Author: David Rasmussen
Date: 11:23:44 12/03/01
Go up one level in this thread
On December 03, 2001 at 07:54:00, Gian-Carlo Pascutto wrote: >On December 03, 2001 at 07:08:43, David Rasmussen wrote: > >>If you have collisions, you _will_ fuck up evaluation. > >True. But one needs to think about how much this matters. > >I did a test (and posted the numbers here) of 32 bit hashing. > >I was not, and have never been, able to determine any kind >of weakening of my program when using only 32 bits hashing. > >The amount of collisions when hashing only pawns will be >significantly less. > >You need to think about, and preferably test, whether it >is beneficial to take the few extra collisions and make >hash updates + storage significantly cheaper, or not. > >There is a common misconception that collisions are so >evil that they will automatically weaken your program >when they happen. That is totally false. > I understand that, but the speed gain isn't that big from 64 bits to 32 bits, and while collisions might not themselves be evil, the increase in complexity of debugging etc. sure is. A very important invariant, that a position's pawnstructure is always evaluated the same way, doesn't hold. I don't like that for a small speed gain. But sure, it's a design choice. /David
This page took 0.01 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.