Author: David Rasmussen
Date: 03:23:14 12/06/01
Go up one level in this thread
On December 06, 2001 at 05:40:54, Sune Fischer wrote: >On December 06, 2001 at 05:06:49, David Rasmussen wrote: >>No it doesn't. How do you know? Have you done any tests yourself? No. You just >>theoreticize. And falsely too, since my findings have been confirmed. > >I have too, didn't you read my post followup? > Not with tests from Crafty, no. Or any other program that uses a similar hashing scheme. >>>Well for pawn tables I believe this is the same. Since there are so few >>>different posistions, even one collision could mean a thousand collisions >>>because it is reused so many times. >>> >> >>Apparently not. Crafty has this problem. Still, it is pretty strong. Now that >>Bob changes back to 64-bit keys in light of my findings, I guess it will only >>get stronger, but the question is "how much stronger?". > >Yes I read that he he "confirmed" your results, but if you really had 300 >collisions you would have a really poor pawn eval(). There _are_ 300 collisions. Do you want me to show you the 600 positions in question that has the same pawn hashkey (that is the definition of a collision), and which comes in the first 3 minutes of search, and therefore causes true collisions? I have already posted one pair. I have verified the first 10-15 pairs. They were all collisions. You theoreticize. You say "because I can't understand it, and it doesn't work with my theory, it can't be true". Like the clericals and scholars, when Galileio asked them to look in his newly invented telescope, to see with their own eyes the imperfections of the moon, the moons of jupiter, that they didn't believe could be true. They said "It doesn't fit with out theory, so it can't be true. And we don't care for hard evidence." >I think some things has >changed in crafty since last Robert tested, perhaps a bad PRNG? >32 bits is enough, I'm pretty sure of it (both theoretical and emirical), unless >you have too many positions. > >>>Huh? >>>Neither me or Hyatt can confirm your findings, and since Bruce is also using >> >>Yes, Hyatt has. > >Give him more time, perhaps he will find the bug :) I give up. You'll never believe that 32 bits cause collisions, will you?
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.