Don Dailey: <banlktikkhjtgs9wq0g8wgopxtosk6xf...@mail.gmail.com>: >On Fri, Jun 17, 2011 at 9:23 PM, Hideki Kato <hideki_ka...@ybb.ne.jp> wrote: > >> Jean-loup Gailly: <banlktinl-bdioawtjhjc14xekx02b6g...@mail.gmail.com>: >> >> One advantage however to involving a lot of people and their computers >> is >> >that >> >> I COULD do the study out to enormous numbers of playouts, given enough >> >help. >> > >> >Going beyond 8M playouts/move would only be useful, in my opinion, if we >> can >> >test >> >stronger programs. It's not as bad as self play, but you don't learn much >> >from beating >> >a weaker opponent more than 95% of the time. I would love to be able to >> play >> >thousands >> >of games against Zen, which is currently about 2 stones stronger than >> Pachi. >> >> Unfortunatelly, copying the executable binary of Zen to other machines >> is not allowed by the contract... >> > >So you simply specify which machines have Zen installed. In my >configuration file, assuming we use my tester you basically configure the >machine like this:
Unfortunatelly again, all of my pcs are busy for my research work/experiments in this year or more (perhaps, until I'll finish my thesis). #Ah, a dual core Athlon-64 pc, being used for the anchors on CGOS, might be usable (I have to ask Yamato and the sponsor before joinning). The pc has only 1 MB of RAM, though. One question: What's the object of this project? Just to measure the scalability of modern MCTS programs, or to prove (or find some support for) your word "It converges at perfect play"? If the latter, I'd like to say that's almost useless because benchmarks on 19x19 board cannot prove that. Even Zen, current strongest, can explore a small part of the state space of the game of Go. Even for 9x9 board, Zen is not enough, I believe. So, first of all, we'd better try solved 5x5 board to test if the mix of current bots can reach the perfect play. Hideki >player = Zen-4k >desc = Zen at 4 playouts per move >invoke = ssh foobar.com /home/joe/bin/zen -playouts 4000 >level = 4000 >Hash = 128 > >Of course this might look a lot different for Go, and I have no idea how >zen is configured for level, etc. > >At the top of the configuration file you specify cpus because the tester is >designed for multi-core and you might want to limit the number of >simulatanteous games to 4 on a quad core, etc. But for this application >that is not so relevant, so we would want to provide for a way to limit the >number of running instances of ANY program on the host machine. For >instance we might not want to be running more than 4 programs at any given >time on the foobar.com machine. > >I think it might be easier to manually set these matches up with twogmp or >something like that too. That would be easy. > >Don > > > > > > > name > > > >> >> Hideki >> >> >Also it's not enough to double playouts you must also double memory usage, >> >and you >> >may have trouble finding enough machines with terabytes of ram. >> > >> >> I would probably have multiple instance of EACH program, not a single >> >fixed >> >> opponent for one program but that does require a lot more games. >> > >> >Same problem. Beating multiple weaker programs won't tell you much. We >> need >> >multiple stronger programs. >> > >> >Jean-loup >> >---- inline file >> >_______________________________________________ >> >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 >> >---- inline file >_______________________________________________ >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
