Author: Ingo Lindam
Date: 03:02:37 12/10/02
Hello all, in addition to the 'testing position criteria' discussion, (see http://www.talkchess.com/forums/1/message.html?269784 ) where the focus is in particular on testing positional skills of chess engines I would like to give my oppinion that searching for pure positional positions (that still are interesting enough for tests) can't be successful, because every position you can't just evaluate by knowing the result win/loss/draw without to obtain anything includes tactics if you just look deep enough into the tree (or even if you just have a look into the tree of any depth). So if you want to evaluate positional skills you have (in my oppinion) just the following choices: 1) accept that you can't distinguish between positional and tactical skills while testing 2) making a complex analysis on a complex answer (atleast an treelike analysis) to the choosen test positions to distinguish as far as possible between positional and tactical aspects in the given solution AND being sure that it will be a very complex method of testing AND being sure that you will probable never get more feedback on your testing like 'this is right' than you get feedback like 'you are wrong!' OR 3) just test the engines by evaluating the scores/evaluations it gives to test positions without obtaining a tree (= without generating moves) in the test position. Then ofcourse it might be more usefull to give the machine rather a lot of positions from different lines of (a human) analysis that should be clear to evaluate without looking forward in meaning of generating moves... than giving the machine the key position where it needs to find the 'one and only' and 'very surprising' 'key move' and the 'corresponding PV' to evaluate the given position right. (BTW: that definition of good test positions always sounds strongly like tactics to me) Again: For me 'positional' means making (or giving weights to) decissions/evaluations without obtaining variations and on basis of empirical knowledge that is always linked with a likelihood for being true in this very special case. Internette Gruesse, Ingo
This page took 0.01 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.