Author: Ron Murawski
Date: 07:03:30 08/17/02
Go up one level in this thread
On August 17, 2002 at 03:20:19, Gian-Carlo Pascutto wrote: >On August 17, 2002 at 00:51:59, Will Singleton wrote: > >>I don't have my source code with me now, but I remember that OR'ing the two >>32-bit values will give very bad results. > >Why? > >-- >GCP Trying to build a 64-bit random number sequence from several smaller sources is fraught with danger. I remember testing a bad 32-bit psuedo-random number generator for dice simulations and it failed miserably due to serial correlation. Serial correlation is a formal test of successive random number sequence pairs reduced to several ranges for long sequence runs. For my dice simulation it turned out that rolling a "1" on the first die increased the likelihood of rolling a "6" on the 2nd die and lessened the likelihood of rolling another "1". For Zobrist hashing a test for hamming distance of successive numbers would be a start. Testing every other number would follow, etc. Any psuedo random number generator that suffers from serial correlation will yield less than optimal random numbers. The question whether these "less than optimal numbers" are bad for Zobrist hashing is unanswered, but I tend to mistrust them. A very good 64-bit random number generator with C-source can be found here: http://burtleburtle.net/bob/rand/isaacafa.html Look for ISAAC-64. Ron
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.