Author: Gian-Carlo Pascutto
Date: 12:36:19 07/10/00
Go up one level in this thread
On July 10, 2000 at 14:39:02, Dann Corbit wrote: >>PostModernist 2414 2444 | 2404 +3 | 45% 26 687 178 >>Falke 2453 2438 | 2447 +2 | 64% 3 53 14 >>QueenMaxima 2355 2385 | 2369 -5 | 73% 51 1153 122 >>Hossa 2347 2380 | 2300 +15 | 51% 71 1831 171 >>Sjeng 2257 2298 | 2139 +39 | 76% 20 166 0 > >That's a pretty big leap up. Do you have an explanation for a 40 point rise? I fixed positional learning, finetuned the eval, added new endgame and material inequality code, made the search less risky, added some root- processing, evaluation caches for standard chess, optimized the movegenerator and do not allow different types of eval during a single search to keep the scores more consistent. I now use an automatically generated opening book out of a 200Mb PGN database, immediately switch moves on a fail-high and try to do something sensefull on a fail-low. According to some highly scientific analysis on the WinBoard forum, the result is a program that plays significantly worse than TSCP. Most of these improvements are also in the version I submitted for BOTC. -- GCP
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.