Re: [computer-go] May KGS bot tournament: full-sized boards, fast
Reminder - it starts in a few hours (13:00 GMT), five hours after the time of posting this). Nick In message <[EMAIL PROTECTED]>, Nick Wedd <[EMAIL PROTECTED]> writes Registration is now open for this Sunday's bot tournament. It will use 19x19 boards for both divisions, which will be 6-round Swiss. It will start at 15:00 GMT, and take place in the Asian night, European evening, and American morning. Time limits will be 18 minutes each, sudden death, for both divisions. It will end about 19:00 GMT. Registration is as described at http://www.weddslist.com/kgs/how/index.html As last month, when you register you should tell me the processor power (number of processors, processor speed, and any other significant details) of the platform that it will be running on. This is so that the processor power can be stated on my report of the event, making comparisons between programs more meaningful for anyone reading the report. The tournaments themselves are on the KGS site at http://www.gokgs.com/tournInfo.jsp?id=380 and http://www.gokgs.com/tournInfo.jsp?id=381. Nick -- Nick Wedd[EMAIL PROTECTED] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] May KGS bot tournament: full-sized boards, fast
>From the website http://www.gokgs.com/tournInfo.jsp?id=380 and the fact that it hasn't started, I deduce that it starts at 1500 GMT, or about 40 minutes time. On Sun, 2008-05-04 at 11:00 +0100, Nick Wedd wrote: > Reminder - it starts in a few hours (13:00 GMT), five hours after the > time of posting this). > > Nick > > > In message <[EMAIL PROTECTED]>, Nick Wedd > <[EMAIL PROTECTED]> writes > >Registration is now open for this Sunday's bot tournament. It will use > >19x19 boards for both divisions, which will be 6-round Swiss. It will > >start at 15:00 GMT, and take place in the Asian night, European > >evening, and American morning. Time limits will be 18 minutes each, > >sudden death, for both divisions. It will end about 19:00 GMT. > > > >Registration is as described at > >http://www.weddslist.com/kgs/how/index.html > > > >As last month, when you register you should tell me the processor power > >(number of processors, processor speed, and any other significant > >details) of the platform that it will be running on. This is so that > >the processor power can be stated on my report of the event, making > >comparisons between programs more meaningful for anyone reading the > >report. > > > >The tournaments themselves are on the KGS site at > >http://www.gokgs.com/tournInfo.jsp?id=380 and > >http://www.gokgs.com/tournInfo.jsp?id=381. > > > >Nick > ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] May KGS bot tournament: full-sized boards, fast
On Sun, 2008-05-04 at 14:22 +0100, Tom wrote: > >From the website http://www.gokgs.com/tournInfo.jsp?id=380 and the fact > that it hasn't started, I deduce that it starts at 1500 GMT, or about 40 > minutes time. I think you mean 1h40m? > > On Sun, 2008-05-04 at 11:00 +0100, Nick Wedd wrote: > > Reminder - it starts in a few hours (13:00 GMT), five hours after the > > time of posting this). > > > > Nick > > > > > > In message <[EMAIL PROTECTED]>, Nick Wedd > > <[EMAIL PROTECTED]> writes > > >Registration is now open for this Sunday's bot tournament. It will use > > >19x19 boards for both divisions, which will be 6-round Swiss. It will > > >start at 15:00 GMT, and take place in the Asian night, European > > >evening, and American morning. Time limits will be 18 minutes each, > > >sudden death, for both divisions. It will end about 19:00 GMT. > > > > > >Registration is as described at > > >http://www.weddslist.com/kgs/how/index.html > > > > > >As last month, when you register you should tell me the processor power > > >(number of processors, processor speed, and any other significant > > >details) of the platform that it will be running on. This is so that > > >the processor power can be stated on my report of the event, making > > >comparisons between programs more meaningful for anyone reading the > > >report. > > > > > >The tournaments themselves are on the KGS site at > > >http://www.gokgs.com/tournInfo.jsp?id=380 and > > >http://www.gokgs.com/tournInfo.jsp?id=381. > > > > > >Nick > > > > ___ > 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] May KGS bot tournament: full-sized boards, fast
In message <[EMAIL PROTECTED]>, Jason House <[EMAIL PROTECTED]> writes On Sun, 2008-05-04 at 14:22 +0100, Tom wrote: >From the website http://www.gokgs.com/tournInfo.jsp?id=380 and the fact that it hasn't started, I deduce that it starts at 1500 GMT, or about 40 minutes time. I think you mean 1h40m? I also find summer-time (daylight saving) really confusing. Nick -- Nick Wedd[EMAIL PROTECTED] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Mogo scalability
The scalability study showed that mogo gains almost 100 rating points per doubling of performance. But on CGOS, there is mogo 1 CPU and mogo 4 CPU. I would expect the 4 CPU mogo to be 150 rating points or more higher than 1 CPU. But it is actually only 40 points higher. I guess this is because in the rating study mogo was mostly playing other mogo versions. A the faster mogo can see everything the weaker mogo sees, and more. But against other programs on CGOS, there will be weaknesses common to both fast and slow mogo, that other programs might exploit. Is the true scalability of mogo (against a variety of programs, or against people) less than the rating study indicates? David ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Mogo scalability
