Author: Robert Hyatt
Date: 06:48:54 05/02/00
Go up one level in this thread
On May 02, 2000 at 02:29:39, Chessfun wrote: >On May 02, 2000 at 01:50:33, blass uri wrote: > >>On May 01, 2000 at 21:50:12, Robert Hyatt wrote: >> >>>On May 01, 2000 at 20:55:15, Christophe Theron wrote: >>> >>>>On April 30, 2000 at 22:58:11, Robert Hyatt wrote: >>>> >>>>>On April 30, 2000 at 22:38:10, Christophe Theron wrote: >>>>> >>>>>>On April 30, 2000 at 19:04:19, Chris Taylor wrote: >>>>>> >>>>>>>And now with some text in it? >>>>>>> >>>>>>>I restart both computers after each match? A re-boot. Do not ask me why? >>>>>>>Mayhap, to clear settings?? >>>>>>> >>>>>>>Chris Taylor >>>>>> >>>>>> >>>>>>But is it enough? I have heard that Crafty learns by position. That is it is >>>>>>able to remember the positions of a lost game and to avoid playing again the >>>>>>moves that go into these positions. >>>>>> >>>>>>In this case, replaying a match with the same time controls gives an advantage >>>>>>to Crafty. >>>>>> >>>>>>I don't know if Fritz has a similar system. >>>>>> >>>>>>In the case of learning by position, it should be necessary to erase the "learn" >>>>>>file when one wants to reproduce a match... >>>>>> >>>>>> >>>>>> Christophe >>>>> >>>>> >>>>>Why is that so. When I play multiple matches against a human, no one asks to >>>>>hypotize me to clear my memory. Do we now suddenly declare that _all_ learning >>>>>functions are 'NFG' and should not be allowed? >>>> >>>> >>>> >>>>The purpose of the experiment is to see what happens when the engines have to >>>>play from given positions that they have never seen before. >>>> >>>>We try to reproduce what happens in a match. >>>> >>>>You could criticize the small number of games for example, but not the fact that >>>>we want to see "fresh" engines in the match. >>>> >>>> >>>> >>>> Christophe >>> >>> >>>I criticize the fact that the concept of a "fresh engine" is nonsense. The >>>match already starts off in a position not of the program's choosing, since >>>no opening book or book learning is available, even though it is an _integral_ >>>part of the engine. >>> >>>Here is what I have heard, so far: >>> >>>1. Can't use crafty's SMP search. That isn't fair. >>> >>>2. Can't use crafty's position learning, that isn't fair. >>> >>>3. Can't use crafty's book learning, that isn't fair. >>> >>>maybe it is time we define "fair". IE is not fair something I do that others >>>don't do? Or is fair only the things that others do? The Nunn test is simply >>>'flawed'. Interesting, yes. But flawed. >> >>You can call the game that the programs play nunnchess instead of chess but >>it does not change the fact that the game is fair because both sides start from >>the same position. >> >>I agree that the result in the chess game may be different but it does not >>change the fact that both programs play a fair game. >> >>Uri > > >The Nunn 1 positions are not that odd, most programs have all >of them within there opening books. I agree to me it is as fair >as it can get. > >We all see enough posts about losing while in book, here we have >a given set of positions. This is similar to playing openings on >computers and reversing colors for the second game, which also >IMHO is far better for measuring strength than the current method. > >Thanks. So you think that there are only about 10 reasonable positions that can be reached by move 10-15? Or are there more than a few trillion? And picking 10 positions means you for sure get 10 positions that both programs will actually reach using their 'native' book? I don't think so...
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.