Re: [Computer-go] Orego 7.08 released
Thank you, professor. I am looking forward to your upcoming paper. Thanks, Aja - Original Message - From: Peter Drake To: Aja ; computer-go@dvandva.org Sent: Tuesday, January 11, 2011 4:21 PM Subject: Re: [Computer-go] Orego 7.08 released This version's default player (Lgrf2Player) uses the last-good-reply policy, and in fact an improved version described in an upcoming paper. We do not use the Elo-based heavy playouts, because we were never able to get them to run quickly enough to really offer an improvement. Peter Drake http://www.lclark.edu/~drake/ On Jan 10, 2011, at 8:27 PM, Aja wrote: Hi professor Drake, I read your paper THE LAST-GOOD-REPLY POLICY FOR MONTE-CARLO GO and was very surprised with the performance of the heuristic The Last-Good-Reply Policy. In your experiment, it boosts the wining rate from around 40% to almost 60%. I wonder does this version of Orego feature this heuristic? Or maybe you have combine this heuristic with the Elo-Based Heavy Playouts described in your paper Investigating the E ects of Playout Strength in Monte-Carlo Go? Thanks, Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Orego 7.08 released
Yes, the Power of Forgetting paper. Peter Drake http://www.lclark.edu/~drake/ On Jan 11, 2011, at 3:47 AM, Aja wrote: I discover that this upcoming paper is already available in your website http://legacy.lclark.edu/~drake/Orego.html I will try this interesting idea in Erica. Aja - Original Message - From: Aja To: Peter Drake ; computer-go@dvandva.org Sent: Tuesday, January 11, 2011 5:53 PM Subject: Re: [Computer-go] Orego 7.08 released Thank you, professor. I am looking forward to your upcoming paper. Thanks, Aja - Original Message - From: Peter Drake To: Aja ; computer-go@dvandva.org Sent: Tuesday, January 11, 2011 4:21 PM Subject: Re: [Computer-go] Orego 7.08 released This version's default player (Lgrf2Player) uses the last-good-reply policy, and in fact an improved version described in an upcoming paper. We do not use the Elo-based heavy playouts, because we were never able to get them to run quickly enough to really offer an improvement. Peter Drake http://www.lclark.edu/~drake/ On Jan 10, 2011, at 8:27 PM, Aja wrote: Hi professor Drake, I read your paper THE LAST-GOOD-REPLY POLICY FOR MONTE-CARLO GO and was very surprised with the performance of the heuristic The Last-Good-Reply Policy. In your experiment, it boosts the wining rate from around 40% to almost 60%. I wonder does this version of Orego feature this heuristic? Or maybe you have combine this heuristic with the Elo-Based Heavy Playouts described in your paper Investigating the E ects of Playout Strength in Monte-Carlo Go? Thanks, Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Orego 7.08 released
On Tue, Jan 11, 2011 at 08:45:31AM -0800, terry mcintyre wrote: Drake, has your research proved the proverb Learn Joseki, Lose Three Ranks? I have a theory, based on my own experience with joseki. Knowing the right move is not enough; one must know what to do with the wrong moves also. Joseki often skate at the edge between good for black and good for white, and they also tend to be strongly influenced by conditions such as ladders. In Pachi, I'm using joseki sequences that were automatically extracted from Kogo branches marked as GOOD VARIATION. The overall efect against other programs has been mostly neutral, possibly a decrease by very few elo points. However, my subjective impression has been that the program has more success against humans with joseki enabled, and doubtlessly its play is much more pleasant to watch and play against, so I think it is worth the tradeoff. Note that I am not using joseki unconditionally, just as another heuristic among all the others. So if Pachi does not like the joseki, it is free to play elsewhere. Sometimes yes, it gets into a sequence that it ultimately misplays, but it seems not to be the majority of cases. So, probably a very boring neutral rating at least from me. :-) -- Petr Pasky Baudis Computer science education cannot make an expert programmer any more than studying brushes and pigment can make an expert painter. --esr ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Orego 7.08 released
Many Faces does something similar, but I also include all the responses to bad moves, so when the opponent makes a bad move the program knows how to punish it. I include all moves from every joseki book published in English through about 2003. I added a few joseki from a 5 volume Japanese joseki dictionary too. I also found that it makes no real difference to strength against computers, but helps a little against people, and makes the games much more peasant to watch. Part of my goal is to make a program that people can learn from, so this is another reason for it to know and play joseki. David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Petr Baudis Sent: Tuesday, January 11, 2011 9:23 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] Orego 7.08 released On Tue, Jan 11, 2011 at 08:45:31AM -0800, terry mcintyre wrote: Drake, has your research proved the proverb Learn Joseki, Lose Three Ranks? I have a theory, based on my own experience with joseki. Knowing the right move is not enough; one must know what to do with the wrong moves also. Joseki often skate at the edge between good for black and good for white, and they also tend to be strongly influenced by conditions such as ladders. In Pachi, I'm using joseki sequences that were automatically extracted from Kogo branches marked as GOOD VARIATION. The overall efect against other programs has been mostly neutral, possibly a decrease by very few elo points. However, my subjective impression has been that the program has more success against humans with joseki enabled, and doubtlessly its play is much more pleasant to watch and play against, so I think it is worth the tradeoff. Note that I am not using joseki unconditionally, just as another heuristic among all the others. So if Pachi does not like the joseki, it is free to play elsewhere. Sometimes yes, it gets into a sequence that it ultimately misplays, but it seems not to be the majority of cases. So, probably a very boring neutral rating at least from me. :-) -- Petr Pasky Baudis Computer science education cannot make an expert programmer any more than studying brushes and pigment can make an expert painter. --esr ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Running automated bot matches?