David Fotland wrote: The scalability study showed that mogo gains almost 100 rating points per doubling of performance. But on CGOS, there is mogo 1 CPU and mogo 4 CPU. I would expect the 4 CPU mogo to be 150 rating points or more higher than 1 CPU. But it is actually only 40 points higher. I guess this is because in the rating study mogo was mostly playing other mogo versions. A the faster mogo can see everything the weaker mogo sees, and more. But against other programs on CGOS, there will be weaknesses common to both fast and slow mogo, that other programs might exploit. Is the true scalability of mogo (against a variety of programs, or against people) less than the rating study indicates? It's possible. Most people believe that when the only difference between 2 programs is how deep they look, the less deep version gets victimized fairly badly.I'm undecided about how strong this effect is, although I do believe it's a factor. On CGOS most of the programs, while not identical, are very similar. Does that also have an impact on the rating pool? If that is impacting it too, that would decrease the mogo vs mogo effect. The rating study includes games against Leela too. Next week when I get back from my vacation, I will try voiding games not played between different opponents, and see how it affects the curve.It probably won't answer any questions but it might be a clue. - Don David ___ 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: Mogo scalability
David Fotland: <[EMAIL PROTECTED]>: >The scalability study showed that mogo gains almost 100 rating points per >doubling of performance. > >But on CGOS, there is mogo 1 CPU and mogo 4 CPU. I would expect the 4 CPU >mogo to be 150 rating points or more higher than 1 CPU. But it is actually >only 40 points higher. I'm also wondering this for months. One reason of this discount is that the MoGo bros running on cgos are the _big versions. By my obserbation (they are running on my pcs and both are Q6600/3GHz with different mother boards), mogo_big_4core's perallelism is around 300% (by top command), perhaps due to its heavier uct part (just my guess). >I guess this is because in the rating study mogo was mostly playing other >mogo versions. A the faster mogo can see everything the weaker mogo sees, >and more. I'm not sure. Mogo_big_4core's winning rate against mogo_big_1core is about 60% (1641 / 2743) which corresponds only +70 ELO and current rating difference on cgos is about 40, as you wrote. According to the scaling rule of mc playouts, it should be 160 ELO (log2(3) * 100 = 158.5). >But against other programs on CGOS, there will be weaknesses common to both >fast and slow mogo, that other programs might exploit. > >Is the true scalability of mogo (against a variety of programs, or against >people) less than the rating study indicates? I believe their current ratings on cgos indicate so. Hideki >David > > >___ >computer-go mailing list >computer-go@computer-go.org >http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Mogo scalability
Hideki Kato wrote: David Fotland: <[EMAIL PROTECTED]>: The scalability study showed that mogo gains almost 100 rating points per doubling of performance. But on CGOS, there is mogo 1 CPU and mogo 4 CPU. I would expect the 4 CPU mogo to be 150 rating points or more higher than 1 CPU. But it is actually only 40 points higher. I'm also wondering this for months. One reason of this discount is that the MoGo bros running on cgos are the _big versions. By my obserbation (they are running on my pcs and both are Q6600/3GHz with different mother boards), mogo_big_4core's perallelism is around 300% (by top command), perhaps due to its heavier uct part (just my guess). I guess this is because in the rating study mogo was mostly playing other mogo versions. A the faster mogo can see everything the weaker mogo sees, and more. I'm not sure. Mogo_big_4core's winning rate against mogo_big_1core is about 60% (1641 / 2743) which corresponds only +70 ELO and current rating difference on cgos is about 40, as you wrote. According to the scaling rule of mc playouts, it should be 160 ELO (log2(3) * 100 = 158.5). But against other programs on CGOS, there will be weaknesses common to both fast and slow mogo, that other programs might exploit. Is the true scalability of mogo (against a variety of programs, or against people) less than the rating study indicates? I believe their current ratings on cgos indicate so. You should look on the All Time Ratings list to get a more correct picture. I will try to update it today but we still have 551 games on the all time list for mogo-big-4 core: It showsmogo-big-4core as 2563 +35 -35 and mogo-big-1core as 2480 +24 -24 Also, don't forget that 4 cores are not as efficient as 1 core, so we shouldn't expect this to be as good as the 1 core version running 4x longer. I would expect it to be much better than 2x though - perhaps the mogo team can comment on that. - Don Hideki David ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ 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: Mogo scalability
