Author: Moritz Berger
Date: 04:13:49 07/27/98
Go up one level in this thread
On July 27, 1998 at 06:25:55, Don Dailey wrote: >On July 27, 1998 at 02:03:33, Moritz Berger wrote: > >>On July 26, 1998 at 23:37:23, Mark Young wrote: >> >>>I think it is wrong for people to think >>>fritz 5 is best at fast time controls. I found the program much stronger at >>>slower time controls. > >Don't all programs play much stronger at slower time controls? The >question is does Fritz improve more at slower time controls than >Junior, the program in question. It wouldn't suprise me if you >were right on this however. Each ply of search does A LOT to improve >your chess, and this doesn't just apply to tactics as many people seem >to think. > > >> >>I subscribe to this statement. Fritz plays best at 40/120 where search depth >>often compensates for positional misevaluations (that can be observed in all >>chess programs). One striking example was a French game I played a year ago >>(Fritz was black) where Fritz wanted to blunder away the game by opening the >>position against itself - only at the last ply (which was 12 or 13 AFAIR on my >>P166) it always so the refutation after a couple of minutes and finally went >>ahead and won the game. >> >>Moritz The point is that knowledge in most programs doesn't consist of exact plans but of mere rules of thumb that are correct only most of the time (as the authors certainly hope). Some people confuse rules of thumb with real understanding when there is a huge grey area of uncertainity. The deeper you search, the more likely it is that an "exact" search will outperform "fuzzy" knowledge with a shallower search. The key is being smart at pruning the search tree and being selective enough to get that extra plys (one ply won't be enough). Fritz is not only very fast (lean evaluation) but also very much selective in its search which I consider to be a different approach of implicit "pruning knowledge". It usually gets 2-4 ply deeper than others in middlegame positions which often neutralises the "knowledge" in the statical (more expensive) evaluations of other programs (that might be worth 2-4 ply as well from a positional point of view but doesn't give them the 2-4 tactical plies Fritz gets). The big problem is how to implement strategical plans - I believe that most programs just survive as long as there is some tangible positional shift within their search horizon. Statical knowledge is insufficient to give a good enough representation of the reality on the chessboard to work with e.g. a 1 ply search and win against strong opponents. 40 ply plans are still more likely to emerge from the experience and intuition of the very best human Grandmasters. Junior and The King are very successful by (as far as I understand it) not pruning the entire search tree but using all kind of extensions on top of a shallower search. Please note that I don't claim to know for certain that these programs work that way, it could be very different in reality but this is how I perceive how they work. Moritz
This page took 0.01 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.