I've used gomill (http://mjw.woodcraft.me.uk/gomill/). It's a collection of python scripts to automate running tests. The tournaments are organized by defining a control file that's just pure python, which is nice. I believe one of the common programs (libego or fuego maybe) have scripts for running gtp games as well. - James On Tue, Jan 11, 2011 at 9:41 AM, Joona Kiiski joona.kii...@gmail.comwrote: Hi everyone, During the last week I've been examining sources of different open source go-engines (fuego, pachi, orego). Now I'd like to start making some simple modifications to some of them (not yet decided which one) and see how it goes (likely my 50 first tries will fail miserably, but it's okay). In computer chess programming, it's nowadays a widely accepted fact that only reasonable way to test changes is to run a huge number of test games between original and modified version. I assume that same applies also for go-programming. So, let us have open-source program X and slightly modified version it X'. What is the easiest way to run say 1000 super-fast games between them? I hope there already exists some scripts or programs to do this. My OS is Linux if it matters. Thanks for your help! ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Running automated bot matches?
Orego contains programs (in the orego.experiment package) for running a bunch of test programs against a standard opponent such as GNU Go. There's no reason that standard opponent couldn't be another version of Orego. The experiment scripts used to be in Python, but I changed them to Java in this version so that new researchers wouldn't have to learn another language. Peter Drake http://www.lclark.edu/~drake/ On Jan 11, 2011, at 9:41 AM, Joona Kiiski wrote: Hi everyone, During the last week I've been examining sources of different open source go-engines (fuego, pachi, orego). Now I'd like to start making some simple modifications to some of them (not yet decided which one) and see how it goes (likely my 50 first tries will fail miserably, but it's okay). In computer chess programming, it's nowadays a widely accepted fact that only reasonable way to test changes is to run a huge number of test games between original and modified version. I assume that same applies also for go-programming. So, let us have open-source program X and slightly modified version it X'. What is the easiest way to run say 1000 super-fast games between them? I hope there already exists some scripts or programs to do this. My OS is Linux if it matters. Thanks for your help! ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Running automated bot matches?
http://www.lysator.liu.se/~gunnar/gtp/ id your source for GTP related stuff. Contains both Perl and Python script for running loads of games. I do wonder how you plan to make it fast though :) Petri 2011/1/11 Joona Kiiski joona.kii...@gmail.com Hi everyone, During the last week I've been examining sources of different open source go-engines (fuego, pachi, orego). Now I'd like to start making some simple modifications to some of them (not yet decided which one) and see how it goes (likely my 50 first tries will fail miserably, but it's okay). In computer chess programming, it's nowadays a widely accepted fact that only reasonable way to test changes is to run a huge number of test games between original and modified version. I assume that same applies also for go-programming. So, let us have open-source program X and slightly modified version it X'. What is the easiest way to run say 1000 super-fast games between them? I hope there already exists some scripts or programs to do this. My OS is Linux if it matters. Thanks for your help! ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Orego 7.08 released : Power of Forgetting paper
The discussion changed the subject to joseki, but the paper is not about joseki at all. (There is a poster about joseki in the same website.) The Power of Forgetting is an improvement to the previous Last-Good Reply idea. The results are spectacular and implementation is super-simple. Looks like RAVE applied to playouts, the simple heuristic that beats more ambitious ideas. It improves a MoGo-like policy: (capture, escape, 3x3) from ~10% to ~35% with 8K playouts and from ~25% to ~65% with 32K playouts. (Winrate against GnuGo) in 19x19! Something i guess, everybody will want to try. I will. Jacques. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Primble: Scrabble with prime numbers
