Re: [computer-go] Fuego technical report
Overall, it's a good read. Nitpicks: • The scalability graphs need to be clearer. Maybe add a caption or change the single-threaded label? I looked at the graphs first and took a bit to figure out why "single-threaded" outperformed all else. • The RAVE section wasn't all too clear. I think it tried to explain too much or tried too hard to avoid math. • The scalability graphs aren't very informative once the win rate gets too high. It'd be better if the reference had something like four seconds per move instead of one • The placement of graphs at the end felt awkward, but it may be the best place for them since I believe the paper's goal is ease of reading. Maybe make the references to the graphs less explicit in the text? Sent from my iPhone On May 4, 2009, at 1:54 PM, Martin Mueller wrote: Our technical report describing the Fuego framework is now available on http://www.cs.ualberta.ca/research/techreports/2009/TR09-08.php I will probably make at least one more revision, so all feedback and suggestions are welcome. Thank you Martin ___ 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] Choosing moves in playouts.
Dear Isaac, I'm about to work on heavy playouts, and I'm not sure how to choose a move during the playout. I intend to have weights for various features. I thought about 3 versions: 1. In a position, calculate all the weights and the total weight. Then, play one move i with the probability weight_i/total_weight. Keeping cumulative weights, as Alvaro suggested, is one way to go. You can improve #1 by choosing a possible play randomly, and then making the play with the probability weight/maximum_weight. Bill Spight ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Incorporating a prior estimate
2009/5/1 Brian Sheppard : > In reading Sylvain Gelly's thesis, it seemed that incorporating a prior > estimate of winning percentage is > very important to the practical strength of Mogo. > > E.g., with 1 trials, Mogo achieved 2110 rating on CGOS, whereas my > program attempts to > reproduce existing research and is (maybe) 1900 rating with 2 to 3 > trials. The use of a > prior is an important difference, so I want to understand it more deeply. > > Some questions: > > 1) When you create a node, do you initialize > > number of simulations = C > number of wins = C * PriorEstimate() > > where C is a constant > 0? In Sylvain's thesis, the optimal C = 50, > suggesting that > incorporating a prior estimate was the equivalent of 50 UCT-RAVE trials. Yes, but for "number of RAVE simulations" and "number of RAVE wins". I think the optimal range was between 20 and 50 (you can test values in that range). The actual value certainly depends on your actual prior. > 2) Two variations were suggested. In one variation, the prior was > incorporated into the UCT > statistics of the node. In the other, the prior was incorporated into the > RAVE statistics. Charts > in the thesis do not confirm which was actually being measured. In some > cases it appears to > be the UCT version, but elsewhere it seems to be the RAVE version. Does > anyone know > what was really done? Doing it on the RAVE statistics is what is working best. > 3) Elsewhere I have seen information suggesting that Mogo initializes RAVE > statistics to > implement progressive widening. Does that conflict with the use of a prior > for RAVE initialization, > or is it in addition to the use of a prior for RAVE initialization? Progressive widening and prior for RAVE initialization serve the same purpose. The prior is maybe smoother but they should be more or less equivalent in practice. > 4) When creating a node, do you estimate the prior for that node , or for > that node's children? I estimated the prior for all move for that node (I stored the RAVE values in the node, not in the children). Sylvain > Thanks in advance, > Brian > > ___ > 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] Choosing moves in playouts.
You have the most control with option 1. You can implement this fast by keeping the sum of the weights for each row and for the total board. You then "roll" a number between 0 and total_weight, and advance through the rows subtracting the probability of each row until you would cross 0, then go along the row subtracting the probability of each point, until you would cross zero. Pick the point where the process ends. I initially implemented a similar scheme using a binary tree, and I think it was Rémi who told me about this method, which is simpler and faster in practice. You may have problems with floating-point precission doing this. The easy solution is using integers for weights, but perhaps there are ways to make the code robust while keeping the more natural floating-point values. Álvaro. On Mon, May 4, 2009 at 2:25 PM, Isaac Deutsch wrote: > Hello, > > I'm about to work on heavy playouts, and I'm not sure how to choose a move > during the playout. I intend to have weights for various features. I thought > about 3 versions: > > 1. In a position, calculate all the weights and the total weight. Then, play > one move i with the probability weight_i/total_weight. > > 2. Select a move randomly. Calculate the weight of it, then squash that > weight in the [0,1] range. Play that move with that "probability". > > 3. Same as 2., but play that move if the "probability" is higher than a > certain treshold. > > Which one do you think works best? I'm looking forward to other ideas, too. :) > > -Isaac > -- > Neu: GMX FreeDSL Komplettanschluss mit DSL 6.000 Flatrate + Telefonanschluss > für nur 17,95 Euro/mtl.!* > http://dslspecial.gmx.de/freedsl-surfflat/?ac=OM.AD.PD003K11308T4569a > ___ > 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] Choosing moves in playouts.
