Author: Dan Newman
Date: 15:16:19 04/05/01
Go up one level in this thread
On April 05, 2001 at 10:21:04, Brian Richardson wrote: >Perhaps the definition of a "collision" is an issue. I use a 64 bit key and in >testing have not hit a true collision when the full 64 bit key is the same for >two different board positions (tested by also storing the entire board). In >actual practice, I DO get numerous "dupe" hash entries, where the first n bits >(n depends on the size of the hash table index) of the index into the hash table >are the same... I've always defined "collision" to mean two different positions hashing to the same table entry. I call two different positions hashing to the same hash code a "hashing error" or an "undetected collision". But the common usage here seems to be different... I've never been able to detect any errors with 64-bit hash codes. I've seen estimates of about 1 error/day for 64-bit codes and was able to detect about one error/s using 32-bit hash codes. If adding a bit reduces the error rate to about half (which seems reasonable, but I don't think is actually correct), then the error rate would be something like one error in 100 years with 64-bit codes. I still check the hash table move for legality though... -Dan.
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.