Re: [computer-go] Joseki Book
Only papers I can recall are from seventies (assuming you mean academic papers) from Wilcoxx. I may have electrical copies. Not sure though. I managed to find some of them from ACM site. That paper described position based approach where each and every stage was stored into datastructure, kinda like huge pattern matching library. Was called lenses if I remember correctly. More common way is store joseki moves as a tree. Biggest issue is always hos key in all those variations. Petri 2009/11/9 Jessica Mullins jmull...@lclark.edu Hi, I am wondering what is the best way to build a Joseki Book? I am a student at Lewis Clark College and am working with Professor Peter Drake to build a Joseki Book for the program Orego.sed aproach i.e each state of joske Right now I am extracting moves from professor players and saving those into a database. Then if during game play a position is contained in the database, play the response move like the professional. I am just wondering what other people have done to build a Joseki Book, or if anyone knows of any papers that might be helpful. Thank you, Jessica Mullins Lewis Clark College '10 ___ 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: [SPAM] Re: [computer-go] Joseki Book
There is a paper about that in http://hal.inria.fr/inria-00369783/en/ and Tristan Cazenave published something around that also. (these two works are about the automatic building of opening book in self-play) See also the references in the PDF above. Best regards, Olivier Only papers I can recall are from seventies (assuming you mean academic papers) from Wilcoxx. I may have electrical copies. Not sure though. I managed to find some of them from ACM site. That paper described position based approach where each and every stage was stored into datastructure, kinda like huge pattern matching library. Was called lenses if I remember correctly. More common way is store joseki moves as a tree. Biggest issue is always hos key in all those variations. Petri 2009/11/9 Jessica Mullins jmull...@lclark.edu Hi, I am wondering what is the best way to build a Joseki Book? I am a student at Lewis Clark College and am working with Professor Peter Drake to build a Joseki Book for the program Orego.sed aproach i.e each state of joske Right now I am extracting moves from professor players and saving those into a database. Then if during game play a position is contained in the database, play the response move like the professional. I am just wondering what other people have done to build a Joseki Book, or if anyone knows of any papers that might be helpful. Thank you, Jessica Mullins Lewis Clark College '10 ___ 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/ -- = Olivier Teytaud (TAO-inria) olivier.teyt...@inria.fr Tel (33)169154231 / Fax (33)169156586 Equipe TAO (Inria-Futurs), LRI, UMR 8623(CNRS - Universite Paris-Sud), bat 490 Universite Paris-Sud 91405 Orsay Cedex France (one of the 56.5 % of french who did not vote for Sarkozy in 2007) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Joseki Book
In message 1257750292.4af7bf140b...@webmail.lclark.edu, Jessica Mullins jmull...@lclark.edu writes Hi, I am wondering what is the best way to build a Joseki Book? I am a student at Lewis Clark College and am working with Professor Peter Drake to build a Joseki Book for the program Orego. Right now I am extracting moves from professor players and saving those into a database. Then if during game play a position is contained in the database, play the response move like the professional. I am just wondering what other people have done to build a Joseki Book, or if anyone knows of any papers that might be helpful. A simpler approach might be to find a free resource which has already done the extraction and built a tree from it, and copy that. On a broader level - it depends what you are trying to do. If you want Orego to play well in the long term, getting it to play good moves (what a professional would acknowledge as good) in the josekis must be a good thing. But there is the more difficult question of how to get the josekis in the four corners to relate to one another. A move can be good in the context of the local corner but bad in the context of the whole game. Dealing with this is, I think, difficult. Another approach is highly specific and short-term. Suppose you want to be able to beat a particular opponent, MFoG say. Then you can study MFoG's recent games, with the help of a strong player, and find positions where, just out of its joseki book, MFoG makes a bad move. Then you can tune Orego to play such josekis, and to follow up MFoG's bad move by knowing how to punish it. (I don't recommend this approach.) On an intermediate level, a joseki that is good for a professional is not necessarily so good for a kyu player. Professionals are better than weak players at using thickness, whereas solid territory is worth much the same to both. So if your objective is for Orego to become 1-dan, you should tend to prefer josekis which give low solid territorial positions, leaving the hard-to-use outer influence for its opponents. Nick -- Nick Weddn...@maproom.co.uk ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [SPAM] Re: [computer-go] Joseki Book
