Author: Robert Hyatt
Date: 17:15:08 06/01/04
Go up one level in this thread
On June 01, 2004 at 19:11:51, Sune Fischer wrote: >On June 01, 2004 at 18:37:33, Robert Hyatt wrote: > >>On June 01, 2004 at 18:31:48, Sune Fischer wrote: >> >>>On June 01, 2004 at 18:16:25, Robert Hyatt wrote: >>> >>>>On June 01, 2004 at 17:55:14, Sune Fischer wrote: >>>> >>>>>On June 01, 2004 at 13:56:37, Robert Hyatt wrote: >>>>> >>>>>>On June 01, 2004 at 12:03:44, Sune Fischer wrote: >>>>>> >>>>>>>On June 01, 2004 at 11:52:50, Robert Hyatt wrote: >>>>>>> >>>>>>>>> >>>>>>>>>As for pondering you obviously can't play with ponder on at a uni-processor, so >>>>>>>>>I don't see how that can come as a surprise. >>>>>>>> >>>>>>>>I do it all the time with no problems whatsoever. So what if each program gets >>>>>>>>1/2 of the processor? >>>>>>> >>>>>>>1/2 cpu, exactly, would be no problem. >>>>>>>But what if one engine decides to "ponder" with 10 threads, or if the threads >>>>>>>don't run at the same priority? >>>>>>> >>>>>>>What if one engine decides to skip pondering for one move, then the other gets >>>>>>>100%. That's double punishment. >>>>>> >>>>>>That's a stupid engine, too. :) >>>>> >>>>>So? >>>>>No reason to punish it twice, that just forces everyone to do stupid hacks to >>>>>keep them at full load. >>>>> >>>>>There are other issues as well, ie. if one engine starts hitting TBs heavily, >>>>>how does that influence cpu load between the programs? >>>>> >>>>>What about trashing the cache? >>>>>Author of engine X has spend many hours fine tuning his memory footprint to fix >>>>>exactly into the 256 kb. Running a second program completely cripples his >>>>>engine, he claims, this was _not_ what it was designed for. >>>>> >>>>>-S. >>>> >>>> >>>>That is why testing on _one_ computer is generally wrong. :) >>> >>>What's wrong with it if you turn pondering off? >>> >>>-S. >> >> >>Perhaps an engine is not well tested in that mode? > >Good argument, but what if I reverse and say some engines might not be well >tested with ponder on? There is no cure for stupidity. If someone tests their engine in sub-optimal configurations, then they could expect problems. I don't see a solution for stupidity. :) > >You can't play a fair tournament that makes everybody happy. actually you can. Play "run what ya brung" and let the chips fall where they may. If someone gets ripped by a learner program, they'll probably be learning by next year. If someone gets ripped by a killer book, they'll probably fix it to avoid getting ripped next year. "lowest common denominator" is not very interesting. > >As for pondering in particular, the protocol specificly has easy and hard >commands to turn pondering on and off. Thus any xboard compliant engine should >know how to play it both ways, if some engines don't then it can simply be >considered a weakness, IMO. Nobody in their right mind considers ponder=off to be "full strength" however. So weakening the engine by doing so is a "caveat emptor" situation. > >>IMHO an engine should be tested "as is". > >What if two engines have conflicting "as is" settings? > Impossible. Play mine "as is" against yours "as is". There can be no conflict there... >Would playing an aggressive learner against a non-learner be an interesting >experiment at all? > No. But if the non-learner stays a non-learner, he can expect the same result over and over until he chooses to fix it. You can either take direct action yourself to fix a problem, or you can manually cripple the opponent to "equalize things". But the latter is not possible at (say) the WCCC, or on the SSDF list. There, discretion is advised and learning is recommended as a key survival tool. >> If you want to twiddle with a new >>"personality" then that is fine. Everyone is doing that but they are making it >>clear that things are far from "normal" by naming the personality they are >>creating, to make it distinct from the default personality. > >They are making it clear, everyone AFAICT remembers to post the tournament >conditions. Not so clear to 99% of the people. If they see "ponder=off" they say "so what?" Ditto for "learn=off". > >But by all means, let the record show that the author of Crafty feels the >program is heavily weakened when using ponder off and learning off. > >There, happy now? :) > >-S. Perhaps. learn=off won't be a problem soon. ponder=off is needed for lots of reasons so it can stick around... :)
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.