes except that we humans seems to be offended if things are rounded nicely for us. So we could accept those values as is, or we could round it to what to our sensibilities seems somehow more normal and use 5 minutes 10 minutes and 20 minutes for 9x9, 13x13 and 19x19 respectively. If we want to speed things up a bit, we might consider going from 3.7 seconds per move to 2.5 seconds per move. This gives the following approximate table: 9x9 - 202.5 seconds or 3 minutes, 22 seconds 13x13 - 422.5 seconds or 7 minutes 2 seconds 19x19 - 902.5 seconds or 15 minutes 2 seconds These could be rounded to 3 minutes, 7 minutes and 15 minutes or kept as is. There is some argument for making the bigger boards play faster based on the notion that you SHOULD play faster since the game will have a lot more moves in it. In this case, the time control could be set the same for all board sizes, perhaps 15 minutes per game or even 10 minutes per game. There is some appeal to having this kind of consistency, but of course the quality of the games on the big boards would suffer accordingly. Of course we don't care about absolute quality since we are testing programs against each other and we accept that they play much better at longer time controls. But we could set the average time per move faster if we were not comfortable with just making them all the same. We could do something like 5, 10, 15 or something like that. In addition to the time control, there is currently a 0.75 second gift which is configurable. The gift makes it possible for some programs with high latency connection issues to finish ridiculously long games without defaulting on time despite the fact that they are playing instantly. So fast time controls shouldn't be dominated by network speed considerations. My current default choice is: 9x9 - 5 minutes. (to keep it the same as it is.) 13x13 - 10 minutes. 19x19 - 15 or 20 minutes. Feedback? - Don - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing
[computer-go] CGOS up and running
CGOS is up and running again, start your bots! It's possible we lost 1 or 2 of the last round games. I'm not sure of this. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] CGOS up and running - NOT
Sorry, there are some problems I have to iron out. Unfortunately, I have an important commitment and I won't be able to get to this until later in the afternoon. So please bear with me. It will probably be much later today. - Don On Sat, 2008-08-02 at 08:30 -0400, Don Dailey wrote: CGOS is up and running again, start your bots! It's possible we lost 1 or 2 of the last round games. I'm not sure of this. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] cgos playing games
Ok, it seems to be playing games now. Somehow the database was corrupted so I just removed it so that it gets recreated. It is basically archived data and we have the games stored elsewhere, so no real loss. I can even recover this data if I want by writing some scripts but I have lost timing information (which I don't currently use anyway.)I think what happens is that you cannot load old games in the viewer, which is a feature I doubt many of you use. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] cgos playing games
What's the connection info for the 3 new servers? I looked at the boardspace page, but it was out of date Sent from my iPhone On Aug 2, 2008, at 9:09 AM, Don Dailey [EMAIL PROTECTED] wrote: Ok, it seems to be playing games now. Somehow the database was corrupted so I just removed it so that it gets recreated. It is basically archived data and we have the games stored elsewhere, so no real loss. I can even recover this data if I want by writing some scripts but I have lost timing information (which I don't currently use anyway.)I think what happens is that you cannot load old games in the viewer, which is a feature I doubt many of you use. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Ladders and UCT again
On Aug 2, 2008, at 4:31 AM, Gunnar Farnebäck [EMAIL PROTECTED] wrote: It's often a good idea to bias capturing moves in the playouts, regardless whether it's a ladder or not. This would result in those stones being captured in most simulations. What method do people use for finding capture moves in playouts? Pseudo liberties can miss simple stuff like open triangles and one- eyed groups. Additionally, some literature discusses captures to add group liberties. What's the preferred method to detect that?___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Ladders and UCT again
