Re: [computer-go] BOINC
2007/10/29, Ben Lambrechts [EMAIL PROTECTED]: Has anyone tried to program Go via BOINC? Like, trying to solve 7x7 by distributed computing? That would be interesting. (Although I'm skeptical about participation.) -- Seo Sanghyeon ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
Don Dailey wrote: Of course that's better, but I'm talking about a quick and dirty solution. I may never implement handicap games since it's tricky with ELO ratings. This can also be done by the programmers. E.g. If CrazyStone is too strong, Rèmi can introduce a CrazyStoneH3 which passes 3 times at the beginning. But not at the first move, to avoid smart tricks. If CrazyStoneH3 is given white and plays: 2. whatever 4. pass 6. pass 8. pass. Black cannot pass and win the game because of komi. This way you are not forcing programs to accept handicap (some may suffer more than others). It is just a program who always gets white and does not play against other handicap programs. Just an idea. Jacques. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Crazy Stone on 19x19 CGOS
How does one configure MoGo to do a fixed number of playouts per move? I saw only time-based command line options. On 10/27/07, Rémi Coulom [EMAIL PROTECTED] wrote: Hi, I have just connected Crazy Stone (CS-8-26-10k-1CPU). It uses 10,000 playouts per move, and runs on 1 CPU. It should finish all its games in less than 5 minutes. In my tests, it scores 41% against GNU Go 3.6 Level 10, and 73.5% against MoGo_release3 at 10k playouts per move (the playouts of Mogo are about 10% slower than those of Crazy Stone). These tests were run over 600 games, starting from 300 positions with two stones located at random (but not on the first two lines), and alternating colors. (computational power provided by the Grid5000 project: https://www.grid5000.fr/mediawiki/index.php/Grid5000:Home (they ask for advertisement in exchange)) I am a bit surprised that Crazy Stone won so easily against MoGo, because on the old server, it looked much stronger: http://cgos.boardspace.net/19x19/standings.html Olivier, do the numbers there indicate the number of playouts? Or is it playouts per processor? Maybe I messed up something. The log of Mogo indicates: 1 simulations(average length:0) done, time used: 2.94 seconds.( 3401.4 games/sec) So, it looks OK. I have the feeling that, maybe, MoGo overfits GNU more than Crazy Stone does. In particular, MoGo's romantic opening style is completely confusing for GNU, but Crazy Stone has no problem with it. I'll run Mogo 10k against GNU to check. Rémi ___ 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] Finding older CGOS games
How can I find and view older games on CGOS once they scroll off of http://cgos.boardspace.net/9x9/standings.html? I know I can recreate URL's such as http://cgos.boardspace.net/9x9/SGF/2007/10/29/176900.sgf with some pain and download all the games for the day(s) of interest. My problem is then that I don't have any of the summary info that is included on the standings page and I don't know which files to look at. Is there a better way? (I ask because I kicked off a bot 11 hours ago. It got disconnected after only 22 games and they've all scrolled off the recent game summary. Given that there was 21 losses, I'd like to check out what was going wrong, if anything) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Where is Mogo?
I don't see Mogo on the server?Where is Mogo? However CrazyStone is there to represent the Monte Carlo programs and seems to be doing a very good job indeed! CS-8-26-2CPU http://www.lri.fr/%7Eteytaud/cross/CS-8-26-2CPU.html is doing absurdly well for far, winning almost every game it plays. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] BOINC
G'day guys, I'm involved in the development of a very powerful and flexible grid software, which we plan to release in January. It is all java based. http://www-nereus.physics.ox.ac.uk/ (bear in mind you can't download it yet and the website is out of date) One of the things I'd like to do on it, once it is finished, is some kind of attack on Go. I've messed around trying to genetically generate algorithms to play go. However this has had to go on the back burner for the moment. The brief attempt I made had no way of storing data between games (I ran out of time) and the best algorithm it came up with was a purely random algorithm... :-) our group is also the one that is doing JPC - http://www-jpc.physics.ox.ac.uk/ I'd love to hear about anyone else distributed attacks on Go. cheers, Ian On 10/29/07, Sanghyeon Seo [EMAIL PROTECTED] wrote: 2007/10/29, Ben Lambrechts [EMAIL PROTECTED]: Has anyone tried to program Go via BOINC? Like, trying to solve 7x7 by distributed computing? That would be interesting. (Although I'm skeptical about participation.) -- Seo Sanghyeon ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- Fundraising Coordinator OUDC (Dancesport) [EMAIL PROTECTED] http://club.oudancesport.co.uk/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] BOINC
Not that I'm aware, but the engine I'm working on will be parallelized. Depending on time and finances I'm even considering going down the route of custom fpga based boards but that's on the dream list so far, and isn't planned for RC1. -Josh On 10/29/07, Ben Lambrechts [EMAIL PROTECTED] wrote: Hi, Has anyone tried to program Go via BOINC? regards, Ben http://boinc.berkeley.edu/ ___ 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] Where is Mogo?
