Author: Tord Romstad
Date: 06:33:22 04/07/99
Go up one level in this thread
On April 07, 1999 at 02:39:10, blass uri wrote: > >On April 07, 1999 at 00:59:33, Ted Sutton wrote: > >>Uri, >> have you found positions where Junior5 finds the best move? >> What sort of postion does that tend to be? > >I remember that there was one tactical position when Junior was the only program >that could find kasparov's move against ivantuk. > >I cannot say that Junior is better at tactics and it is only about >this specific position. > >Junior also found the best move in a study because it had a different evaluation >function and evaluated the rook more relatively to the passed pawns. > >Fritz5.32 and Genius5 could not find the right move in the same position because >they believed that the passed pawns are better than the rook. > >Junior can also find the best moves in some endgames because it gives endgames >of KRN vs KR or KRB vs KR evaluation that is not a big advanatge for the better >side. > >Junior can find better move than Fritz sometimes because Junior is not a stupid >root processor like Fritz. > >Fritz can trade queens and understand only after trading queens that it was a >bad idea and not because it sees more plies. > >Something like this cannot happen to Junior. I am not sure about this --- I very much doubt that Junior is doing a full eval at the leaves (it is simply too fast for this). My guess is that Junior uses some kind of hybrid approach. A preprocessor is probably applied somewhere _within_ the three. When Junior does an n-ply search, I believe that the preprocessor is applied at all positions of depth n-d (for some small value of d) and that the eval is updated incrementally for the last d plies (plus extensions). This is certainly a very oversimplified explanation of what Junior does (just thinking about all problems this would give in the implementation of hash tables gives me headache, for instance), but I don't think I am missing the mark completely. If I am right in my above speculations, the problem you describe in Fritz will sometimes also occur in Junior, but of course not nearly as often. Tord
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.