On Sat, Aug 2, 2008 at 9:43 AM, Jason House [EMAIL PROTECTED] wrote: On Aug 2, 2008, at 4:31 AM, Gunnar Farnebäck [EMAIL PROTECTED] wrote: It's often a good idea to bias capturing moves in the playouts, regardless whether it's a ladder or not. This would result in those stones being captured in most simulations. What method do people use for finding capture moves in playouts? Pseudo liberties can miss simple stuff like open triangles and one-eyed groups. Additionally, some literature discusses captures to add group liberties. What's the preferred method to detect that?___ When we wrote dimwit, John Tromp found a fast method that I described here: http://computer-go.org/pipermail/computer-go/2007-November/012342.html However, my current thinking is that it's probably best to just keep a real liberty count and a list of liberties for each chain. This way you can also find atari moves, which would be very hard to do if you only keep pseudo-liberties. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Ladders and UCT again
Depending on your implementation, it may be faster to re-derive the liberty list when you need it. For example, if your playout move suggester only suggests capturing the last stone played, you don't need to do all of the work to update liberty counts for any other chains. Thanks a lot for everyone's help here. I think I have my ladder- reader working, although my program was still playing out ladders. The solution turned out to be something completely different: turn up the exploration coefficient in the UCT formula. If it's too low, the program commits to a particular move too early, builds up a lot of playouts through that move, and keeps playing the move even though it has read a sequence of follow-ups that lead to a bad result. Peter Drake http://www.lclark.edu/~drake/ On Aug 2, 2008, at 7:06 AM, Álvaro Begué wrote: On Sat, Aug 2, 2008 at 9:43 AM, Jason House [EMAIL PROTECTED] wrote: On Aug 2, 2008, at 4:31 AM, Gunnar Farnebäck [EMAIL PROTECTED] wrote: It's often a good idea to bias capturing moves in the playouts, regardless whether it's a ladder or not. This would result in those stones being captured in most simulations. What method do people use for finding capture moves in playouts? Pseudo liberties can miss simple stuff like open triangles and one-eyed groups. Additionally, some literature discusses captures to add group liberties. What's the preferred method to detect that?___ When we wrote dimwit, John Tromp found a fast method that I described here: http://computer-go.org/pipermail/computer-go/2007-November/ 012342.html However, my current thinking is that it's probably best to just keep a real liberty count and a list of liberties for each chain. This way you can also find atari moves, which would be very hard to do if you only keep pseudo-liberties. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Ladders and UCT again
I keep liberty counts. David -Original Message- From: [EMAIL PROTECTED] [mailto:computer-go- [EMAIL PROTECTED] On Behalf Of Jason House Sent: Saturday, August 02, 2008 6:43 AM To: computer-go Subject: Re: [computer-go] Ladders and UCT again On Aug 2, 2008, at 4:31 AM, Gunnar Farneb�ck [EMAIL PROTECTED] wrote: It's often a good idea to bias capturing moves in the playouts, regardless whether it's a ladder or not. This would result in those stones being captured in most simulations. What method do people use for finding capture moves in playouts? Pseudo liberties can miss simple stuff like open triangles and one- eyed groups. Additionally, some literature discusses captures to add group liberties. What's the preferred method to detect that?___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] komi for 13x13 and 19x19
To you go experts: Does it make sense to use a komi of 7.5 for 13x13 and 19x19 under CGOS rules? I'm trying to set up an experimental server now for 13x13. The first will be experimental with a super fast time control then I will clear it and start the official 13x13 server.I just want to make sure everything is set up right. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Ladders and UCT again
David Fotland: [EMAIL PROTECTED]: I keep liberty counts. Me too. Also is Hiroshi. -Hideki David -Original Message- From: [EMAIL PROTECTED] [mailto:computer-go- [EMAIL PROTECTED] On Behalf Of Jason House Sent: Saturday, August 02, 2008 6:43 AM To: computer-go Subject: Re: [computer-go] Ladders and UCT again On Aug 2, 2008, at 4:31 AM, Gunnar Farnebck [EMAIL PROTECTED] wrote: It's often a good idea to bias capturing moves in the playouts, regardless whether it's a ladder or not. This would result in those stones being captured in most simulations. What method do people use for finding capture moves in playouts? Pseudo liberties can miss simple stuff like open triangles and one- eyed groups. Additionally, some literature discusses captures to add group liberties. What's the preferred method to detect that?___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ 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/