I'm trying to update the all-time ratings list for CGOS but I have a 1 gig limit on the server and CGOS has reached it.So I'm going to prepare this off-line. So it should be up later. I'm not sure if CGOS will continue to operate during the week, but I don't have time to clean it up and I will be gone for the week. Hopefully, it will continue to run until I get back.I may be able to do some minor cleanups in order to help it continue. - Don Don Dailey wrote: Hideki Kato wrote: David Fotland: <[EMAIL PROTECTED]>: The scalability study showed that mogo gains almost 100 rating points per doubling of performance. But on CGOS, there is mogo 1 CPU and mogo 4 CPU. I would expect the 4 CPU mogo to be 150 rating points or more higher than 1 CPU. But it is actually only 40 points higher. I'm also wondering this for months. One reason of this discount is that the MoGo bros running on cgos are the _big versions. By my obserbation (they are running on my pcs and both are Q6600/3GHz with different mother boards), mogo_big_4core's perallelism is around 300% (by top command), perhaps due to its heavier uct part (just my guess). I guess this is because in the rating study mogo was mostly playing other mogo versions. A the faster mogo can see everything the weaker mogo sees, and more. I'm not sure. Mogo_big_4core's winning rate against mogo_big_1core is about 60% (1641 / 2743) which corresponds only +70 ELO and current rating difference on cgos is about 40, as you wrote. According to the scaling rule of mc playouts, it should be 160 ELO (log2(3) * 100 = 158.5). But against other programs on CGOS, there will be weaknesses common to both fast and slow mogo, that other programs might exploit. Is the true scalability of mogo (against a variety of programs, or against people) less than the rating study indicates? I believe their current ratings on cgos indicate so. You should look on the All Time Ratings list to get a more correct picture. I will try to update it today but we still have 551 games on the all time list for mogo-big-4 core: It showsmogo-big-4core as 2563 +35 -35 and mogo-big-1core as 2480 +24 -24 Also, don't forget that 4 cores are not as efficient as 1 core, so we shouldn't expect this to be as good as the 1 core version running 4x longer. I would expect it to be much better than 2x though - perhaps the mogo team can comment on that. - Don Hideki David ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ 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: Mogo scalability
One reason of this discount is that the MoGo bros running on cgos are the _big versions. By my obserbation (they are running on my pcs and both are Q6600/3GHz with different mother boards), mogo_big_4core's perallelism is around 300% (by top command), perhaps due to its heavier uct part (just my guess). The top might overestimate the parallelization, as it depends the results of top depend on how the locks are implemented. I don't know how was this part at the time of the release. With 4 cores, the current version of mogo is close to a speed-up four. I believed that the released version was quite similar from that point of view. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Mogo scalability
Is the true scalability of mogo (against a variety of programs, or against people) less than the rating study indicates? Unfortunately, I think that the answer is yes (very very). In particular the tuning has been performed on small games, and I guess the new very fast (parallel) versions of mogo would be better with - more exploration; - more diversity in the MC part. In particular, some of the last improvements of mogo were negligible (if positive) for small numbers of simulations, but are incredibly efficient for large numbers of simulations. It's not very surprising: if you have more computational power, you have to explore more, and you can afford a bigger diversity in the MC-analysis. The parallel version has a huge speed-up, which has also an effect on the ranking vs humans, but the effect is much lower. What we need now is probably much more a qualitative change in the MC, whenever this qualitative change is computationally expensive. Thanks to clusters, we can afford expensive Monte-Carlo. In my humble opinion, if in the same code we have: - the mogo parallelization (30 lines of MPI, 30 lines of C++, not more...) - very good MC simulations (e.g. the ones with the best results for the "tsumego") - a big cluster then you have a fantastic Go program. Olivier ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] The effect of the UCT-constant on Valkyria
David Fotland wrote: > So I'm curious then. With simple UCT (no rave, no priors, no progressive > widening), many people said the best constant was about 0.45. What are the > new concepts that let you avoid the constant? Whatever concepts are used it must indirectly be a question of improved move ordering. The better the move ordering, the smaller the need to do exploration. > Is it RAVE, because the information gathered during the search lets you > focus the search accurately without the UCT term? Many people have said > that RAVE has no benefit for them. > > Do most of the strongest programs use RAVE? I think from Crazystone's > papers, that it does not use RAVE. Gnugomc does not use rave. I've never had success with RAVE but I might make a new attempt for GNU Go some time. > Is it the prior values from go knowledge, like opening books, reading > tactics before the search etc? Do all of the top programs have opening > books now? I know mogo does. The MonteGNU account on CGOS (9x9) has a self-learnt opening book with currently slightly more than 16000 moves. Over the last 1000 games it has played on average 4 moves (own moves that is, opponent moves not counted) from the book. The record is 22 moves from book. > Do most of the top programs read tactics before the search? I know Aya > does. GNU Go in Monte Carlo mode reads lots of tactics before the MC search. But it doesn't use the tactics for the MC search. :-/ > Does it matter how prior values are used to guide the search? I think mogo > uses prior knowledge to initialize the RAVE values. Do other programs > include it some other way, by initializing the FPU value, or by initializing > the UCT visits and confidence, or some extra, "prior" term in the equation? > > Are there other techniques (not RAVE) that people are using to get > information from the search to guide the move ordering? I think crazystone > estimates ownership of each point and uses it to set prior values in some > way. GNU Go uses a global move ordering shared by all nodes in the tree and initialized from the results of the normal move generation. /Gunnar ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Mogo scalability
