Author: Angrim
Date: 13:22:34 07/10/02
Go up one level in this thread
On July 09, 2002 at 17:59:17, Robert Hyatt wrote: >I have been running some tests after prompting by Bruce, and the results >have been interesting. > >The question posed by him was "how many hash collisions (signatures match >but positions do not) can we tolerate in a search without having serious >problems?" > >I did the following: I made my HashProbe() routine use the node counter, >and every N nodes, I would "match" a hash entry even if the signatures were >not the same, simulating a collision. I got all the way down to a collision >every 1000 nodes without seeing even a score change at the root, which was >surprising. Do you really mean "a false match every 1000 nodes" or did you mean "a false match every 1000 collisions" which is what it sounds like you tested? There is a huge difference between these two error rates, especially for short searches which barely fill the hash table, and thus have quite few collisions. If your goal was to test "a false match every 1000 nodes" then I would suggest having HashProbe() return hash entry #0 every 1000 nodes even if there was a valid match. Angrim
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.