you could always take a joseki dictionary and build the trees by hand, if you feel that you're strong enough to work out the most common variations for the most common opening situations. s. 2009/11/9 Olivier Teytaud teyt...@lri.fr: There is a paper about that in http://hal.inria.fr/inria-00369783/en/ and Tristan Cazenave published something around that also. (these two works are about the automatic building of opening book in self-play) See also the references in the PDF above. Best regards, Olivier Only papers I can recall are from seventies (assuming you mean academic papers) from Wilcoxx. I may have electrical copies. Not sure though. I managed to find some of them from ACM site. That paper described position based approach where each and every stage was stored into datastructure, kinda like huge pattern matching library. Was called lenses if I remember correctly. More common way is store joseki moves as a tree. Biggest issue is always hos key in all those variations. Petri 2009/11/9 Jessica Mullins jmull...@lclark.edu Hi, I am wondering what is the best way to build a Joseki Book? I am a student at Lewis Clark College and am working with Professor Peter Drake to build a Joseki Book for the program Orego.sed aproach i.e each state of joske Right now I am extracting moves from professor players and saving those into a database. Then if during game play a position is contained in the database, play the response move like the professional. I am just wondering what other people have done to build a Joseki Book, or if anyone knows of any papers that might be helpful. Thank you, Jessica Mullins Lewis Clark College '10 ___ 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/ -- = Olivier Teytaud (TAO-inria) olivier.teyt...@inria.fr Tel (33)169154231 / Fax (33)169156586 Equipe TAO (Inria-Futurs), LRI, UMR 8623(CNRS - Universite Paris-Sud), bat 490 Universite Paris-Sud 91405 Orsay Cedex France (one of the 56.5 % of french who did not vote for Sarkozy in 2007) ___ 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] Joseki Book
Le 09/11/2009 à 08:04, Jessica Mullins a écrit : Hi, I am wondering what is the best way to build a Joseki Book? I am a student at Lewis Clark College and am working with Professor Peter Drake to build a Joseki Book for the program Orego. Right now I am extracting moves from professor players and saving those into a database. Then if during game play a position is contained in the database, play the response move like the professional. I am just wondering what other people have done to build a Joseki Book, or if anyone knows of any papers that might be helpful. I don't knwo how to build such a book, but Kogo's Joseki dictionnary is a huge .sgf file containging joseki + trick moves and punishment. Maybe it can be parsed to extract only joskis. Gnugo includes a joseki db, with tags (joseki, hamete ...) so maybe it can be used more easily. my 2 cents. Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Joseki Book
On Mon, Nov 09, 2009 at 02:46:11PM +0100, Alain Baeckeroot wrote: Le 09/11/2009 à 08:04, Jessica Mullins a écrit : Hi, I am wondering what is the best way to build a Joseki Book? I am a student at Lewis Clark College and am working with Professor Peter Drake to build a Joseki Book for the program Orego. Right now I am extracting moves from professor players and saving those into a database. Then if during game play a position is contained in the database, play the response move like the professional. I am just wondering what other people have done to build a Joseki Book, or if anyone knows of any papers that might be helpful. I don't knwo how to build such a book, but Kogo's Joseki dictionnary is a huge .sgf file containging joseki + trick moves and punishment. Maybe it can be parsed to extract only joskis. It has * GOOD VARIATION * and * BAD VARIATION * marks in some branches, I think you should get a pretty good coveraged by just taking these branches into consideration. But for something basic, I think re-using Gnugo database would be easiest. The other extreme is automatic pattern extraction from professional games database, either by extracting actual general patterns (see Remi Colulom's seminal paper) or by attempting to isolate only joseki sequences (a way to define a joseki sequence could be something like: subsequence of a game starting in an empty corner and with each move in distance =2 of some previous joseki move, terminated as soon as a move appears that could either be part of two joseki sequences or appears in distance 3=x=5 of the joseki stones). -- Petr Pasky Baudis A lot of people have my books on their bookshelves. That's the problem, they need to read them. -- Don Knuth ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Joseki Book