Just an observation... On the cross-table page, the back to standings link is incorrect. It should point to http://www.lri.fr/~teytaud/cgosStandings.html On 10/29/07, Don Dailey [EMAIL PROTECTED] wrote: I don't see Mogo on the server?Where is Mogo? However CrazyStone is there to represent the Monte Carlo programs and seems to be doing a very good job indeed! CS-8-26-2CPU http://www.lri.fr/%7Eteytaud/cross/CS-8-26-2CPU.html is doing absurdly well for far, winning almost every game it plays. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] BOINC
We have used as many as 72 CPUs running SlugGo, but our algorithm did not scale well and we found that after 16 there was no benefit. We are trying new things, and will report positive or negative results as we get them. Cheers, David On 29, Oct 2007, at 8:03 AM, Ian Preston wrote: I'd love to hear about anyone else distributed attacks on Go. cheers, Ian ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Where is Mogo?
Olivier just configured the server wrong.In the configuration file you have to point things to the right directory and some of this is relative. I have no control over this (which is how I want it!) - Don Jason House wrote: Just an observation... On the cross-table page, the back to standings link is incorrect. It should point to http://www.lri.fr/~teytaud/cgosStandings.html http://www.lri.fr/%7Eteytaud/cgosStandings.html On 10/29/07, *Don Dailey* [EMAIL PROTECTED] mailto:[EMAIL PROTECTED] wrote: I don't see Mogo on the server?Where is Mogo? However CrazyStone is there to represent the Monte Carlo programs and seems to be doing a very good job indeed! CS-8-26-2CPU http://www.lri.fr/%7Eteytaud/cross/CS-8-26-2CPU.html is doing absurdly well for far, winning almost every game it plays. - Don ___ computer-go mailing list computer-go@computer-go.org mailto: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/
[computer-go] Standard references on CGOS
For all of us in the bot-making kiddie pool, it's exceptionally helpful to have reference implementations of basic algorithms running on the server. When playing with AMAF, I found the reference AMAF bots very helpful. Now that I'm playing with UCT, references for UCT would be helpful. I have the resources that I could run one reference bot 24/7 (I can do AMAF with 1k simulations per move). Are there others that are willing to do similar? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Where is Mogo?
