Author: Wayne Lowrance
Date: 15:59:45 08/01/01
Go up one level in this thread
On August 01, 2001 at 13:56:27, Uri Blass wrote: >On August 01, 2001 at 10:45:41, Wayne Lowrance wrote: > >>On August 01, 2001 at 10:37:39, Wayne Lowrance wrote: >> >>>On August 01, 2001 at 03:54:23, Uri Blass wrote: >>> >>>>On August 01, 2001 at 02:06:06, Dann Corbit wrote: >>>> >>>>>On August 01, 2001 at 02:01:45, Uri Blass wrote: >>>>> >>>>>>On August 01, 2001 at 01:44:33, Dann Corbit wrote: >>>>>> >>>>>>>On August 01, 2001 at 01:40:03, Uri Blass wrote: >>>>>>> >>>>>>>>On July 31, 2001 at 18:49:37, Dann Corbit wrote: >>>>>>>> >>>>>>>>>On July 31, 2001 at 18:36:53, Otello Gnaramori wrote: >>>>>>>>> >>>>>>>>>><snip> >>>>>>>>>>Franz Morsch saying the main difference between this version of Fritz compared >>>>>>>>>>to its predecessors did not lie so much in greater chess knowledge but more due >>>>>>>>>>to the machines newfound ability to deal with anti-computer chess strategy, and >>>>>>>>>>to learn from its mistakes. He also said that he believed that this incarnation >>>>>>>>>>of Fritz is every bit as strong as the Deep Blue II that defeated Kasparov and >>>>>>>>>>has far greater "chess knowledge". >>>>>>>>>><snip> >>>>>>>>> >>>>>>>>>For a machine with many thousands of tunable chess parameters, carefully >>>>>>>>>adjusted by teams of programmers and GM's >>>>>>>> >>>>>>>> >>>>>>>>carefully adjusted by programmers and GM's? >>>>>>>>I doubt it. >>>>>>> >>>>>>>The programmers first used a gradient method, which was based upon thousands of >>>>>>>GM games. Then, individual parameters were hand-tuned by the GM's advice. >>>>>> >>>>>>I doubt if using GM's games to adjust the evaluation is a good idea. >>>>>>I also doubt if GM's advise for parameters is a good idea because I believe that >>>>>>the GM's do not think in the numbers of computers in games. >>>>>> >>>>>>The GM's have not time to calculate evaluation in the computer way and I do not >>>>>>think the advantage of GM's against computer is because of a better static >>>>>>evaluation. >>>>>> >>>>>>The advantage of humans against computers is their ability to think and change >>>>>>their evaluation during the game and the ability to plan that is not about >>>>>>evaluation. >>>>> >>>>>I might be wrong, but I rather suspect that all of the best computer chess >>>>>programs have advice incorporated from GM's. Perhaps some of it is indirect >>>>>(e.g. someone advised Robert Hyatt, someone looked at Hyatt's code). >>>> >>>>I suspect that it is only a small part of the work about chess programs. >>>>The biggest part is testing changes and I guess that it is done by games or test >>>>positions and not by GM's. >>>> >>>>Building the right test positions may be also a lot of work because good >>>>test positions should be taken from games when in part of the cases there is no >>>>tactics and the problem is to find a good positional move. >>>> >>>>Uri >>> >>>I hope I am quoting Christophe correct. He posted that he does not use game to >>>evaluate results. He has spent time to write testing code (part or all of which >>>is embedded in his development chess program to evaluate changes. Sounds like >>>the correct way to go if you have confidence in the evaluation code. >>>Wayne >> >>The evaluation of the testing code seems easy. >>a) set up a board position of the typcal problem. >>b) Nest the testing code without changing the chess playing program. It must not >>affect the before-after eval. >>c) Make the changes thought to improved the program. >>d) Check the eval and the board to observe the change in play if any. >> >>To simplistic Christopohe ? >>Wayne > >Not so easy >You need to find the right board positions. > >Uri Uri I was sort of thinking of reported and verified bugs which means that board positions are more or less defined. Such as GM program killer concepts that the program is being defeated by GM's Wayne
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.