Re: [computer-go] komi for 13x13 and 19x19
On Aug 2, 2008, at 10:34 AM, Don Dailey wrote: Does it make sense to use a komi of 7.5 for 13x13 and 19x19 under CGOS rules? I don't know about 13x13, but for 19x19 you should use 6.5. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] anchor for 19x19
Does anyone remember what anchor 19x19 was using? I want to use the same one. I think it was some version of gnugo fixed at 1800 ELO right? Also, will need some volunteers to run gnugo as an anchor on 13x13 and 19x19. It's ok to have 2 or 3 anchors and you don't absolutely have to have it running 24/7 but it's good if at least 1 is running most of the time and 2 is better.They should be run with identical settings of course. ( I would also like to know the settings used on the previous server but it's probably not a huge deal if I can't get that info), we can take a guess about version and settings and it will probably be very close. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] 13x13 server up and running
Ok, the 13x13 server is up and running. Here are some temporary instructions that will probably be understandable for those with bots already running: Everything remains basically the same except the port and the location of the web pages. SIZE PORT - - 9x9 6867 13x136813 19x196819 (not up yet) The web pages are not linked yet from the main page. However the URL differs only in the initial directory path, it will be 9x9, 13x13 or 19x19 (when it's ready) and here is the main standings page for the 13x13 server as an example: http://cgos.boardspace.net/13x13/standings.html It would be nice to get a few bots on 13x13 to get it started off. The instructions for the viewers are on the main web page, but to refresh your memory here is how to view the 13x13 games: /home/drd/bin/cgosview.kit -server cgos.boardspace.net -port 6813 - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 13x13 server up and running
On Aug 2, 2008, at 1:48 PM, Don Dailey wrote: Ok, the 13x13 server is up and running. Here are some temporary instructions that will probably be understandable for those with bots already running: would be nice to get a few bots on 13x13 to get it started off. myCtest-10k-UCT is running ... Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 13x13 server up and running
On Aug 2, 2008, at 2:23 PM, Christoph Birk wrote: would be nice to get a few bots on 13x13 to get it started off. myCtest-10k-UCT is running ... Weired. I got disconnected during my first game (12) but CGOS does not mention this game as a loss for myCtest ... it ignored it entirely and the website does not show game-12. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 13x13 server up and running
I think someone rebooted the server. It was odd because I got the standard disconnect message but when I logged back in the uptime for the machine is over 1 day. Maybe the uptime is for the physical machine and not the virtual machine? - Don On Sat, 2008-08-02 at 14:36 -0700, Christoph Birk wrote: On Aug 2, 2008, at 2:23 PM, Christoph Birk wrote: would be nice to get a few bots on 13x13 to get it started off. myCtest-10k-UCT is running ... Weired. I got disconnected during my first game (12) but CGOS does not mention this game as a loss for myCtest ... it ignored it entirely and the website does not show game-12. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 13x13 server up and running
This reminds of that I have always used an ancient client to connect to the 9x9 go servers. The link for connecting to the 19x19 on http://cgos.boardspace.net/ is for an older clinet I think. So I would like to know how to connect using cgosgtp.tcl in windows with a configuraton file. I have .bat containing the single line tclkit cgosGtp.tcl -c config13.txt -k Quit.txt which leads to tcl popping up the error msg: This isn't a TK applicatinBad window path name config13.txt (The missing space and missing s in windows is as it is shown) Any hints are wellcome! Quoting Don Dailey [EMAIL PROTECTED]: Ok, the 13x13 server is up and running. Here are some temporary instructions that will probably be understandable for those with bots already running: Everything remains basically the same except the port and the location of the