Author: Robert Hyatt
Date: 21:23:42 01/14/98
Go up one level in this thread
On January 14, 1998 at 22:20:57, Stuart Cracraft wrote: >That endgame tablebase search you posted was certainly interesting. >Can you digress a little bit about what you did to integrate tablebases >in standard play. To the endgame tablebase layperson, it would seem >impossible: the files are so large as to not be comfortable in memory >and fseeking a disk file during tree search at end points or in the >search would be too slow. What's the trick? > >Also, with maximum tablebases enabled, what do you think is the >increase in ratings point strength on ICS? > >--Stuart > no magic. I do use fseek() deep in the search tree, but there is one trick. I only probe immediately after a capture. In fact, I only probe after captures that take me to the specific 5 piece files I have, or any 3-4 piece files. I probe anywhere depth is > 0, which means *not* in the q-search. It slows down by maybe 50% (worst case) but the scores are 100% correct, so the trade-off of speed for accuracy is worth it. No idea about a rating improvement, but it is obviously better with than without. KRP vs KR has been a big winner or drawer. KRB vs KR and KRN vs KR have also been useful since those are pretty common.
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.