Author: Thomas Mayer
Date: 04:27:14 02/02/05
Go up one level in this thread
Hi Tony, On February 02, 2005 at 02:24:30, Tony Werten wrote: >On February 01, 2005 at 16:44:48, Thomas Mayer wrote: > >>Hi, >> >>of course this is not perfect, but sufficent, I think... In the line >>you see, that Quark did understand it. I believe that other engines will >>produce similar lines, but I have doubts that they score it near to that. >>You are welcome to test this with some engines... >> > >> >>It does really work... >>Well, Vincent might be right -> anyway, I think the nearly understanding >>of such positions MUST have some influence on it's endgame strength, I >>will continue to work on this... > >My feelings are about the same on this stuff, but be ready to be disappointed. > >XiniX has some of this knowledge fe KBBKP. Althought it picks Ka6 from the >testposition from BT2630 immediately, it needs 22+ ply to return a checkmate >score. > >Same with KNNKP. In a position posted here a few days ago, XiniX finds Ne3 in 3 >ply :) with a +0.4 score :( >Although it recognizes about 70% of all wins for KNN, alpha beta seems very >resourcefull to find some silly line that isn't recognized. > >BTW branchingfactor at ply 20 is still only 1.8, I wonder how high the >recognition percentage has to be to get it down to 1. ( Not only will a lower bf >enabled XiniX to search deeper, it will also give it less possibilities to find >silly lines that escape recognition) > >Anyway, fun stuff, but I don't expect it to make XiniX 50 elo stronger. well, I don't expect it either to make Quark 50 Elo stronger... but I believe it has some influence. Especially Quark was so far quite stupid in some easy endgames, e.g. it did still statically score KBKP as little advantage for the bishop side when it does not have TBs handy... it had no idea about KPK when P is on either a or h file etc... Also I have now an environment with which I can add special programmed endgames without losing speed. (Usually I gain speed, because the main eval function gets less complicate and the special evaluation is simply faster) Of course currently this is 3000 lines new code which has introduced some new +/- bugs, but I have started to debug them... Well, and the side effect that it evaluates some studies better then most other programs is something nice to have, even when it does not help in strength... :) Greets, Thomas P.S.: I do not think that I have anything in mind that can make Quark 50 Elo stronger... hehe, only possibility would be an idea what Stefan is doing in his Shredder to get so much deeper... :) PPS.: Well, it would be better to work on Quarks middlegame evaluation, but I am simply to bad in chess to make much progress there, so I start with endgames... :)
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.