On Nov 9, 2009, at 2:39 AM, Nick Wedd wrote: On a broader level - it depends what you are trying to do. If you want Orego to play well in the long term, getting it to play good moves (what a professional would acknowledge as good) in the josekis must be a good thing. But there is the more difficult question of how to get the josekis in the four corners to relate to one another. A move can be good in the context of the local corner but bad in the context of the whole game. Dealing with this is, I think, difficult. True, but at the moment we're just interested in getting Orego to play ANY joseki, i.e., a reasonable move in some corner, rather than a disastrous tenuki. Finding the right joseki will be future work. (Orego also has a small fuseki book, which we're working to expand.) On an intermediate level, a joseki that is good for a professional is not necessarily so good for a kyu player. Professionals are better than weak players at using thickness, whereas solid territory is worth much the same to both. So if your objective is for Orego to become 1-dan, you should tend to prefer josekis which give low solid territorial positions, leaving the hard-to-use outer influence for its opponents. Good advice! We have, of course, linked goals: we want to make the program stronger, but we also want to discover techniques that might be relevant to problems other than Go; thus the focus on automatically extracting joseki. Peter Drake http://www.lclark.edu/~drake/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Citation on zero exploration?
Many of us have concluded that, with RAVE, there is no need for a UCT exploration term: http://computer-go.org/pipermail/computer-go/2009-June/018773.html Is there a published source on this result that I could cite? Thanks, Peter Drake http://www.lclark.edu/~drake/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [SPAM] Re: [computer-go] Joseki Book
One approach might be to combine some well-known joseki and fuseki books with such books as 100 tips for Amateur Players, which explain some of the pitfalls, tricks, and traps behind popular joseki. Nihon Kiin publishes some detailed and thorough joseki books. Slate and Shell published a series of workbooks by Yilun Yang; one of those suggests a few guidelines for when to tenuki, that is, when it is ok for one to ignore an approach move and play elsewhere. These might help guide a program to know when to stay on the main line of a joseki, and when to switch to another important point. There is a proverb, study joseki, lose two stones.(in strength) I don't know if anybody has used MCTS to blend local joseki knowledge with strategic knowledge to determine which joseki is most appropriate; that might be an interesting line of approach. If you store the joseki as a tree, it might make sense to expand not just one, but the several main lines in the tree, then compare their merits. One might even expand the trick play lines. I once was told by a Japanese Go professional that he was given two very strict commands when he began studying as an insei. The first was, never, ever study joseki. The second was, never ever play his father. ( who was a 3 P professional ). ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Joseki Book
From: Alain Baeckeroot alain.baecker...@laposte.net Le 09/11/2009 à 08:04, Jessica Mullins a écrit : Hi, I am wondering what is the best way to build a Joseki Book? I am a student at Lewis Clark College and am working with Professor Peter Drake to build a Joseki Book for the program Orego. Right now I am extracting moves from professor players and saving those into a database. Then if during game play a position is contained in the database, play the response move like the professional. I am just wondering what other people have done to build a Joseki Book, or if anyone knows of any papers that might be helpful. I don't knwo how to build such a book, but Kogo's Joseki dictionnary is a huge .sgf file containging joseki + trick moves and punishment. Maybe it can be parsed to extract only joskis. I have been told by stronger players that Kogo, while a useful starting point, needs to be supplemented with newer lines of play. Regarding automatic extraction of joseki from pro games - the one pitfall I see is that you'll only discover the surface of a much deeper tree of moves which don't appear in pro games - how best to respond to non-joseki plays. A move is joseki because sound refutations to non-joseki plays exist, but those refutations can be subtle; a characteristic of a trick play is that the refutation is difficult for weaker players. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [SPAM] [computer-go] Citation on zero exploration?
