[Computer-go] Combinatorics of Go
Dear all, When searching for start-of-the-art of Computer Go for my thesis, I discovered a very interesting paper Combinatorics of Go by John Tromp and Gunnar Farneback. I wonder if it is the same John Tromp that played with Many Faces. If I understand correctly, they computed the State-space complexity of 19x19 Go to be 2.08168199382· 10^170, which is really a big number. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Combinatorics of Go
Definitely the same John Tromp. --Bob Solovay On Sat, Jan 1, 2011 at 1:09 AM, Aja ajahu...@gmail.com wrote: Dear all, When searching for start-of-the-art of Computer Go for my thesis, I discovered a very interesting paper Combinatorics of Go by John Tromp and Gunnar Farneback. I wonder if it is the same John Tromp that played with Many Faces. If I understand correctly, they computed the State-space complexity of 19x19 Go to be 2.08168199382· 10^170, which is really a big number. 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] Combinatorics of Go
At 01:09 AM 1/1/2011, you wrote: ... If I understand correctly, they computed the State-space complexity of 19x19 Go to be 2.08168199382· 10^170, which is really a big number. 3^(19*19)=1.740896506590319E172 is all combinations of black, white and vacant intersections on a 19 by 19 board. but some of these are illegal. off the top of my head, that number seems a bit low as it seems to be saying that only about 1.2 percent of the combinations are legal board states. thanks --- co-chair http://ocjug.org/ ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Combinatorics of Go
And to add one more point: He is also the same person that participated in the design of the Tromp-Taylor rule set. So, John is sort of an all-purpose man. Ingo. Original-Nachricht Datum: Sat, 1 Jan 2011 01:40:20 -0800 Von: Robert Solovay solo...@gmail.com An: Aja ajahu...@gmail.com, computer-go@dvandva.org Betreff: Re: [Computer-go] Combinatorics of Go Definitely the same John Tromp. --Bob Solovay On Sat, Jan 1, 2011 at 1:09 AM, Aja ajahu...@gmail.com wrote: Dear all, When searching for start-of-the-art of Computer Go for my thesis, I discovered a very interesting paper Combinatorics of Go by John Tromp and Gunnar Farneback. I wonder if it is the same John Tromp that played with Many Faces. If I understand correctly, they computed the State-space complexity of 19x19 Go to be 2.08168199382· 10^170, which is really a big number. 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 -- NEU: FreePhone - kostenlos mobil telefonieren und surfen! Jetzt informieren: http://www.gmx.net/de/go/freephone ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] News on Tromp-Cook ?
Original-Nachricht Datum: Fri, 31 Dec 2010 21:26:45 +0100 Von: Olivier Teytaud olivier.teyt...@lri.fr ... Also, there are contributors to MCTS older than MCTS - Monte-Carlo people (Cazenave, Bouzy...) and people using tree exploration in planning (Péret Garcia is one of my favorite references); also, quiescent search in alpha-beta, iterative-deepening. Finding one and only one source is a mistake. Amen. Amen. Amen! Best regards, Ingo. PS for Olivier: Prof. Dr. Rudolf Ahlswede passed away on December 18. -- NEU: FreePhone - kostenlos mobil telefonieren und surfen! Jetzt informieren: http://www.gmx.net/de/go/freephone ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Combinatorics of Go
Intriguing! A position is obviously illegal if any point is occupied by a stone surrounded by opposite-colour stones. At the 4 corners, 25 out of 27 combinations will be legal. The proportion (25/27)^4 will survive. At the 68 edges, 79 out of 81: (79/81)^68 will survive. At the 289 interior points, 241 out of 243: (241/243)^289. Multiply those, I get 0.012321913. So presumably the number is on the high side, because this calculation only takes account of single stone blocks illegally on the board. I think it's rather (much) on the low side. The probabilities are not independent. And the correlations are positive: in particular, if a single point is legal, it may be that it is next to a free intersection, which immediately makes legal four (at the center of the board) other points. Jonas ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Combinatorics of Go
On 01.01.2011 15:08, Álvaro Begué wrote: If you don't trust John's numbers It is not about trust but about taking time for understanding his proofs. -- robert jasiek ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Q about CG2006 MCTS paper
On 1 janv. 2011, at 15:13, Fuming Wang wrote: Hi Remi, Thanks for the reply. If I understand correctly, for outcomes of 0 or 1, the formula would become something like the following, right? variance = u - u^2 + 1/S Best regards, Fuming Yes, it is correct. A random variable with binary outcomes has variance = u * (1 - u). 1/S is a term to make sure it is not estimated to zero. Territory variance may change a lot, so it is important to estimate it. But for binary outcome, it is not necessary to worry with such complication, and using a constant value such as 0.25 may even work better in practice. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Fwd: News on Tromp-Cook ?
