Author: Robert Hyatt
Date: 15:31:13 07/03/03
Go up one level in this thread
On July 02, 2003 at 15:24:31, Dieter Buerssner wrote: >On July 02, 2003 at 15:00:33, Russell Reagan wrote: > >>On July 02, 2003 at 14:45:43, Dieter Buerssner wrote: >> >>>How do you come to the conclusion about the 18 plies with HTs? >> >>That is what Dr. Hyatt has said for years. > >I never saw him saying this. I remember him saying, that the reason [for solving >Fine 70 faster than in 26 plies] is bad moveordering (I don't agree). There is no other explanation. Hashing doesn't produce _faster_ solutions unless something is grafted from one part of the three to another part. If you search the tree optimally, this grafting happens, but not in a way that shortens the depth for solution. Here's the explanation I always use: You are searching to a fixed depth of 18 plies. You first search good moves for white and bad moves for black and you find that you reach position B after 10 plies, and you search another 8 plies to see winning the pawn. Now you search good moves for white and black, and you reach position B again, but now at depth 17 since black put up better defense. With the hash table result from the first encounter of B, you discover that B can be won, and you back up the right score. If you searched the _right_ path first, you wouldn't know B is winnable because you reach it with only one ply of depth left. IE the worse your ordering, the more likely you are to find a better solution earlier. Of course, the size of the tree is taking a hit from your bad ordering, but you luck into the solution faster than you should and the net effect is a faster solution. But just for that position, not the general position which takes a hit for bad ordering. > >Anyway, if he said this, I'd be interested in the reason(s) for the 18 plies. I >doubt it. The 18 was what I solve it at, although that is not an absolute. I've had it take from 18 to 22 plies with a known-non-buggy program. It is way sensitive to replacement policy, table size, and even the random numbers used for the Zobrist signature. if it takes 26, hashing isn't helping, but it doesn't mean anything is wrong. If it takes more than 26, there's a bug. :) > >Regards, >Dieter
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.