http://computer-go.org/pipermail/computer-go/2009-June/018773.html As I've often said something related to that (e.g. in http://hal.inria.fr/inria-00369786/fr/ ) I'd like to be more precise. What follows is for binary deterministic games, and I precise at the beginning that this is not intended to make programs much stronger. UCT (with exploration) is consistent, in the sense that asymptotically it will find an optimal move, if any winning move exists. With no exploration, this is no more true, even with rave: there are cases in which rave will only focus on one move and will simulate only this one, in spite of a poor winning rate. However, a simple trick recovers consistency: use (nbWins+K)/(nbSimulations+2K) for example, or other regularization tricks, if the weight of Rave decreases to 0 when the number of simulations goes to infinity. However, if you have patterns with negative weights and the score is biased by these patterns (possibly until a negative value), then you loose consistency whenever the weight of the patterns decrease to 0 when the number of simulations of the corresponding move goes to 0. Then, we might feel that we need more than consistency: we want consistency, and we do not want to simulate all the tree infinitely often. This can be done with some specific rules, and in mogo we have added a simple patch which, roughly, consists in simulating randomly the children nodes when the success rate of a node is very low (below 30%). This ensures that the following rules are enough for ensuring consistency: score(move) - winRate(move) -- 0 as nbSims(move) -- infinity as we optimize automatically patterns and parameters, the advantage is that we can do it without any risk of loosing consistency. More precisely, http://www.lri.fr/~teytaud/consistency.pdf (submitted; not a lot of numerical experiments, it's for theory mainly - yet, for 500 000 simulations per move, there is a significant difference, and for a specific position, it works well - there was absolutely no tuning for this). I don't claim this will make a big change in Monte-Carlo Tree Search - just a bit of theoretical understanding, and the pleasure of asymptotic consistency - maybe it will make a real difference for very long thinking time for which we can't use tuning due to the computational cost :-) and it's the first time I have more than 50% for a modification derived by theoretical ideas :-) Best regards, Olivier ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Joseki Book
Quoting terry mcintyre terrymcint...@yahoo.com: I don't knwo how to build such a book, but Kogo's Joseki dictionnary is a huge .sgf file containging joseki + trick moves and punishment. Maybe it can be parsed to extract only joskis. The problem with josekis are that most of the moves in them are not commented at all, and there are many seemingly reasonable alternatives moves that has to be punished. And just storing one counter move is not enough. And of course the whole board position is most important when a joseki breaks down because of a mistake. What I am trying to say that in order to help a weak program playing well whatever the opponent plays the joseki dictionary has to be enormous. The whole idea behind a joseki is that super strong players have been thinking about what may be playable or not and the the sequence we find in book is just the tip of the iceberg. I think it may make more sense to break down the joseki into common local patterns and let for example MCTS search among those local patterns, sometime reproducing josekis sometimes not. I think someone here wrote long ago that larger patterns do not improve at all on small pattern of some size. A joseki dictionary can be seen as using very large patterns. -Magnus -- 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] Citation on zero exploration?
By result, do you mean this observation or a quest for an explanation? If you merely wish to say that many/most current UCT programs have no need for an exploration term, then that is a context-specific (e.g. not for the E-E in Go paper) heuristic or experimental statement, not a formal one. A source for such a statement has to be more than a paper that simply notices a similar effect for their own application. One would have to reference a larger body of experimentalists or a general consensus. Just my humble opinion, Cenny Wenner On 11/9/09, Peter Drake dr...@lclark.edu wrote: Many of us have concluded that, with RAVE, there is no need for a UCT exploration term: http://computer-go.org/pipermail/computer-go/2009-June/018773.html Is there a published source on this result that I could cite? Thanks, Peter Drake http://www.lclark.edu/~drake/ ___ 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] Joseki Book
Magnus Persson wrote: I think it may make more sense to break down the joseki into common local patterns Patterns are doubtful. Even the best shape can be dead. -- robert jasiek ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [SPAM] Re: [computer-go] Citation on zero exploration?
