Author: Uri Blass
Date: 10:11:16 03/24/03
Go up one level in this thread
On March 24, 2003 at 12:47:06, Will Singleton wrote: >On March 24, 2003 at 11:26:56, Robert Hyatt wrote: > >>On March 24, 2003 at 02:07:31, Will Singleton wrote: >> >>>Position from Baron – Amateur bm Qxc2 >>>[D]2rr3k/pb3ppp/4p3/4P3/1n1B3R/1P4P1/P1q4P/1Q1RKB2 w - - 0 30 >>> >>>For some reason, Ruffian and Crafty have a very tough time with this one. I >>>wonder whether this could be due to some similarity between them. A problem in >>>repetition detection, perhaps? On my machine, amd 1.6ghz: >>> >> >> >>I doubt I have any repetition detection problem. However, for each of the >>programs can >>you give the score for the move _prior_ to finding the correct move? >> >>IE if a program says +.3 then it is _not_ going to want to find a repetition. >>If a different >>program says -.3, then it will go for the repetition as soon as it is found. >> > >The position seems to afford some repetition themes. My thought was that after >Bxa7, Qc3+ might be scored as leading to indeterminate positions due to long >sequences of moves driven by repetition chances (if handled incorrectly). Thus >Qc3+ wouldn't be seen as refuting Bxa7. > >That was my idea after first looking at the position. Then when I saw you had >changed the rep code, the idea seemed justified without further analysis. > >The difference between Crafty 19.2 and 19.3 in this position is rather dramatic. > It doesn't seem likely to be the result of the extension change you mentioned. >Have you found many other positions with such improvements? > >Will I also have a dramatic improvement but thanks to better evaluation. Latest Movei can see the tactical problem at depth 10 when Movei0.07.99 can see the problem only at depth 12. The reason is probably better king safety evaluation. If Crafty19.3 needs smaller depth to see that Bxa7 is losing then my guess is that it is because of some change in the evaluation and not because of repetition code. Uri
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.