Author: Graham Laight
Date: 01:33:31 10/15/04
Go up one level in this thread
On October 14, 2004 at 16:13:48, Sandro Necchi wrote: >On October 13, 2004 at 15:51:02, Robert Hyatt wrote: > >>On October 13, 2004 at 11:11:47, Graham Laight wrote: >> >>>On October 13, 2004 at 10:55:20, Michael Yee wrote: >>> >>>>On October 13, 2004 at 10:42:08, Graham Laight wrote: >>>>>On October 13, 2004 at 10:33:30, Michael Yee wrote: >>>> >>>>>>have 1 "bad" (or underperforming) tournament out of 20, i.e., with low >>>>>>probability. But the rare event *will* (or could) happen at some point. >>>>> >>>>>Please see the answer I gave in >>>>>http://www.talkchess.com/forums/1/message.html?391399 >>>>> >>>>>-g >>>>> >>>>>>Michael >>>> >>>>No offense, but I don't think I understand what your point is. Your simulation >>> >>>My points (made throughout the thread - not just in the previous post in this >>>branch of the thread) are: >>> >>>1. Given the Hydra and Fritz results, the Junior result is unexpectedly low >> >> >>What would you do if you took four humans, and four copies of fritz or hydra and >>played the _same_ event again? And what would you say if one of the copies of >>Fritz produced 3 draws and a loss? "It did poorly?" Or "unexpected random >>chance?" >> >>It is almost a certainty that all 4 copies would _not_ produce the same >>result... >> >Bob, > >you are correct but we are "old fashion". I hope you do not get upset; I mean we >deeply analyze things and try to give explanations to things. > >The "modern" way is to give very quick estimantions/evaluations based on scores >on limited amount of games and or events. > >This is why one program can go down from top to lowest level and the other way >around so easily. This happens on sports too. > >Of course not everybody think this way, but more and more people seems to do it >probably because to understand things need a lot of specific knowledge and >experience which require a lot of time and people do not have or are not willing >to invest the needed time. So it is easier to make fast comments; not so easy to >make deep analysis... In this case, you're not correct, I'm afraid. Unless you do the maths, you could easily fail to realise that a small sample is actually giving you more information than a large sample. For an example of this, please see http://www.talkchess.com/forums/1/message.html?391534 -g >Sandro >> >>> >>>2. The Hydra and Fritz results taken together are an indication of great >>>strength >>> >>>>(or even just a basic probability calculation) shows that a "low" score for an >>>>engine that is assumed to have a certain strength is a rare event. I don't >>>>disagree with that. I'm just confused about what conclusions you're trying to >>>>draw from witnessing a rare event. >>>> >>>>Here's how I might put bilbao in perspective: Suppose we are looking at this >>>>tournament as simply one in a stream of tournaments, and we consider updating >>>>junior's rating (i.e., strength estimate) in a bayesian way. Then junior's past >>>>results would weigh much more heavily than this one new result and the rating >>>>wouldn't change by much. >>>> >>>>What would I conclude? Probably that junior had a (slightly) rare result. >>> >>>The Junior result is probably not too far away from what you'd expect. Perhaps I >>>have been looking in astonishment at the wrong place. Perhaps the astonishment >>>should be focused upon the 7/8 score which Hydra and Fritz achieved - which is >>>highly improbable (I calculated 1/160 in another post in this thread) unless >>>these two computers are substantially better than the opponents that they faced >>>at Bilbao. >>> >>>-g >>> >>>>Michael
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.