Author: J. Wesley Cleveland
Date: 17:23:43 01/28/01
Go up one level in this thread
On January 28, 2001 at 17:12:34, Robert Hyatt wrote: >On January 28, 2001 at 12:57:30, guy haworth wrote: > >> >>Eugene's EGTs are best in real-time because: >> 1) they cover wtm and btm >> 2) physical file-compression and front-end caching have been used >> 3) the indexes are more compact because 'unblockable checks' are avoided >> >>See ICGA_J v23.3 last year. I don't believe that significant improvements can >>be made just by taking these ideas further. There are ideas for reducing the >>index-size of (e.g.) KQR... and KQP.... EGTs but they only save a few %. >> >>It would seem better to preface the Nalimov/Thompson EGTs with a values-only >>EGT, two bits per position (1-0, draw, 0-1 and 'broken') so that it is not >>necessary to go to the disc to find out the 'depth' of moves that reduce the >>theoretical value of the position. >> >>Peter Karrer has revisited ideas first demonstrated by Thompson (KQPKQ, KRPKR) >>and Schaeffer (checkers) whereby positions with the same P-positions form a >>contiguous sub-EGT of the EGT. Thompson also separated positions with different >>patterns of Bishop-square-colours. >> >> >Do you realize how big that would be? the 3/4/5 piece files use 8 bit scores. >If you cut that to 2 bits as you suggest, you _still_ need 2 gigs of memory to >hold them. I/O is going to happen no matter what you do... I suspect that the would be considerably smaller because of better compression, especially tables where one side nearly always wins, or they are nearly all draws, e.g. KQQQk or KNNKNN. This would make the caching more effectiive, and could greatly reduce I/O.
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.