Author: Tim Foden
Date: 03:10:41 02/08/04
Go up one level in this thread
On February 07, 2004 at 18:26:00, Uri Blass wrote: >I started to work on having some evaluation about endgame and >I wrote something to detect draws in KP vs K. > >It does not detect every possible draw but at least hopefully when it detect >draw it is correct if I have no bugs. > >Tablebases is not a solution because they are too slow and I believe that >generally functions are better because even in case of having tablebases if I do >not probe them in the qsearch I may get KP vs K that I need to evaluate without >tablebases and I want to return correct score without looking in tablebases. > >I read that yace is using bitbases even for 4 piece endgames when the bitbases >give only win draw loss information and I guess that the bitbases were >calculated from nalimov tablebases. > >I wonder what other people do in KP vs K endgame in case of not looking in >tablebases(because the program does not support tablebases or because it is a >qsearch node). > >Do they have a special function to detect the result or do they assume that >cases when they get KPK in the qsearch are rare enough when they use the 5 piece >tablebases because in most cases they probe the tablebases earlier. > >Uri Like you at one point I decided to try to code up the knowledge about KP-K endgames into GLC. I tried a number of things, but each time I found bugs in the algoritms. In the end I decided that it would be easier for me to write a KP-K endgame tablebase generator, to generate the KP-K table, which can be kept in memory and probed very fast. It takes only 0.15 seconds (AXP 2.1GHz) for GLC to build this 256K table (it uses more memory than Dieter's one, as it is not a bit base, and I currently don't use any clever indexing scheme at all.) Also, it isn't a distance to mate table, rather a distance to *successful* queening. I have since checked it against the Edwards KP-K tablebase and found it to be correct. Cheers, Tim.
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.