Hi Fuming, Most of the current strong programs are using UCT combined with RAVE (a kind of AMAF). The formula is like this (there are many variants), C*RAVE+(1-C)*UCT C is the weight of RAVE. As far as I know, there are at least two useful formula to compute C: 1. The first formula was proposed in the famous paper Combining Online and Offline Knowledge in UCT of Sylvain G. and David S (please refer to Section 6).(http://www.machinelearning.org/proceedings/icml2007/papers/387.pdf). 2. The second (newer) was posted here in the past. Formal reference will be David Silver's PhD thesis. This new formula, according to my testing, is 70 elo stronger than the first one. RAVE is really a big invention. It's a big contribution of Mogo. We must thank Sylvain and David for bringing such powerful method to us. :) Aja - Original Message - From: Fuming Wang To: computer-go@dvandva.org Sent: Saturday, January 01, 2011 10:16 PM Subject: Re: [Computer-go] Fwd: News on Tromp-Cook ? So, the current strong programs are more like AMAF instead of UCT, right? Fuming On Sat, Jan 1, 2011 at 11:32 AM, David Fotland fotl...@smart-games.com wrote: I still have a UCB term, but it's probably because I depend more on Many Face's move generator. I have a rave term, but it's contribution is small. It seems that if the RAVE term is large, then Rave creates enough exploration by itself. David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Petr Baudis Sent: Friday, December 31, 2010 6:27 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Fwd: News on Tromp-Cook ? Hi! On Fri, Dec 31, 2010 at 07:02:35PM +0800, Fuming Wang wrote: Now I know Remi is the first to utilize MCTS. Guess I need to read papers more carefully. I do have a question though. I thought UCT is the foundation of the current strong programs, I know that a RAVE term is added to the original UCB term, i.e. sqrt(t_total/t_i), but the UCB term is still there right? Could you eleborate a bit on why do you say UCT is not good for Go? This is quite contradictory to a lot of material on the internet regarding the lastest bread of go programs. Most likely not all (e.g. it seems not ManyFaces?), but at least many programs use exploration coefficients that are either zero or negligibly small. In Pachi, I'm using 0 as the exploration coefficient in the end, it seems to work the best. But this probably also depends on the fact that I have slight forceful randomization of playouts. 0.02 can work well on 9x9 too, but it also depends on the priors, etc. Overally, it is a question of the overall tuning of the program. But right now, reasonably strong play with only RAVE and no UCB1 is certainly possible. -- 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 -- ___ 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] Combinatorics of Go
It is really an interesting paper. I will try to understand its proof or write a program to verify it. Aja - Original Message - From: Ingo Althöfer 3-hirn-ver...@gmx.de To: computer-go@dvandva.org Sent: Saturday, January 01, 2011 6:23 PM Subject: Re: [Computer-go] Combinatorics of Go And to add one more point: He is also the same person that participated in the design of the Tromp-Taylor rule set. So, John is sort of an all-purpose man. Ingo. Original-Nachricht Datum: Sat, 1 Jan 2011 01:40:20 -0800 Von: Robert Solovay solo...@gmail.com An: Aja ajahu...@gmail.com, computer-go@dvandva.org Betreff: Re: [Computer-go] Combinatorics of Go Definitely the same John Tromp. --Bob Solovay On Sat, Jan 1, 2011 at 1:09 AM, Aja ajahu...@gmail.com wrote: Dear all, When searching for start-of-the-art of Computer Go for my thesis, I discovered a very interesting paper Combinatorics of Go by John Tromp and Gunnar Farneback. I wonder if it is the same John Tromp that played with Many Faces. If I understand correctly, they computed the State-space complexity of 19x19 Go to be 2.08168199382· 10^170, which is really a big number. 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 -- NEU: FreePhone - kostenlos mobil telefonieren und surfen! Jetzt informieren: http://www.gmx.net/de/go/freephone ___ 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] Combinatorics of Go