One more message on Scrable: influenced by Brian's dissertation, one of my students (Thomas Rolle) had to read and to explain the contents of the thesis in his oral exam, in 2002. Thomas caught fire, and soon later wrote a simple bot for a Scrabble variant: (decimal) digits instead of letters, and prime numbers instead of words. So, this is a really international version where each person - independent of his mother tongue - has in principle the same chances. The data base for the bot contained all primes with at most seven digits. Typical semi-full board positions looked like in the picture on http://www.althofer.de/primble.html As a human player I was completely chanceless against the bot... Ingo. Original-Nachricht Datum: Wed, 05 Jan 2011 00:21:14 +0100 Von: Ingo Althöfer 3-hirn-ver...@gmx.de An: computer-go@dvandva.org Betreff: [Computer-go] Special Offer: Towards perfect play of Scrabble Hello, on the website www.icga.org I just found the item Books for sale at ICGA (special offer) in the list on the left. In this books list there are several titles with reduced price. I would like to promote one of them, from the sublist Ph.D. Theses, namely Brian Sheppard: Towards Perfect Play of Scrabble Normal (old price): 25 € Special offer (now): 15 € For ordering information, contact i...@icga.org It is a thesis from 2002 (University of Maastricht). I bought it in that year, and for me it was a high calibre eye opener. It is very smooth to read, and contains a lot of interesting stuff, even for those people who are not especially engaged in computer Scrabble. The core part is Chapter 10, on simulations (called Monte Carlo runs in other fields). I like especially Section 10.5 on the historical development (hand simulations by Ron Tiekert in the 1980's). Without this book I would likely not have started to use pure Monte Carlo in my programs for computer aided game inventing in 2004. Ingo. -- GMX DSL Doppel-Flat ab 19,99 Euro/mtl.! Jetzt mit gratis Handy-Flat! http://portal.gmx.net/de/go/dsl ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Running automated bot matches?
Hi! On Tue, Jan 11, 2011 at 07:41:12PM +0200, Joona Kiiski wrote: So, let us have open-source program X and slightly modified version it X'. What is the easiest way to run say 1000 super-fast games between them? I hope there already exists some scripts or programs to do this. It is better to choose another appropriately strong program as a reference player rather than use self-play - the results tend to be rather misleading, and many people have been bitten by this. The general way to have two programs play against each other is using a 'twogtp' tool. There are several different variations of this tool that differ quite a bit in their usage and featureset; I think most people use gogui-twogtp nowadays. If you want to automate playing many games using twogtp and aggregating the results, aside of gomill, you can use the Autotest framework of Pachi: http://repo.or.cz/w/pachi.git/tree/HEAD:/t-play/autotest (It can be used with other programs than Pachi too.) Both have some advantages and disadvantages, so pick the one you like more. :-) -- Petr Pasky Baudis Computer science education cannot make an expert programmer any more than studying brushes and pigment can make an expert painter. --esr ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Orego 7.08 released : Power of Forgetting paper
This is on my list of things to try, and I am hoping that other people who try it will describe their experiences in this forum. What I perceive is that Orego, while Mogo-like, still has a fairly light playout policy. For example, in the paper, Orego (using the killer-reply-with-forgetting heuristic) defeats Gnugo maybe 87% at 32K trials on 9x9, whereas Pebbles defeats GnuGo 93% using 10K trials. Orego places the killer-reply heuristic as the very first rule applied. I speculate that engines that employ heavier playouts will benefit from placing a killer heuristic farther down the rule set. Anyway, I hope to test something in the next few months, and I will report on results here. Brian -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go-boun...@dvandva.org] On Behalf Of Jacques Basaldúa Sent: Tuesday, January 11, 2011 1:04 PM To: computer-go@dvandva.org Subject: [Computer-go] Orego 7.08 released : Power of Forgetting paper The discussion changed the subject to joseki, but the paper is not about joseki at all. (There is a poster about joseki in the same website.) The Power of Forgetting is an improvement to the previous Last-Good Reply idea. The results are spectacular and implementation is super-simple. Looks like RAVE applied to playouts, the simple heuristic that beats more ambitious ideas. It improves a MoGo-like policy: (capture, escape, 3x3) from ~10% to ~35% with 8K playouts and from ~25% to ~65% with 32K playouts. (Winrate against GnuGo) in 19x19! Something i guess, everybody will want to try. I will. Jacques. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Running automated bot matches?
So, let us have open-source program X and slightly modified version it X'. What is the easiest way to run say 1000 super-fast games between them? I hope there already exists some scripts or programs to do this. For testing Fuego in self-play and against other engines we use gogui-twogtp. http://gogui.sourceforge.net/doc/reference-twogtp.html It can produce nice html output. Here are some sample scripts. http://sourceforge.net/apps/trac/fuego/wiki/RunTestGames Martin___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Question about Zone Mode in MoGo?