I'm actually looking for something weaker than what Olivier has offered: a published report of the empirical finding that (for some programs, at least) an exploration coefficient of zero works best. Peter Drake http://www.lclark.edu/~drake/ On Nov 9, 2009, at 10:15 AM, Olivier Teytaud wrote: Hi; I'd like to answer your post but I must admit I've not clearly understood. My PDF file is essentially a mathematical analysis, proving that we can have consistency with some rules, without having infinitely many visits of the whole tree. UCT has the first property (consistency), but not the second (UCT visits all the tree infinitely often). This is proved under clearly stated assumptions on the problem; including deterministic two-player zero-sum games, and therefore including Go. Best regards, Olivier By result, do you mean this observation or a quest for an explanation? If you merely wish to say that many/most current UCT programs have no need for an exploration term, then that is a context-specific (e.g. not for the E-E in Go paper) heuristic or experimental statement, not a formal one. A source for such a statement has to be more than a paper that simply notices a similar effect for their own application. One would have to reference a larger body of experimentalists or a general consensus. Just my humble opinion, Cenny Wenner On 11/9/09, Peter Drake dr...@lclark.edu wrote: Many of us have concluded that, with RAVE, there is no need for a UCT exploration term: http://computer-go.org/pipermail/computer-go/2009-June/018773.html Is there a published source on this result that I could cite? Thanks, Peter Drake http://www.lclark.edu/~drake/ ___ 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/ -- = Olivier Teytaud (TAO-inria) olivier.teyt...@inria.fr Tel (33)169154231 / Fax (33)169156586 Equipe TAO (Inria-Futurs), LRI, UMR 8623(CNRS - Universite Paris-Sud), bat 490 Universite Paris-Sud 91405 Orsay Cedex France (one of the 56.5 % of french who did not vote for Sarkozy in 2007) ___ 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: [SPAM] Re: [SPAM] Re: [computer-go] Citation on zero exploration?
I don't know if a post in the computer-go mailing list is a report, but you can find numbers in this post: http://computer-go.org/pipermail/computer-go/2008-May/014854.html From the numbers I would say that it shows that all sufficiently small constants are equivalent - maybe more experiments would be interesting. Olivier I'm actually looking for something weaker than what Olivier has offered: a published report of the empirical finding that (for some programs, at least) an exploration coefficient of zero works best. Peter Drake http://www.lclark.edu/~drake/ http://www.lclark.edu/%7Edrake/ On Nov 9, 2009, at 10:15 AM, Olivier Teytaud wrote: Hi; I'd like to answer your post but I must admit I've not clearly understood. My PDF file is essentially a mathematical analysis, proving that we can have consistency with some rules, without having infinitely many visits of the whole tree. UCT has the first property (consistency), but not the second (UCT visits all the tree infinitely often). This is proved under clearly stated assumptions on the problem; including deterministic two-player zero-sum games, and therefore including Go. Best regards, Olivier By result, do you mean this observation or a quest for an explanation? If you merely wish to say that many/most current UCT programs have no need for an exploration term, then that is a context-specific (e.g. not for the E-E in Go paper) heuristic or experimental statement, not a formal one. A source for such a statement has to be more than a paper that simply notices a similar effect for their own application. One would have to reference a larger body of experimentalists or a general consensus. Just my humble opinion, Cenny Wenner On 11/9/09, Peter Drake dr...@lclark.edu wrote: Many of us have concluded that, with RAVE, there is no need for a UCT exploration term: http://computer-go.org/pipermail/computer-go/2009-June/018773.html Is there a published source on this result that I could cite? Thanks, Peter Drake http://www.lclark.edu/~drake/ http://www.lclark.edu/%7Edrake/ ___ 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/ -- = Olivier Teytaud (TAO-inria) olivier.teyt...@inria.fr Tel (33)169154231 / Fax (33)169156586 Equipe TAO (Inria-Futurs), LRI, UMR 8623(CNRS - Universite Paris-Sud), bat 490 Universite Paris-Sud 91405 Orsay Cedex France (one of the 56.5 % of french who did not vote for Sarkozy in 2007) ___ 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/ -- = Olivier Teytaud (TAO-inria) olivier.teyt...@inria.fr Tel (33)169154231 / Fax (33)169156586 Equipe TAO (Inria-Futurs), LRI, UMR 8623(CNRS - Universite Paris-Sud), bat 490 Universite Paris-Sud 91405 Orsay Cedex France (one of the 56.5 % of french who did not vote for Sarkozy in 2007) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [SPAM] Re: [computer-go] Citation on zero exploration?