Ok, I made some more space on the server - we should be good for a few more weeks. - Don Don Dailey wrote: I'm trying to update the all-time ratings list for CGOS but I have a 1 gig limit on the server and CGOS has reached it.So I'm going to prepare this off-line. So it should be up later. I'm not sure if CGOS will continue to operate during the week, but I don't have time to clean it up and I will be gone for the week. Hopefully, it will continue to run until I get back.I may be able to do some minor cleanups in order to help it continue. - Don Don Dailey wrote: Hideki Kato wrote: David Fotland: <[EMAIL PROTECTED]>: The scalability study showed that mogo gains almost 100 rating points per doubling of performance. But on CGOS, there is mogo 1 CPU and mogo 4 CPU. I would expect the 4 CPU mogo to be 150 rating points or more higher than 1 CPU. But it is actually only 40 points higher. I'm also wondering this for months. One reason of this discount is that the MoGo bros running on cgos are the _big versions. By my obserbation (they are running on my pcs and both are Q6600/3GHz with different mother boards), mogo_big_4core's perallelism is around 300% (by top command), perhaps due to its heavier uct part (just my guess). I guess this is because in the rating study mogo was mostly playing other mogo versions. A the faster mogo can see everything the weaker mogo sees, and more. I'm not sure. Mogo_big_4core's winning rate against mogo_big_1core is about 60% (1641 / 2743) which corresponds only +70 ELO and current rating difference on cgos is about 40, as you wrote. According to the scaling rule of mc playouts, it should be 160 ELO (log2(3) * 100 = 158.5). But against other programs on CGOS, there will be weaknesses common to both fast and slow mogo, that other programs might exploit. Is the true scalability of mogo (against a variety of programs, or against people) less than the rating study indicates? I believe their current ratings on cgos indicate so. You should look on the All Time Ratings list to get a more correct picture. I will try to update it today but we still have 551 games on the all time list for mogo-big-4 core: It showsmogo-big-4core as 2563 +35 -35 and mogo-big-1core as 2480 +24 -24 Also, don't forget that 4 cores are not as efficient as 1 core, so we shouldn't expect this to be as good as the 1 core version running 4x longer. I would expect it to be much better than 2x though - perhaps the mogo team can comment on that. - Don Hideki David ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ 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/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] updates
Ok, I've done the following: 1. free up space on CGOS - should be good for a few weeks/months 2. updated allTime list. 3. updated archives with march and april games. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Mogo scalability
Olivier Teytaud: <[EMAIL PROTECTED]>: >> One reason of this discount is that the MoGo bros running on cgos are >> the _big versions. By my obserbation (they are running on my pcs and >> both are Q6600/3GHz with different mother boards), mogo_big_4core's >> perallelism is around 300% (by top command), perhaps due to its >> heavier uct part (just my guess). > >The top might overestimate the parallelization, as it depends the results >of top depend on how the locks are implemented. I don't know how was this >part at the time of the release. > >With 4 cores, the >current version of mogo is close to a speed-up four. I believed that the >released version was quite similar from that point of view. I recently checked so-called public relase 3 (essentially the same as the initial release, right?) as well and its 4 core version shows over 330% in top command. So, I believe the big version doesn't have the same parallelism. Does MoGo use simple mutex or some smarter locks? I'd like to run newer version on cgos anyway :) Hideki >___ >computer-go mailing list >computer-go@computer-go.org >http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Mogo scalability
I recently checked so-called public relase 3 (essentially the same as the initial release, right?) as well and its 4 core version shows over 330% in top command. So, I believe the big version doesn't have the same parallelism. This % is quite boring and complicated :-) The % decreases with time. Longer games, lower %. It also strongly depends on whether it is compiled with float or doubles (due to concurrency in memory access for AMAF values, strongly reduced with float), and we have several improvements (not in the release) which improve this % of a few % each. And I have no idea of what is in the release versions. I should find time for a release with everything fine or I should make this code GPL :-) Olivier ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