I don't see Mogo on the server?Where is Mogo? we will come :-) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Standard references on CGOS
On Oct 29, 2007, at 8:39 AM, Jason House wrote: For all of us in the bot-making kiddie pool, it's exceptionally helpful to have reference implementations of basic algorithms running on the server. When playing with AMAF, I found the reference AMAF bots very helpful. Now that I'm playing with UCT, references for UCT would be helpful. 'myCtest-V-0003' is running 50k UCT. Pure random playouts guided by a UCT search with theses parameters: # playouts before expanding = 50 node-score = win_ratio + 0.5 * sqrt(log(N)/n); I will start it under the nam 'myCtest-50k-UCT' later today running 24/7. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Standard references on CGOS
Thanks! I'm not sure if my engine will support 50k simulations without running out of time in long games. Is it possible to do 10k? My engine does about 2k playouts per second. This may be a side-effect of the language I'm using. On 10/29/07, Christoph Birk [EMAIL PROTECTED] wrote: On Oct 29, 2007, at 8:39 AM, Jason House wrote: For all of us in the bot-making kiddie pool, it's exceptionally helpful to have reference implementations of basic algorithms running on the server. When playing with AMAF, I found the reference AMAF bots very helpful. Now that I'm playing with UCT, references for UCT would be helpful. 'myCtest-V-0003' is running 50k UCT. Pure random playouts guided by a UCT search with theses parameters: # playouts before expanding = 50 node-score = win_ratio + 0.5 * sqrt(log(N)/n); I will start it under the nam 'myCtest-50k-UCT' later today running 24/7. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Standard references on CGOS
On 10/29/07, Christoph Birk [EMAIL PROTECTED] wrote: I'm not sure if my engine will support 50k simulations without running out of time in long games. Is it possible to do 10k? no problem. I will start 'myCtest-10k-UCT' later today. Christoph How does this compare to myCtest-10k that previously ran on CGOS? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Standard references on CGOS
On Mon, 29 Oct 2007, Jason House wrote: no problem. I will start 'myCtest-10k-UCT' later today. How does this compare to myCtest-10k that previously ran on CGOS? myCtest-10k: 1 random playouts (1050 ELO) myCtest-10k-UCT: 1 random playouts guided by a UCT search (1350 ELO) * nodes are expanded after 50 runs through them * UCT_score = win_ratio + 0.5 * sqrt(log(N)/n) Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] BOINC
G'day guys, I'm involved in the development of a very powerful and flexible grid software, which we plan to release in January. It is all java based. http://www-nereus.physics.ox.ac.uk/ (bear in mind you can't download it yet and the website is out of date) One of the things I'd like to do on it, once it is finished, is some kind of attack on Go. I've messed around trying to genetically generate algorithms to play go. However this has had to go on the back burner for the moment. The brief attempt I made had no way of storing data between games (I ran out of time) and the best algorithm it came up with was a purely random algorithm... :-) our group is also the one that is doing JPC - http://www-jpc.physics.ox.ac.uk/ I'd love to hear about anyone else distributed attacks on Go. cheers, Ian Hi, Funny: I was just getting ready to dust off my GA again. It sounds like you are looking for an interesting application that will motivate people to volunteer computer time and draw some notice to your project. If it were me, I would want to make a distributed version of a Monte-Carlo/UCT go engine and have it play some exhibition games against strong human players. The social engineering qualities are good. I think?there would be a fair amount of (short term) interest from people who could have their computer helping in such a contest. And it would be easy to give them some feedback about how the game was going and what their computer was doing. Play-by-mail, one turn a day, might be acceptable. Tournament time limits would be much more dramatic. milestone 1: All network-nodes compute pure Monte-Carlo (no search tree) scores for the possible moves, the scores are combined centrally to pick the move. It's easy, it will wring out the system, and the bandwidth is low. The playing performance will always be poor because this algorithm doesn't scale well. milestone 2: Each network-node builds its own tree using UCT, but information is only combined at the root. This version will play much better because each node is smarter. The bandwidth will be higher. I can only guess at the scaling behavior, but this milestone might be the 80% solution. milestone 3: Information from the search-nodes is shared between network-nodes, but only for search-nodes close to the root of the tree. Sounds innocent enough. You just limit the shared nodes to the first couple of plys. But it's a trap that will suck you in: best scaling behavior requires too much communication-but what if you made each Monte-Carlo simulation smarter...? ??? I'm just throwing the idea out there. I expect and invite others on the list to point out its flaws. - Dave Hillis . Check Out the new free AIM(R) Mail -- Unlimited storage and industry-leading spam and email virus protection. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] BOINC
