Author: Omid David Tabibi
Date: 11:25:17 01/17/03
Go up one level in this thread
On January 17, 2003 at 12:52:59, scott farrell wrote: >On January 17, 2003 at 01:20:23, Omid David Tabibi wrote: > >>On January 17, 2003 at 00:48:48, Robert Hyatt wrote: >> >>>On January 16, 2003 at 20:29:54, Omid David Tabibi wrote: >>> >>>>[D] 4k3/8/8/p1p1p1p1/PpPpPpPp/1P1P1P1P/8/4K3 w - - 0 1 >>>> >>>>At the above position, some programs (e.g. Fritz, Hiarcs, Crafty) reach their >>>>maximum depth in a second or two (see the Crafty analysis below), while others >>>>(e.g. Tiger, Shredder) need more time as they search deeper (Tiger for example >>>>quickly reaches depth 20, and then slowly goes deeper and deeper, see the Tiger >>>>analysis below. For Genesis it takes about 4 minutes to reach depth 60). What is >>>>different in the former class which enables them to quickly reach the max depth? >>>> >>> >>>This is a hashing problem. There are not many unique positions, so hashing >>>should cause the search to be very efficient... nothing but king moves helps >>>this even more. >>> >> >>There are only 16 (white) x 24 (black) x 2 (sides) = 768 unique positions. But >>still normally you cannot retrieve everything from hash table: Let's say you >>have a position stored in hash table with depth 30, but now you need a depth 31 >>search, so you cannot use the hash result; after the next iteration you will >>need a result for depth 32 and so on and so on... >> >>Genesis gets about 30% hash hits from the above position. >> >> >>> >>>>Another interesting point is that Crafty (and many others) gradually reduce the >>>>score as they searches deeper (in the below log you can see that at depth 47 >>>>Crafty gives the score 0), while some others like Tiger, remain with their fixed >>>>score (-0.24 in the log below). Based on what factors is the score reduced in >>>>Crafty? >>>> >>> >>>Probably draw by repetition. The side that is behind won't move his king very >>>far from the original square. The other side will try every possible square >>>but after 40+ plies he runs out of new squares and the other side can force a >>>repetition. >> >>If so, then why doesn't that happen with all the engines? Tiger never changes >>the score; Genesis reached the depth 80 and never changed the score (and to the >>best of my knowledge Genesis doesn't have any repetition detection bugs). >>Besides, the score drop is usually gradual, not a sudden change to 0. > >My chompster engine has some experimental blockade detection, the score starts >around0.05 and gets to 0.0 by ply 8 with about 2000 nodes. And depth20 has only >20000 nodse, and a score of 0.0 still. > >I am thinking other commerical might have bloackade detection, and if they did, >they would behave markedly different to those that did not. > I also have a blockage detection system, managing to detect the most complex King and Pawn blockages even despite some dynamic pawns' presence, but the above example is interesting for hash table debugging purposes. >Scott
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.