Author: Inmann Werner
Date: 03:43:06 03/10/00
Yesterday, I absolutly was unhappy about the evaluation function of my engine, so I trashed it and left only the material balance. (I found some serious bug in evaluation which, when corrected had no effect in my testsuites..) For fun, I started the LCTII Testsuite for Tactics and..... I had an "all times best!!!!" The engine solved all the same positions as with evaluation, but much quicker!! The endgame positions did not work as good, cause a little knowledge about passed pawns there is needed (although endgame Tablebases was used) but it scored about 50% as if with evaluation. Till now, I much tested with tactical positions..... Conclusion for me: If I improve my search algo, I still will use the tactiacl positions. But if I put some knowledge in, I only will look, if the knowledge disturbs the tactical search. What I need now: Positions for computers, where you can test knowledge. No very hard positions, but clear!! and simple!! positional things, which never can be found tactical. 1) king safety and threat out of horizon.(>ply 18) 2) rook on free lines.. etc etc (I am too poor in chess, to tell you the rest...) The goal: If I put in knowledge, and the tactics do not suffer too much, but the positional positions are found quicker, the engine is better. For each problem should be some different positions, cause it is very simple to tune a engine for one position with "knowledge", which is bad for the same problem in other positions. I know, the end test will be games, but ....... In many suites, there are a lot of sacrifice positions. I do not like them very much. The best solver for it is my "without knowledge" engine, if it finds the payback. It do not like the idea to accept sacrifice for some positional advantage, the program not really understands and can push.... Werner
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.