milestone 2: Each network-node builds its own tree using UCT, but information is only combined at the root. This version will play much better because each node is smarter. The bandwidth will be higher. I can only guess at the scaling behavior, but this milestone might be the 80% solution. I am currently working at this level. I have a (small) network of some 10 CPUs for tests. Once it is debugged I was hoping to borrow someone's (David?) cluster. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Standard references on CGOS
On Oct 29, 2007, at 10:35 AM, Jason House wrote: Thanks! I'm not sure if my engine will support 50k simulations without running out of time in long games. Is it possible to do 10k? no problem. I will start 'myCtest-10k-UCT' later today. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] BOINC
On 10/29/07, [EMAIL PROTECTED] [EMAIL PROTECTED] wrote: milestone 1: All network-nodes compute pure Monte-Carlo (no search tree) scores for the possible moves, the scores are combined centrally to pick the move. It's easy, it will wring out the system, and the bandwidth is low. The playing performance will always be poor because this algorithm doesn't scale well. It scales, reasonably, but there's a maximum total work to do before any extra becomes useless. milestone 2: Each network-node builds its own tree using UCT, but information is only combined at the root. This version will play much better because each node is smarter. The bandwidth will be higher. I can only guess at the scaling behavior, but this milestone might be the 80% solution. This is what I'm currently aiming for with my bot Currently limited to a single machine since I'm not using a fancier API. milestone 3: Information from the search-nodes is shared between network-nodes, but only for search-nodes close to the root of the tree. Sounds innocent enough. You just limit the shared nodes to the first couple of plys. But it's a trap that will suck you in: best scaling behavior requires too much communication-but what if you made each Monte-Carlo simulation smarter...? I'd probably suggest something making the network nodes mirror the search tree. As results from children get aggregated, the parent node can repartition what fraction of its resources to dedicate to each subtree. As long as the number of children per node are bound, this should give reasonable performance that scales. I'm just throwing the idea out there. I expect and invite others on the list to point out its flaws. - Dave Hillis ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] BOINC
On 10/29/07, steve uurtamo [EMAIL PROTECTED] wrote: As results from children get aggregated, the parent node can repartition what fraction of its resources to dedicate to each subtree. um, doesn't this mean sending out messages to every child for every repartitioning? I was thinking of something along the lines of Hey child, you had control of 10 network nodes, please return 2 to me followed by Hey other child, here's two new nodes to use. As each controlling node performs the reallocation, only the children that are affected will have to react. Reallocation of shallow trees should work relatively well. Reallocation of deep trees should be rarer. I don't know what kind of limitations on reallocation frequency might be needed. While a network node may get detached for a while, a simple way to keep work going is to have it keep processing its old task and give one final update before switching to the new controller. I haven't done any formal analysis, but it seems like a reasonable strategy. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On 10/29/07, Christoph Birk [EMAIL PROTECTED] wrote: On Mon, 29 Oct 2007, Jacques Basaldúa wrote: This can also be done by the programmers. E.g. If CrazyStone is too strong, Rèmi can introduce a CrazyStoneH3 which passes 3 times at the beginning. But not at the first move, to avoid smart tricks. If CrazyStoneH3 is given white and plays: 2. whatever 4. pass 6. pass 8. pass. Black cannot pass and win the game because of komi. I like this idea. This actually might work even without a server change. The self-handicapped program can even play black and pass on the 2nd,3rd,... move. White still cannot win by passing since there are live black stones on the board. I don't think black can ever pass in any simple manner to give handicap. After playing one stone, the remaining empty territory *should* look like nobody's territory. The score would then be black stones vs white stones + komi. I'd argue that the right is always greater than the left until territory begins to form. Similarly, white can safely pass right out of the gate since a black pass would make it win. White can't, however, pass more than 7 times with a komi of 7.5. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
It would be easy to change the cgos3.tcl script to enable self-handicap in this way. I would make this change if crazy-stone or mogo would agree to put up a copy. - Don Christoph Birk wrote: On Mon, 29 Oct 2007, Jacques Basaldúa wrote: This can also be done by the programmers. E.g. If CrazyStone is too strong, Rèmi can introduce a CrazyStoneH3 which passes 3 times at the beginning. But not at the first move, to avoid smart tricks. If CrazyStoneH3 is given white and plays: 2. whatever 4. pass 6. pass 8. pass. Black cannot pass and win the game because of komi. I like this idea. This actually might work even without a server change. The self-handicapped program can even play black and pass on the 2nd,3rd,... move. White still cannot win by passing since there are live black stones on the board. It is not exactly like real handicaps but it would be interesting to see. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Standard references on CGOS
