How to Extract Relevant Knowledge from Grandmaster Games

Subtitle: Part 1: Grandmasters have insights – the problem is what to incorporate into practical programs

Article type: Research Article

Authors: Hartmann, Dap

Affiliations: Leiden Observatory, The Netherlands

Abstract: Evaluation functions of chess programs usually consist of different notions that are weighted, and added to yield one value that describes all the features of a given position in terms of material equivalents. In this paper, results are presented of the analysis of 62,965 positions that occurred in grandmaster games. Ten notions, some frequently used in evaluation functions, have been tested against these positions, and the results are shown as the average values of these notions for winners and losers separately. Also, the advantage of winners over losers by individual notion is shown. With this method, called the ‘Dap Tap’, grandmaster knowledge that is condensed in games can be tapped to test the importance of certain notions as sub-goals on the way to victory. Although this method does not yield reliable information on the relative importance of the notions, it is an excellent tool to test their relevance.

DOI: 10.3233/ICG-1987-10103

Journal: ICGA Journal, vol. 10, no. 1, pp. 14-36, 1987