Hi Alvaro, I think you have perhaps misunderstood. As I read it, Arthur was refering to his own analytic result (1.232) as being on the high side, not John's result in the paper. Arthur is implicitly assuming that John's number is correct (which I think we all are), and then rationalising what the discrepancy is between his analytic result and John's. Personally I found his analysis very helpful. The way I read the reply from Jonas, he is similarly is referring to Arthur's calculation, not John's (specifically he is referring to Arthur's reasoning), other than that he also is assuming implicitly that John's calculation is correct. I hope this helps smooth over ruffled feathers. Happy New Year =) Raffles On 01/01/2011 14:08, Álvaro Begué wrote: The people that think the number is low or high have bad intuitions, that's all. Writing a program that generates random configurations and checks whether they are valid is fairly trivial. If you don't trust John's numbers, that's what you can do. Alvaro. On Saturday, January 1, 2011, Kahn Jonasjonas.k...@math.u-psud.fr wrote: Intriguing! A position is obviously illegal if any point is occupied by a stone surrounded by opposite-colour stones. At the 4 corners, 25 out of 27 combinations will be legal. The proportion (25/27)^4 will survive. At the 68 edges, 79 out of 81: (79/81)^68 will survive. At the 289 interior points, 241 out of 243: (241/243)^289. Multiply those, I get 0.012321913. So presumably the number is on the high side, because this calculation only takes account of single stone blocks illegally on the board. I think it's rather (much) on the low side. The probabilities are not independent. And the correlations are positive: in particular, if a single point is legal, it may be that it is next to a free intersection, which immediately makes legal four (at the center of the board) other points. Jonas ___ 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 mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Fwd: [computer-go] Experiments with UCT
Got it. Thx. Fuming On Fri, Dec 31, 2010 at 10:25 PM, Go Fast fas...@gmail.com wrote: -- Forwarded message -- From: Rémi Coulom remi.cou...@free.fr Date: Tue, Jul 25, 2006 at 8:22 AM Subject: [computer-go] Experiments with UCT To: computer-go computer...@computer-go.org Hi, I mentioned UCT in one of my previous messages to the list: http://zaphod.aml.sztaki.hu/papers/ecml06.pdf I tried it in Crazy Stone. I found that the algorithm described in the paper does not work well, but I managed to improve it a lot with a small change: I used 1/sqrt(20) instead of 1/sqrt(2) for the C_p constant. It now seems to work very well. Here is a summary of how it works: - Use probability of winning as score, not territory - Use the average outcome as position value - Select the move that maximizes v + sqrt((2*log(t))/(10*n)) v is the value of the move (average outcome, between 0 and 1), n the number of simulations of this move, and t the total number of simulations at the current position. In case a move has n = 0, it is selected first. Here are experiment results with Crazy Stone. 170 games are played against GNU Go 3.6 at level 10, from 85 different starting positions, alternating colors, at various time control (time per game), 1 CPU at 2.2 GHz. version 0005 UCT 2 min 40% 46.7% 4 min 48.2% 56.6% 8 min 52.9% 64.7% 16 min 57.4% 67.6% 32 min 66.6% 71.6% I have tried hard to improve it, but it seems very difficult. Using a more clever backup operator may help, but I have not managed to measure a significant difference yet. I thank Yizao for letting me know about UCT. His program, MoGo, seems to be doing very well on CGOS. Maybe Yizao can tell us more about his experiments. Rémi ___ computer-go mailing list computer...@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ 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] Combinatorics of Go
I think you have perhaps misunderstood. As I read it, Arthur was refering to his own analytic result (1.232) as being on the high side, not John's result in the paper. Arthur is implicitly assuming that John's number is correct (which I think we all are), and then rationalising what the discrepancy is between his analytic result and John's. Personally I found his analysis very helpful. The way I read the reply from Jonas, he is similarly is referring to Arthur's calculation, not John's (specifically he is referring to Arthur's reasoning), other than that he also is assuming implicitly that John's calculation is correct. I cannot speak for Arthur, but my answer was indeed specific to the quick and (not so?) dirty estimate by Arthur. I did not refer in any way to Tromp's result, nor did I try to get a real estimate. Jonas ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Combinatorics of Go