On Mon, 29 Oct 2007, Christoph Birk wrote: myCtest-10k-UCT: 1 random playouts guided by a UCT search (1350 ELO) * nodes are expanded after 50 runs through them * UCT_score = win_ratio + 0.5 * sqrt(log(N)/n) I added variants with different expansion thresholds 'myCtest-V-0006': 25 'myCtest-V-0007': 10 Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] BOINC
As results from children get aggregated, the parent node can repartition what fraction of its resources to dedicate to each subtree. um, doesn't this mean sending out messages to every child for every repartitioning? s. __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On Mon, 29 Oct 2007, Jacques Basaldúa wrote: This can also be done by the programmers. E.g. If CrazyStone is too strong, Rèmi can introduce a CrazyStoneH3 which passes 3 times at the beginning. But not at the first move, to avoid smart tricks. If CrazyStoneH3 is given white and plays: 2. whatever 4. pass 6. pass 8. pass. Black cannot pass and win the game because of komi. I like this idea. This actually might work even without a server change. The self-handicapped program can even play black and pass on the 2nd,3rd,... move. White still cannot win by passing since there are live black stones on the board. It is not exactly like real handicaps but it would be interesting to see. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On Mon, 29 Oct 2007, steve uurtamo wrote: there's not really much sense in a game 'won' in the first 10 moves. i.e. i mean that it doesn't have much intrinsic meaning. i think it's fair to throw away game results that have this feature to them, then only cooperating programs will have their results counted. I agree. The only change to the server would be to NOT stop games after 2 consecutive passes if the were less than 10 moves played. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
One way to handle handicaps without a server change which could be easily implemented with the client is to to simply make the first N moves random - but it would not resemble a traditional handicap system in any way. Plus the first N moves might end up being pretty good moves so it would be applied unevenly. The problem with any black pass moves is that white immediately passes too and wins on komi points.Unless there are no white stones on the board of course. There is no simple way to fake it with combinations of pass moves with programs cooperating. Probably most programs won't pass on the first 20 moves or so, but we can't count on that behavior because it's incorrect. You should always pass immediately if it wins the game outright. - Don John Tromp wrote: On 10/29/07, Christoph Birk [EMAIL PROTECTED] wrote: On Mon, 29 Oct 2007, Jacques Basaldúa wrote: This can also be done by the programmers. E.g. If CrazyStone is too strong, Rèmi can introduce a CrazyStoneH3 which passes 3 times at the beginning. But not at the first move, to avoid smart tricks. If CrazyStoneH3 is given white and plays: 2. whatever 4. pass 6. pass 8. pass. Black cannot pass and win the game because of komi. But to avoid Black from winning by capturing white's stone and passing, white needs to make sure to play her stone where it has 4 liberties. Even that is not sufficient; white has to play this stone on the 3rd line or higher (exercise for the reader: how cld black take advantage of some 2nd line moves?) regards, -John ___ 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] 19x19 CGOS
why not just ignore game results that took place in fewer than 10 moves? then black can play his handicap stones, white can pass, and everyone's cool. s. - Original Message From: Jason House [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Monday, October 29, 2007 4:28:44 PM Subject: Re: [computer-go] 19x19 CGOS On 10/29/07, Christoph Birk [EMAIL PROTECTED] wrote: On Mon, 29 Oct 2007, Jacques Basaldúa wrote: This can also be done by the programmers. E.g. If CrazyStone is too strong, Rèmi can introduce a CrazyStoneH3 which passes 3 times at the beginning. But not at the first move, to avoid smart tricks. If CrazyStoneH3 is given white and plays: 2. whatever 4. pass 6. pass 8. pass. Black cannot pass and win the game because of komi. I like this idea. This actually might work even without a server change. The self-handicapped program can even play black and pass on the 2nd,3rd,... move. White still cannot win by passing since there are live black stones on the board. I don't think black can ever pass in any simple manner to give handicap. After playing one stone, the remaining empty territory *should* look like nobody's territory. The score would then be black stones vs white stones + komi. I'd argue that the right is always greater than the left until territory begins to form. Similarly, white can safely pass right out of the gate since a black pass would make it win. White can't, however, pass more than 7 times with a komi of 7.5. __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
