Author: Robert Hyatt
Date: 17:45:59 05/20/00
Go up one level in this thread
On May 20, 2000 at 16:38:06, Fernando Villegas wrote: >On May 19, 2000 at 21:12:43, Robert Hyatt wrote: > >>On May 19, 2000 at 14:09:57, Fernando Villegas wrote: >> >>>On May 19, 2000 at 12:33:26, Enrique Irazoqui wrote: >>> >>>>On May 19, 2000 at 12:27:41, Ed Schröder wrote: >>>> >>>>>On May 19, 2000 at 12:13:44, Enrique Irazoqui wrote: >>>>> >>>>>>On May 19, 2000 at 12:04:05, Chris Carson wrote: >>>>>> >>>>>>>The only loss due to HW was the Reb-Cen Hoffman. >>>>>>> >>>>>>>How do I justify removing games for Shredder and >>>>>>>Junior due to operator error and IP failure (which >>>>>>>I think should be removed from the TPR, but listed >>>>>>>in the results) but not the HW failure for Rebel? >>>>>>>It may be a won position for Hoffman, it was right >>>>>>>to declare him the winner with prize money, but having >>>>>>>a won position and winning are two different things. >>>>>>>I think this game should be left out of the TPR >>>>>>>calculation. Just 2 cents. :) >>>>>> >>>>>>I always counted it because Rebel was indeed lost. Ed agrees, by the way. But >>>>>>since you have a different opinion, why don't you make 2 lists, one with ant the >>>>>>other without this game? >>>>> >>>>>2 points... a) I think Rebel would have lost the Hoffman game by good >>>>>play of Hoffman but how can you be sure of that? >>>> >>>>That's Uri's point. I am personally sure enough, but even in case of uncertainty >>>>the game should count, or else no games played on the Kyro thing should count. I >>>>mean, if we count points scored by Rebel on that particular overclocked (faster) >>>>machine, we should also count the games lost because of failure due to >>>>overclocking. >>> >>> >>>There is not logical simetry between winning with some device and losing because >>>a failure of it. The normal thing for a device is good operation, the >>>malfunctioning is an accident. Only with the average good functioning you have a >>>measure of what a progran can do, not with the accident. I insist that a defeat >>>because of the accident must be taken into account in a tornament, just bad luck >>>as the flu of a human, but if we are doing an experiment, and that is what Chris >>>is doing, it has not sense to include not operatives episodes of the experiment. >>>Fernando >>> >>> >> >> >>That doesn't make sense. The "strength" of a computer is a measure of how it >>does against human competition. If it loses 1 game out of every 4 due to >>hardware failures, then those games belong in the rating formula. Because >>hardware failures are a fact of life. A human gets sick in the middle of a >>game, he loses. If he doesn't he plays on. Some people get sick more >>frequently than others. This should be reflected in their rating since there >>is a definite chance they will get sick when they play me... >> >>You have to take the good with the bad, or else it is just "cherry-picking". >> >>If you cherry-pick, you can produce any rating you want... > > >Hi Bob: >You would be right IF we already were living in another stage of history of >chess computer where strenght compared with GM or whatever was not anymore an >incognite, a concern; then the issue just would be to play tournaments and win >or lose due even to failures, BUT in this stage of the history the entire point >is this, duifferent one: which is program-GM relation of forces. That is what >all of us are asking. In other words, is a non solved issue and as such requires >special measures of evaluation, an experiment, a clear cut conditioning of the >game or experiment, not just the real thing. Specifically the incognite is how >good programs -not the full unit program-hardware- at least so I understand the >issue- are if playing at his best. Could they in that BEST condition be so good >or not? And so and so. And as much that the question is precisely that, then >just to take into acount games where the program did not suffer from hardware >has plenty of sense. I hope I make me clear. >Cheers >Fernando >> >> I understand.. but it is simply the wrong way to analyze program strength. IE take the same program, one on a stock PIII/800, the other on a super-cooled, overclocked PIII/1600. Obviously the overclocked CPU will be stronger. But it will likely fail more frequently. Isn't that part of the risk in using faster hardware? And if it loses one game out of 10 by crashing, doesn't that 1 game count since it _did_ lose it. And it will lose similar games in the future... USCF ratings simply count all games played... you crash you lose... as that is part of a "computer program". >> >> >>> >>>> >>>>Enrique >>>> >>>>> and b) the hardware >>>>>collapsed, better show up with good hardware. The machine had a flu so >>>>>can humans have, so always count the game in case of hardware failures. >>>>> >>>>>Ed >>>>> >>>>>>Enrique >>>>>> >>>>>>>Best Regards, >>>>>>>Chris Carson
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.