Watching StoneGrid aimlessly playing on the large board, I am thinking of implementing the Zone mode mentioned in the MoGo's original paper on 19x19 board. I am thinking of applying the zone mode in the tree part only, i.e. giving higher weight to the zone where the last move of the node belongs to when evaluating the final position of each simulation. Has anyone tried the zone mode other than the MoGo team? Is it successful? ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Zen!
In message c3521a2f56794359829e0e86ff328...@homepce1bd7763, Aja ajahu...@gmail.com writes Hi Nick, Thanks for the report. I have one correction: Erica was running on i7 950, 2 cores with 4 threads, **2.70 GHz, not 2.7 GHz. Thanks, fixed. In the report you said pachi was running on 20-core system after round 4. Is that true? I remember Petr said pachi was running on single core for the whole tournament, including the game against Zen. Maybe I am wrong. Both pasky and Jean-loup Gailly (who was actually running pachi2) told me that the problem was discovered and fixed before the start of round 4. Nick Aja - Original Message - From: Nick Wedd n...@maproom.co.uk To: computer-go@dvandva.org Sent: Tuesday, January 11, 2011 3:52 AM Subject: [Computer-go] Congratulations to Zen! Congratulations to Zen, undefeated winner of yesterday's KGS bot tournament! My report is at http://www.weddslist.com/kgs/past/67/index.html As usual, I welcome your comments and corrections. My report says little about the actual games; most of the programs are several stones stronger than me, and I have nothing constructive to say. Nick -- Nick Weddn...@maproom.co.uk ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Nick Weddn...@maproom.co.uk ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Orego 7.08 released
I re-post because the format seems to be in a mess. Hi David, I also found that it makes no real difference to strength against computers, but helps a little against people, and makes the games much more peasant to watch. I haven't try joseki in Erica, but it looks strange to me that you said joseki makes no real difference to strength against computers, but helps a little against people. At least, in the game mfgo against Erica in this KGS tournament, mfgo was leading from the beginning mainly because of good joseki replies in each corner. I think joseki is very important for Go programs as soon as they reach 1d level. I believe, a Go program will never reach stable high dan (=KGS 4d) without joseki knowledge. This is the same with the situations of human learning. When a player is weaker than 1d, joseki is not so important, because if he is leading 10 points in the opening stage, the game might be reversed by losing 20 points in an easy semeai of middle game. But, when a player is improved to 1d or 2d, joseki starts to make sense, since his reading ability makes the semeai big loss much fewer. For me, I can't imagine to beat a 6d player without joseki knowledge. When I lose 10 points in the opening, that is almost decisive. That's why pros sometimes resign early and immediately after wrong joseki playing, because there is no chance to reverse, in their view. The stronger the playing strengh, the more important the opening play. 9x9 Go is exactly a good example for statement. Do you think mfgo, on 9x9, can beat a strong program, if the first move is played at the first line? :) Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Orego 7.08 released
The stronger the playing strengh, the more important the opening play. 9x9 Go is exactly a good example for statement. Do you think mfgo, on 9x9, can beat a strong program, if the first move is played at the first line? :) Aja No, but that's not joseki. Ordinary search finds 5-5 the best move. You may be right that joseki helps a lot now. Years ago, I don't think joseki helped, but the program was weaker. David ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Orego 7.08 released
Zen uses no opening book for 19x19 (but some joseki knowledge must provided by the patterns acquired from game records) Hello Kato-san, Does Zen use patterns bigger than 3x3 then? (And if so, in the playouts too, or just in the MCTS tree?) On the subject of joseki, it seemed Many Faces came off equal or distinctly worse in the joseki in the games against John Tromp. So, I think it needs still more joseki knowledge? By the way, in game 1 John played a move (G15) that was not joseki (F15 is apparently the joseki move). John read the KGS comments between games, and played the correct move when the same pattern came up in game 2 :-) Darren Yamato once tried but made Zen weaker in benchmarks, possibly due to a mismatching of the playing style. Hideki This is the same with the situations of human learning. When a player is weaker than 1d, joseki is not so important, because if he is leading 10 points in the opening stage, the game might be reversed by losing 20 points in an easy semeai of middle game. But, when a player is improved to 1d or 2d, joseki starts to make sense, since his reading ability makes the semeai big loss much fewer. For me, I can't imagine to beat a 6d player without joseki knowledge. When I lose 10 points in the opening, that is almost decisive. That's why pros sometimes resign early and immediately after wrong joseki playing, because there is no chance to reverse, in their view. The stronger the playing strengh, the more important the opening play. 9x9 Go is exactly a good example for statement. Do you think mfgo, on 9x9, can beat a strong program, if the first move is played at the first line? :) Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Darren Cook, Software Researcher/Developer http://dcook.org/work/ (About me and my work) http://dcook.org/blogs.html (My blogs and articles) ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