there's not really much sense in a game 'won' in the first 10 moves. i.e. i mean that it doesn't have much intrinsic meaning. i think it's fair to throw away game results that have this feature to them, then only cooperating programs will have their results counted. s. - Original Message From: Don Dailey [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Monday, October 29, 2007 5:23:46 PM Subject: Re: [computer-go] 19x19 CGOS One way to handle handicaps without a server change which could be easily implemented with the client is to to simply make the first N moves random - but it would not resemble a traditional handicap system in any way. Plus the first N moves might end up being pretty good moves so it would be applied unevenly. The problem with any black pass moves is that white immediately passes too and wins on komi points.Unless there are no white stones on the board of course. There is no simple way to fake it with combinations of pass moves with programs cooperating. Probably most programs won't pass on the first 20 moves or so, but we can't count on that behavior because it's incorrect. You should always pass immediately if it wins the game outright. - Don John Tromp wrote: On 10/29/07, Christoph Birk [EMAIL PROTECTED] wrote: On Mon, 29 Oct 2007, Jacques Basaldúa wrote: This can also be done by the programmers. E.g. If CrazyStone is too strong, Rèmi can introduce a CrazyStoneH3 which passes 3 times at the beginning. But not at the first move, to avoid smart tricks. If CrazyStoneH3 is given white and plays: 2. whatever 4. pass 6. pass 8. pass. Black cannot pass and win the game because of komi. But to avoid Black from winning by capturing white's stone and passing, white needs to make sure to play her stone where it has 4 liberties. Even that is not sufficient; white has to play this stone on the 3rd line or higher (exercise for the reader: how cld black take advantage of some 2nd line moves?) regards, -John ___ 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/ __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On 10/29/07, Christoph Birk [EMAIL PROTECTED] wrote: On Mon, 29 Oct 2007, Jacques Basaldúa wrote: This can also be done by the programmers. E.g. If CrazyStone is too strong, Rèmi can introduce a CrazyStoneH3 which passes 3 times at the beginning. But not at the first move, to avoid smart tricks. If CrazyStoneH3 is given white and plays: 2. whatever 4. pass 6. pass 8. pass. Black cannot pass and win the game because of komi. But to avoid Black from winning by capturing white's stone and passing, white needs to make sure to play her stone where it has 4 liberties. Even that is not sufficient; white has to play this stone on the 3rd line or higher (exercise for the reader: how cld black take advantage of some 2nd line moves?) regards, -John ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] CGOS
It appears as if both CGOS servers crashed ... ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
The whole idea is to not have to change the server. If I'm going to change the server I might as well do handicap the right way. I remember us talking about this before - we went back and forth on how to implement handicap with chinese scoring and CGOS but I don't remember what conclusion I came to. Let's review this: 1. We would still attempt to schedule opponents near equal strength. 2. We would still compute ELO ratings. 3. Some calculation (perhaps a constant such as 100 at first) to equate ELO difference to stone handicaps. 4. At rating time I would make the ELO compensation based on handicap and rate accordingly. For the handicap system, I have been checking around at various systems and the GTP protocol.I think the best way which is likely to cause the least amount of agony among programmers is to have the server just send the appropriate play b commands to set up the position. The GTP says your engine is supposed to accept moves out of order. I would use traditional handicap placement and no compensation (remember that discussion?) - Don Christoph Birk wrote: On Mon, 29 Oct 2007, steve uurtamo wrote: or to simply not include the results of such games, so as not to break the protocol for machines that wanted to have such games take place. What would break? Server - clientB: genmove clientB - Server:PASS server - clientW: play PASS server - clientW: genmove clientW - Server:PASS (W tries to be smart and win) server - clientB: play PASS(the server does NOT stop the game) server - clientB: genmove clientB - Server:d4 server - clientW: play d4 (W should accept that move) server - clientW: genmove (W should generate a move) ... All that happens is that White would have wasted its move. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On Mon, 29 Oct 2007, Don Dailey wrote: The whole idea is to not have to change the server. If I'm going to change the server I might as well do handicap the right way. But this is a trivial change compared to dealing with an ad hoc ELO/handicap conversion. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On Mon, 29 Oct 2007, steve uurtamo wrote: or to simply not include the results of such games, so as not to break the protocol for machines that wanted to have such games take place. What would break? Server - clientB: genmove clientB - Server:PASS server - clientW: play PASS server - clientW: genmove clientW - Server:PASS (W tries to be smart and win) server - clientB: play PASS(the server does NOT stop the game) server - clientB: genmove clientB - Server:d4 server - clientW: play d4 (W should accept that move) server - clientW: genmove (W should generate a move) ... All that happens is that White would have wasted its move. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