Happy New Year to all Just a note: As a go historian, I interviewed John and summarized his findings along with my other articles that have short interviews with Olivier, Remi and Dave at www.usgo.org/bobhighlibrary. Peter Shotwell ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] 2011 KGS bot tournaments
Now that I am back from London (the man/machine challenge, and refereeing the London Open Go Tournament), I must urgently plan the schedule of KGS bot tournaments for 2011. It will be something like the 2010 schedule (see http://www.weddslist.com/kgs/future.html ), but with four slow tournaments instead of one. The time limits for these four tournaments will be 12 hours, eight hours, six hours, and four hours per player. At the request of some European players, these will start at two hours before midnight GMT, rather than at midnight as in the past. The regular monthly tournaments will as usual be on Sundays, one per month, generally early in each month. There will be six 19x19, probably two 13x13, and probably four 9x9. There will be a variety of time settings, much as in 2010. I expect the next monthly tournament to be on Sunday January 8th, and to be 19x19. If anyone wishes to influence my decisions about the schedule, please let me know very soon. Nick -- 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] 2011 KGS bot tournaments
I like this proposal. I hope you also continue the annual championship. That should boost participation. These tournaments are a tremendous boon to the computer go community, and I'm really happy you are continuing them. I prefer the January tournament to be 19x19, because otherwise I have to spend a lot of time this week working on a 9x9 book. I'd prefer to have more time to develop and test a book before the next 9x9 tournament. David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Nick Wedd Sent: Saturday, January 01, 2011 12:03 PM To: computer-go@dvandva.org Subject: [Computer-go] 2011 KGS bot tournaments Now that I am back from London (the man/machine challenge, and refereeing the London Open Go Tournament), I must urgently plan the schedule of KGS bot tournaments for 2011. It will be something like the 2010 schedule (see http://www.weddslist.com/kgs/future.html ), but with four slow tournaments instead of one. The time limits for these four tournaments will be 12 hours, eight hours, six hours, and four hours per player. At the request of some European players, these will start at two hours before midnight GMT, rather than at midnight as in the past. The regular monthly tournaments will as usual be on Sundays, one per month, generally early in each month. There will be six 19x19, probably two 13x13, and probably four 9x9. There will be a variety of time settings, much as in 2010. I expect the next monthly tournament to be on Sunday January 8th, and to be 19x19. If anyone wishes to influence my decisions about the schedule, please let me know very soon. 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
Re: [Computer-go] News on Tromp-Cook ?
On Fri, Dec 31, 2010 at 1:37 PM, Jeff Nowakowski j...@dilacero.org wrote: In Computer Olympiad 2007, Steenvreter was gold medal on 9x9. ... Obviously it was following MoGo's lead with UCT (the tournament was held in June 2007, well after the remarkable success of MoGo). I don't mean to discredit Steenvreter, CrazyStone, or any other program. I'm just focusing on MoGo so much because it set the bar so high and got everybody chasing it. Not exactly; I already knew UCT before MoGo even existed. Levente's ecml paper was not the first one on UCT. He had already submitted other work (*) where he specifically applied UCT to the domain of computer games. At the time I was even considering to add UCT to Magog and have a nice reunion at the 2006 Olympiad in Turin (my former colleagues Levente Kocsis and Mark Winands were also co-authors of Magog). Unfortunately then some conference organizer screwed up... Mogo's biggest contributions, so far, in my view, are 1.Applied UCT to computer Go, and such application came from the idea MCTS that proposed in 2006 by Remi Coulom. No, it came directly from Levente. Several people got access to his paper around the same time as Remi's paper. It is more like they independently proposed similar ideas. Erik * Improved Monte-Carlo Search by Levente Kocsis, Csaba Szepesvári and Jan Willemson ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] News on Tromp-Cook ?
