Author: Ed Schröder
Date: 07:15:54 10/18/98
Go up one level in this thread
>>Since ages I use the following formula for creating a set of random
>>integers for hashing at program start:
>>#define LENGTH 64*12
>>int random [LENGTH];
>>int a,b,c,d,x;
>>srand(9); // initialize prime number
>>for (x=0; x<LENGTGH; x++)
>> { a=rand(); b=rand(); if (b & 1) a=a | 0x8000;
>> c=rand(); d=rand(); if (d & 1) c=c | 0x8000;
>> random[x]=(a<<16)+c;
>> }
>>I wonder how good such a system is and how others do it.
>>- Ed -
>>From the definition of LENGTH I guess these are piece-square tables you are
>generating.
Yes.
>I do the same, but I generate 48-bit numbers (actually only 45-bit, because I
>don't bother completing the MSbit), not 32-bit as you do. I think it's
>generally acknowledged that 32-bit is not enough ?
Let me give an example with a 32-bit hash-key:
Total positions searched : 1,386,498
Hash errors : 453
These 453 errors are discovered checking the positions when a hash hit occurs
so they are 100% examples the hash algorithm fails. I find this number quite
(much too) high.
When I increase the number of bits to 40 or 48 the number of errors drops to
(say) 200 / 150. Still much to high IMO.
So my suspect goes to random generator.
- Ed -
>Amir
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.