Hello, I'm about to work on heavy playouts, and I'm not sure how to choose a move during the playout. I intend to have weights for various features. I thought about 3 versions: 1. In a position, calculate all the weights and the total weight. Then, play one move i with the probability weight_i/total_weight. 2. Select a move randomly. Calculate the weight of it, then squash that weight in the [0,1] range. Play that move with that "probability". 3. Same as 2., but play that move if the "probability" is higher than a certain treshold. Which one do you think works best? I'm looking forward to other ideas, too. :) -Isaac -- Neu: GMX FreeDSL Komplettanschluss mit DSL 6.000 Flatrate + Telefonanschluss für nur 17,95 Euro/mtl.!* http://dslspecial.gmx.de/freedsl-surfflat/?ac=OM.AD.PD003K11308T4569a ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: Fuego technical report
Hi Martin, thanks for the information and the report. In the abstract you write "...Fuego includes a Go engine with a playing strength that is competitive with the top programs in 9x9 Go, ..." I want to support this claim. Over the weekend I had the fun to watch some free 9x9 games of Fuego on KGS, and I was really impressed by its playing strength. Good luck for Pamplona! Ingo . -- Neu: GMX FreeDSL Komplettanschluss mit DSL 6.000 Flatrate + Telefonanschluss für nur 17,95 Euro/mtl.!* http://dslspecial.gmx.de/freedsl-surfflat/?ac=OM.AD.PD003K11308T4569a ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Fuego technical report
Our technical report describing the Fuego framework is now available on http://www.cs.ualberta.ca/research/techreports/2009/TR09-08.php I will probably make at least one more revision, so all feedback and suggestions are welcome. Thank you Martin ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Monte-Carlo Simulation Balancing
Hi, We used alpha=0.1. There may well be a better setting of alpha, but this appeared to work nicely in our experiments. -Dave On 3-May-09, at 2:01 AM, elife wrote: Hi Dave, In your experiments what's the constant value alpha you set? Thanks. 2009/5/1 David Silver : Yes, in our experiments they were just constant numbers M=N=100. ___ 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] Older archives?
And just in case, here are some alternate links to the same files: * MailArchive04052003.txt http://www.filefactory.com/file/agf0d17/n/MailArchive04052003_txt * MailArchive04052003.zip http://www.filefactory.com/file/agf0dh4/n/MailArchive04052003_zip * MailArchive04052003.7z http://www.filefactory.com/file/agf0dhe/n/MailArchive04052003_7z - Original Message From: Ben Shoemaker To: computer-go Sent: Monday, May 4, 2009 9:25:46 AM Subject: Re: [computer-go] Older archives? Here's a link to the archives of the computer go mailing list from 1993 - 2003 all in one file: (available in zip, 7-zip and uncompressed) http://www.mediafire.com/?sharekey=5e8b5601844d16558d78a0e5552916099b61fa34587d11e9c95965eaa7bc68bc Ben Shoemaker. - Original Message From: Darren Cook To: computer-go Sent: Sunday, May 3, 2009 10:38:44 PM Subject: [computer-go] Older archives? The archives for this list are here: http://computer-go.org/pipermail/computer-go/ But they only go back to August 2003. Does anyone know where the older archives are to be found? Google is coming up blank. And my own archive of selected posts only goes back to 2002 for some reason. I know the list goes back to at least 1990, and has full archives, as I remember spending a good number of hours reading them all after I joined the list in around 1995. I'm interested in tracking down some posts from 1998 to 2000. Thanks, Darren -- Darren Cook, Software Researcher/Developer http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic open source dictionary/semantic network) http://dcook.org/work/ (About me and my work) http://dcook.org/blogs.html (My blogs and articles) ___ 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] Re: Monte Carlo on GPU
Michael Williams wrote: > Michael Williams wrote: >> See the April 30 2009 posting: http://www.tobiaspreis.de/ >> > > > The CUDA SDK also comes with a sample called "Monte-Carlo Option Pricing" I don't think there is much more relevance to Go than "it also uses random numbers somewhere". -- GCP ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Older archives?
Here's a link to the archives of the computer go mailing list from 1993 - 2003 all in one file: (available in zip, 7-zip and uncompressed) http://www.mediafire.com/?sharekey=5e8b5601844d16558d78a0e5552916099b61fa34587d11e9c95965eaa7bc68bc Ben Shoemaker. - Original Message From: Darren Cook To: computer-go Sent: Sunday, May 3, 2009 10:38:44 PM Subject: [computer-go] Older archives? The archives for this list are here: http://computer-go.org/pipermail/computer-go/ But they only go back to August 2003. Does anyone know where the older archives are to be found? Google is coming up blank. And my own archive of selected posts only goes back to 2002 for some reason. I know the list goes back to at least 1990, and has full archives, as I remember spending a good number of hours reading them all after I joined the list in around 1995. I'm interested in tracking down some posts from 1998 to 2000. Thanks, Darren -- Darren Cook, Software Researcher/Developer http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic open source dictionary/semantic network) http://dcook.org/work/ (About me and my work) http://dcook.org/blogs.html (My blogs and articles) ___ 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] Re: Monte Carlo on GPU
Michael Williams wrote: See the April 30 2009 posting: http://www.tobiaspreis.de/ The CUDA SDK also comes with a sample called "Monte-Carlo Option Pricing" ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Monte Carlo on GPU
See the April 30 2009 posting: http://www.tobiaspreis.de/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