Usually AMAF refers to an engine that does not build a tree. On Sat, Jan 1, 2011 at 7:20 PM, Aja ajahu...@gmail.com wrote: Hi Erik, Thanks a lot. The state-of-the-art part of Compuetr Go in my thesis will be more accurate. Do you mean the whole MCTS scheme combined with UCB formula proposed by Mogo is completely inspried by Levente's work? If I understand Remi's paper correctly, Remi can change Crazy Stone's MCTS to Mogo's one by 10 mins work (including just change the selection formula to UCB)... Aja - Original Message - From: Erik van der Werf erikvanderw...@gmail.com To: computer-go@dvandva.org Sent: Sunday, January 02, 2011 6:11 AM Subject: Re: [Computer-go] News on Tromp-Cook ? On Fri, Dec 31, 2010 at 1:37 PM, Jeff Nowakowski j...@dilacero.org wrote: In Computer Olympiad 2007, Steenvreter was gold medal on 9x9. ... Obviously it was following MoGo's lead with UCT (the tournament was held in June 2007, well after the remarkable success of MoGo). I don't mean to discredit Steenvreter, CrazyStone, or any other program. I'm just focusing on MoGo so much because it set the bar so high and got everybody chasing it. Not exactly; I already knew UCT before MoGo even existed. Levente's ecml paper was not the first one on UCT. He had already submitted other work (*) where he specifically applied UCT to the domain of computer games. At the time I was even considering to add UCT to Magog and have a nice reunion at the 2006 Olympiad in Turin (my former colleagues Levente Kocsis and Mark Winands were also co-authors of Magog). Unfortunately then some conference organizer screwed up... Mogo's biggest contributions, so far, in my view, are 1.Applied UCT to computer Go, and such application came from the idea MCTS that proposed in 2006 by Remi Coulom. No, it came directly from Levente. Several people got access to his paper around the same time as Remi's paper. It is more like they independently proposed similar ideas. Erik * Improved Monte-Carlo Search by Levente Kocsis, Csaba Szepesvári and Jan Willemson ___ 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 mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] 2011 KGS bot tournaments
David Fotland: 009501cba9f2$45e61f00$d1b25d...@com: I like this proposal. I hope you also continue the annual championship. That should boost participation. These tournaments are a tremendous boon to the computer go community, and I'm really happy you are continuing them. Same for me but I like simpler formula to compute the total points for the Championship. How about ignoring the board sizes, time-settings, and the number of rounds of each tournament, ie, just using the number of players? Hideki I prefer the January tournament to be 19x19, because otherwise I have to spend a lot of time this week working on a 9x9 book. I'd prefer to have more time to develop and test a book before the next 9x9 tournament. David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Nick Wedd Sent: Saturday, January 01, 2011 12:03 PM To: computer-go@dvandva.org Subject: [Computer-go] 2011 KGS bot tournaments Now that I am back from London (the man/machine challenge, and refereeing the London Open Go Tournament), I must urgently plan the schedule of KGS bot tournaments for 2011. It will be something like the 2010 schedule (see http://www.weddslist.com/kgs/future.html ), but with four slow tournaments instead of one. The time limits for these four tournaments will be 12 hours, eight hours, six hours, and four hours per player. At the request of some European players, these will start at two hours before midnight GMT, rather than at midnight as in the past. The regular monthly tournaments will as usual be on Sundays, one per month, generally early in each month. There will be six 19x19, probably two 13x13, and probably four 9x9. There will be a variety of time settings, much as in 2010. I expect the next monthly tournament to be on Sunday January 8th, and to be 19x19. If anyone wishes to influence my decisions about the schedule, please let me know very soon. 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 -- Hideki Kato mailto:hideki_ka...@ybb.ne.jp ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Exploration formulas for UCT
Hi petr, We use the Silver formula: rave_visits / (rave_visits + real_visits + rave_visits * real_visits * 3000) The figure of 3000 is surprisingly resilient. Even with radically different heuristics and playouts, it stays the empirical optimum. Interesting. According to Sylvain's original post here, that means you set bias to sqrt(3000/4)=27.386... But is not bias should be in the range [0,1]? Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
