Author: Robert Hyatt
Date: 14:50:55 08/20/02
Go up one level in this thread
On August 20, 2002 at 16:05:17, David Hanley wrote: >I read with interest the experiments with crafty with introducing hash false >hits artifically into the search tree. I recall the experiments showed that >adding the false hits, even at a very high rate 1 per 1000, had little of no >effect on the search results. > >It would be good for me to use 32-bit signatures in my program, and i'm >wondering if the above result indicates that 32-bit signatures won't matter so >much in my program, which is a slow searcher. Certinaly if i check hamming >distance, 32 bit hash false matches should be at a far lower rate than the one >per thousand that didn't adversely affect crafty. > >dave I am not certain. IE it appears that you will definitely get a number of collisions. But it is not clear if they will hurt you (on average) or not. I don't personally like the idea of any collisions at all, but it is obvious that the large trees we search are very robust in their ability to withstand damage from such errors. Far better than I would have thought...
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.