Author: Angrim
Date: 03:46:50 11/13/99
Go up one level in this thread
On November 13, 1999 at 06:37:20, Angrim wrote: >On November 13, 1999 at 03:52:43, Alexander Kure wrote: > >>This is exactly what Nimzo 7.32 already does. All 3 and 4 piece endings have >>been converted from Nalimov into a special format (NCD - Nimzo compressed data), >>where only the information of win, draw or loosing for each position is stored >>(needs only 2 bits per position). At startup these endings are loaded into >>memory (requiring app. 11 MB for all 3 and 4 piece egtbs) and probed in the >>q-search, which is faster than calling the evaluation function! > >2 bits per position is still more than you need since there are only 3 >values possible. Useing base 3 you can store 5 entries per byte instead >of just 4, saveing 20% on memory. This can be probed just as fast as the >2 bits per entry method since you can still directly calculate which byte >of the table contains each entry. >To extract entry N read byte index N/5 and extract the >(N mod 5)+1st trinary digit from this byte. Oops, changed the meaning of N between the 2 sentences.. following should read: let M be the value of the byte, then the first trinary digit of M is M mod 3, the second is (M/3) mod 3, third is (M/9) mod 3, etc >Where the first digit is N mod 3, the second is (N/3) mod 3, third is >(N/9) mod 3, etc. > >Hope this is helpful, >Angrim
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.