web pages. SIZE PORT - - 9x9 6867 13x136813 19x196819 (not up yet) The web pages are not linked yet from the main page. However the URL differs only in the initial directory path, it will be 9x9, 13x13 or 19x19 (when it's ready) and here is the main standings page for the 13x13 server as an example: http://cgos.boardspace.net/13x13/standings.html It would be nice to get a few bots on 13x13 to get it started off. The instructions for the viewers are on the main web page, but to refresh your memory here is how to view the 13x13 games: /home/drd/bin/cgosview.kit -server cgos.boardspace.net -port 6813 - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- Magnus Persson Berlin, Germany ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 13x13 server up and running
Hi Magnus. I think I have a properly up to date client now. Below is a sample configuration file I am using for gnugo. You can run it with the appropriate cgosGtp.something for your platform. I use cgosGtp.kit and use the tcl runtime for my platform but you can use one of the binaries. Or you can use the pure tcl script if you have tcl installed. You can run cgosGtp -c config_file to make it happen. You can run it without any arguments to get usage instructions. You can make it give you a sample configuration file by doing cgosGtp.tcl -s You can have multiple sections and play different versions or different programs, it will take turns using the specified priorities. If you want 8 out of 10 games to be played with programA you would set priorities 8 and 2 for program A and program B respectively. Or you could set 80 and 20, etc.You can do 8 2 0 if you have 3 programs and you do not want to play one of them. # config file for playing gnugo # - %section server server cgos.boardspace.net port 6813 %section player name Gnugo-3.7.10-a1 password somePassword invokegnugo --mode gtp --score aftermath --capture-all-dead --chinese-rules --min-level 10 --max-level 10 --positional-superko priority 7 On Sun, 2008-08-03 at 00:26 +0200, Magnus Persson wrote: This reminds of that I have always used an ancient client to connect to the 9x9 go servers. The link for connecting to the 19x19 on http://cgos.boardspace.net/ is for an older clinet I think. So I would like to know how to connect using cgosgtp.tcl in windows with a configuraton file. I have .bat containing the single line tclkit cgosGtp.tcl -c config13.txt -k Quit.txt which leads to tcl popping up the error msg: This isn't a TK applicatinBad window path name config13.txt (The missing space and missing s in windows is as it is shown) Any hints are wellcome! Quoting Don Dailey [EMAIL PROTECTED]: Ok, the 13x13 server is up and running. Here are some temporary instructions that will probably be understandable for those with bots already running: Everything remains basically the same except the port and the location of the web pages. SIZE PORT - - 9x9 6867 13x136813 19x196819 (not up yet) The web pages are not linked yet from the main page. However the URL differs only in the initial directory path, it will be 9x9, 13x13 or 19x19 (when it's ready) and here is the main standings page for the 13x13 server as an example: http://cgos.boardspace.net/13x13/standings.html It would be nice to get a few bots on 13x13 to get it started off. The instructions for the viewers are on the main web page, but to refresh your memory here is how to view the 13x13 games: /home/drd/bin/cgosview.kit -server cgos.boardspace.net -port 6813 - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Ladders and UCT again
On Sat, Aug 2, 2008 at 16:06, Álvaro Begué [EMAIL PROTECTED] wrote: On Sat, Aug 2, 2008 at 9:43 AM, Jason House [EMAIL PROTECTED] wrote: On Aug 2, 2008, at 4:31 AM, Gunnar Farnebäck [EMAIL PROTECTED] wrote: It's often a good idea to bias capturing moves in the playouts, regardless whether it's a ladder or not. This would result in those stones being captured in most simulations. What method do people use for finding capture moves in playouts? Pseudo liberties can miss simple stuff like open triangles and one-eyed groups. Additionally, some literature discusses captures to add group liberties. What's the preferred method to detect that?___ When we wrote dimwit, John Tromp found a fast method that I described here: http://computer-go.org/pipermail/computer-go/2007-November/012342.html However, my current thinking is that it's probably best to just keep a real liberty count and a list of liberties for each chain. This way you can also find atari moves, which would be very hard to do if you only keep pseudo-liberties. Do you have any proposal how to find true number of liberties? Cheers, Lukasz ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Ladders and UCT again