On Mon, Nov 09, 2009 at 10:18:25AM -0800, Peter Drake wrote: I'm actually looking for something weaker than what Olivier has offered: a published report of the empirical finding that (for some programs, at least) an exploration coefficient of zero works best. I think you could use Combining expert, offline, transient and online knowledge in Monte-Carlo exploration for that, since there is presented an AMAF equation without any exploration term, and the final equation has no exploration term either. -- Petr Pasky Baudis A lot of people have my books on their bookshelves. That's the problem, they need to read them. -- Don Knuth ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [SPAM] Re: [computer-go] Citation on zero exploration?
My mistake. The comment was directed to the original post and not yours. I was being too slow writing a reply. (yikes, i really dislike the formatting of the http://hal.inria.fr/inria-00369786/fr article) On 11/9/09, Olivier Teytaud olivier.teyt...@lri.fr wrote: Hi; I'd like to answer your post but I must admit I've not clearly understood. My PDF file is essentially a mathematical analysis, proving that we can have consistency with some rules, without having infinitely many visits of the whole tree. UCT has the first property (consistency), but not the second (UCT visits all the tree infinitely often). This is proved under clearly stated assumptions on the problem; including deterministic two-player zero-sum games, and therefore including Go. Best regards, Olivier By result, do you mean this observation or a quest for an explanation? If you merely wish to say that many/most current UCT programs have no need for an exploration term, then that is a context-specific (e.g. not for the E-E in Go paper) heuristic or experimental statement, not a formal one. A source for such a statement has to be more than a paper that simply notices a similar effect for their own application. One would have to reference a larger body of experimentalists or a general consensus. Just my humble opinion, Cenny Wenner On 11/9/09, Peter Drake dr...@lclark.edu wrote: Many of us have concluded that, with RAVE, there is no need for a UCT exploration term: http://computer-go.org/pipermail/computer-go/2009-June/018773.html Is there a published source on this result that I could cite? Thanks, Peter Drake http://www.lclark.edu/~drake/ http://www.lclark.edu/%7Edrake/ ___ 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/ -- = Olivier Teytaud (TAO-inria) olivier.teyt...@inria.fr Tel (33)169154231 / Fax (33)169156586 Equipe TAO (Inria-Futurs), LRI, UMR 8623(CNRS - Universite Paris-Sud), bat 490 Universite Paris-Sud 91405 Orsay Cedex France (one of the 56.5 % of french who did not vote for Sarkozy in 2007) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [SPAM] Re: [computer-go] Citation on zero exploration?
On Mon, Nov 09, 2009 at 07:42:46PM +0100, Cenny Wenner wrote: My mistake. The comment was directed to the original post and not yours. I was being too slow writing a reply. (yikes, i really dislike the formatting of the http://hal.inria.fr/inria-00369786/fr article) Which, incidentally, also is a publication mentioning c=0 favourableness. :-) -- Petr Pasky Baudis A lot of people have my books on their bookshelves. That's the problem, they need to read them. -- Don Knuth ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [SPAM] Re: [computer-go] Citation on zero exploration?