or to simply not include the results of such games, so as not to break the protocol for machines that wanted to have such games take place. s. - Original Message From: Christoph Birk [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Monday, October 29, 2007 5:55:52 PM Subject: Re: [computer-go] 19x19 CGOS On Mon, 29 Oct 2007, steve uurtamo wrote: there's not really much sense in a game 'won' in the first 10 moves. i.e. i mean that it doesn't have much intrinsic meaning. i think it's fair to throw away game results that have this feature to them, then only cooperating programs will have their results counted. I agree. The only change to the server would be to NOT stop games after 2 consecutive passes if the were less than 10 moves played. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
So the suggestion is to throw out games that end in less that 20 moves? Or simply to not rate them? Or is it to not consider 2 passes a draw unless 20 moves have been played? Of course it seems silly to have 2 of these programs play each other - which could easily happen. The game might start like this: pass pass pass etc. - Don Christoph Birk wrote: On Mon, 29 Oct 2007, Don Dailey wrote: The whole idea is to not have to change the server. If I'm going to change the server I might as well do handicap the right way. But this is a trivial change compared to dealing with an ad hoc ELO/handicap conversion. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
ah, well, okay then. :) s. - Original Message From: Christoph Birk [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Monday, October 29, 2007 6:24:41 PM Subject: Re: [computer-go] 19x19 CGOS On Mon, 29 Oct 2007, steve uurtamo wrote: or to simply not include the results of such games, so as not to break the protocol for machines that wanted to have such games take place. What would break? Server - clientB: genmove clientB - Server:PASS server - clientW: play PASS server - clientW: genmove clientW - Server:PASS (W tries to be smart and win) server - clientB: play PASS(the server does NOT stop the game) server - clientB: genmove clientB - Server:d4 server - clientW: play d4 (W should accept that move) server - clientW: genmove (W should generate a move) ... All that happens is that White would have wasted its move. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
Of course it seems silly to have 2 of these programs play each other - which could easily happen. The game might start like this: pass pass pass etc. I think it is very unlikely for any program to pass in the early game (my would not :-) And if, there is no harm done, as at some point the 'self-handicapped' program will start to move and the other will respond. They would both pass if they were playing in self-handicap mode. - Don Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On Mon, 29 Oct 2007, Christoph Birk wrote: Of course it seems silly to have 2 of these programs play each other - which could easily happen. The game might start like this: pass pass pass etc. And if, there is no harm done, as at some point the 'self-handicapped' program will start to move and the other will respond. If two 'self-handicapped' programs play each other the game will look like (eg. 2H): pass pass pass pass d4 ... And it will be an even game; exactly what it should be, right? Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On Mon, 29 Oct 2007, Don Dailey wrote: So the suggestion is to throw out games that end in less that 20 moves? No, just have the server not stop games before move-20. Of course it seems silly to have 2 of these programs play each other - which could easily happen. The game might start like this: pass pass pass etc. I think it is very unlikely for any program to pass in the early game (my would not :-) And if, there is no harm done, as at some point the 'self-handicapped' program will start to move and the other will respond. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
My only arugment is that it would look silly - but it would be correct. But I guess passing on the first few moves will always look silly. - Don Christoph Birk wrote: On Mon, 29 Oct 2007, Christoph Birk wrote: Of course it seems silly to have 2 of these programs play each other - which could easily happen. The game might start like this: pass pass pass etc. And if, there is no harm done, as at some point the 'self-handicapped' program will start to move and the other will respond. If two 'self-handicapped' programs play each other the game will look like (eg. 2H): pass pass pass pass d4 ... And it will be an even game; exactly what it should be, right? Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Mogo fixed playouts parameter (was: Crazy Stone on 19x19 CGOS)
How does one configure MoGo to do a fixed number of playouts per move? I saw only time-based command line options. On Oct 7th Sylvain wrote: --nbTotalSimulations 3000 Once you set this option it ignores all other time settings. Darren ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] BOINC
It depends a great deal on timing. Physics Monte Carlo has been running non-stop for months ... a very big computation that is still quite short on statistics. I will try to find out when that should finish. Cheers, David On 29, Oct 2007, at 11:50 AM, Christoph Birk wrote: Once it is debugged I was hoping to borrow someone's (David?) cluster. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/