Can you describe your algorithms? Cheers, Lukasz On Sat, Aug 2, 2008 at 19:36, Hideki Kato [EMAIL PROTECTED] wrote: David Fotland: [EMAIL PROTECTED]: I keep liberty counts. Me too. Also is Hiroshi. -Hideki David -Original Message- From: [EMAIL PROTECTED] [mailto:computer-go- [EMAIL PROTECTED] On Behalf Of Jason House Sent: Saturday, August 02, 2008 6:43 AM To: computer-go Subject: Re: [computer-go] Ladders and UCT again On Aug 2, 2008, at 4:31 AM, Gunnar Farneb ck [EMAIL PROTECTED] wrote: It's often a good idea to bias capturing moves in the playouts, regardless whether it's a ladder or not. This would result in those stones being captured in most simulations. What method do people use for finding capture moves in playouts? Pseudo liberties can miss simple stuff like open triangles and one- eyed groups. Additionally, some literature discusses captures to add group liberties. What's the preferred method to detect that?___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ 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 mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Ladders and UCT again
I keep a count of the number of liberties for each string of stones. It's updated incrementally when each stone is placed. I only use it for finding captures. I don't read ladders during playouts. Keeping liberty counts is not expensive. I was getting 50k playouts per second on a 2.3 GHz Core2 Duo (1 CPU) just running random playouts without UCT. With UCT, 3x3 patterns, simple local tactical responses, superko test at uct nodes, etc, it slowed down to 30k playouts/second. Incorporating Many Faces and RAVE into the UCT search slowed it down to 12K playouts/second, but made it much stronger. All of these numbers are for 9x9. David -Original Message- From: [EMAIL PROTECTED] [mailto:computer-go- [EMAIL PROTECTED] On Behalf Of Lukasz Lew Sent: Saturday, August 02, 2008 3:50 PM To: computer-go Subject: Re: [computer-go] Ladders and UCT again Can you describe your algorithms? Cheers, Lukasz On Sat, Aug 2, 2008 at 19:36, Hideki Kato [EMAIL PROTECTED] wrote: David Fotland: [EMAIL PROTECTED]: I keep liberty counts. Me too. Also is Hiroshi. -Hideki David -Original Message- From: [EMAIL PROTECTED] [mailto:computer-go- [EMAIL PROTECTED] On Behalf Of Jason House Sent: Saturday, August 02, 2008 6:43 AM To: computer-go Subject: Re: [computer-go] Ladders and UCT again On Aug 2, 2008, at 4:31 AM, Gunnar Farneb ck [EMAIL PROTECTED] wrote: It's often a good idea to bias capturing moves in the playouts, regardless whether it's a ladder or not. This would result in those stones being captured in most simulations. What method do people use for finding capture moves in playouts? Pseudo liberties can miss simple stuff like open triangles and one- eyed groups. Additionally, some literature discusses captures to add group liberties. What's the preferred method to detect that?___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ 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 mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Ladders and UCT again
One is not so effective but simple. Trace all adjacent four intersections of all stones of a string with marking to avoid duplicate. The other is the combination of two dimensional shift, logical or, and logical and operations on bitboards using 128 bit (or wider) registers. -Hideki Łukasz Lew: [EMAIL PROTECTED]: Can you describe your algorithms? Cheers, Lukasz On Sat, Aug 2, 2008 at 19:36, Hideki Kato [EMAIL PROTECTED] wrote: David Fotland: [EMAIL PROTECTED]: I keep liberty counts. Me too. Also is Hiroshi. -Hideki David -Original Message- From: [EMAIL PROTECTED] [mailto:computer-go- [EMAIL PROTECTED] On Behalf Of Jason House Sent: Saturday, August 02, 2008 6:43 AM To: computer-go Subject: Re: [computer-go] Ladders and UCT again On Aug 2, 2008, at 4:31 AM, Gunnar Farneb ck [EMAIL PROTECTED] wrote: It's often a good idea to bias capturing moves in the playouts, regardless whether it's a ladder or not. This would result in those stones being captured in most simulations. What method do people use for finding capture moves in playouts? Pseudo liberties can miss simple stuff like open triangles and one- eyed groups. Additionally, some literature discusses captures to add group liberties. What's the preferred method to detect that?___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ 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 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/