Perfect! The very similar paper (by most of the same authors) Adding expert knowledge and exploration in Monte-Carlo Tree Search contains the key passage: In MoGo, the constant in front of the exploration term was not null before the introduction of RAVE values in [10]; it is now 0. Peter Drake http://www.lclark.edu/~drake/ On Nov 9, 2009, at 10:31 AM, Petr Baudis wrote: On Mon, Nov 09, 2009 at 10:18:25AM -0800, Peter Drake wrote: I'm actually looking for something weaker than what Olivier has offered: a published report of the empirical finding that (for some programs, at least) an exploration coefficient of zero works best. I think you could use Combining expert, offline, transient and online knowledge in Monte-Carlo exploration for that, since there is presented an AMAF equation without any exploration term, and the final equation has no exploration term either. -- Petr Pasky Baudis A lot of people have my books on their bookshelves. That's the problem, they need to read them. -- Don Knuth ___ 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] Joseki Book
From what David Fotland has said, Many Faces will lay out whole josekis as single moves in its searches, which seems like a great way of biasing the mcts tree early on. On Mon, Nov 9, 2009 at 13:13, Robert Jasiek jas...@snafu.de wrote: Magnus Persson wrote: I think it may make more sense to break down the joseki into common local patterns Patterns are doubtful. Even the best shape can be dead. -- robert jasiek ___ 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] Joseki Book
The tradition program did this, because it only did a very shallow global search. So in a two play global search, an entire joseki sequence would be one ply. In the MCTS version, joseki moves get a strong bias, but there is no special handling of sequences. David From: computer-go-boun...@computer-go.org [mailto:computer-go-boun...@computer-go.org] On Behalf Of ? ? Sent: Monday, November 09, 2009 7:54 PM To: computer-go Subject: Re: [computer-go] Joseki Book From what David Fotland has said, Many Faces will lay out whole josekis as single moves in its searches, which seems like a great way of biasing the mcts tree early on. On Mon, Nov 9, 2009 at 13:13, Robert Jasiek jas...@snafu.de wrote: Magnus Persson wrote: I think it may make more sense to break down the joseki into common local patterns Patterns are doubtful. Even the best shape can be dead. -- robert jasiek ___ 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] Joseki Book
David Fotland wrote: in a two play global search, an entire joseki sequence would be one ply. This works only ALA the programs don't depart from stored josekis, right? How could they adapt to non-standard global side-conditions while treating a joseki as fixed one-ply sequence? They must iteratively broaden their search again, at least locally while embedding the local stable results in a global judgement context. So pure one ply seems improper to me, although one might try to start from it using multiple local pseudo-one-ply regions before combining them by means of a possibly / hopefully only / rather global (and therefore relatively thin) search. When you say two play, do you want to stop global search after exactly two moves? Wouldn't that be an exaggeration? -- robert jasiek ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Joseki Book
Two ply (typo) was an example. The original program did one ply global search plus local quiescence. Local quiescence for a joseki move was to complete a few sequences. Obviously not ideal, but better than trying to evaluate a position in the middle of a joseki. David -Original Message- From: computer-go-boun...@computer-go.org [mailto:computer-go- boun...@computer-go.org] On Behalf Of Robert Jasiek Sent: Monday, November 09, 2009 9:38 PM To: computer-go Subject: Re: [computer-go] Joseki Book David Fotland wrote: in a two play global search, an entire joseki sequence would be one ply. This works only ALA the programs don't depart from stored josekis, right? How could they adapt to non-standard global side-conditions while treating a joseki as fixed one-ply sequence? They must iteratively broaden their search again, at least locally while embedding the local stable results in a global judgement context. So pure one ply seems improper to me, although one might try to start from it using multiple local pseudo-one-ply regions before combining them by means of a possibly / hopefully only / rather global (and therefore relatively thin) search. When you say two play, do you want to stop global search after exactly two moves? Wouldn't that be an exaggeration? -- robert jasiek ___ 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/
