Learning Representative Patterns From Real Chess Positions: A Case Study

Warning

This publication doesn't include Faculty of Sports Studies. It includes Faculty of Informatics. Official publication website can be found on muni.cz.
Authors

ŽIŽKA Jan MÁDR Michal

Year of publication 2003
Type Article in Proceedings
Conference Proceedings of the First Indian International Conference on Artificial Intelligence (IICAI-03)
MU Faculty or unit

Faculty of Informatics

Citation
Field Informatics
Keywords pattern recognition; decision trees; classification; representation of examples; relevant attributes
Description This paper deals with a particular pattern recognition by machine learning. The patterns are specific chess positions. A computer learns if a special pattern leads to a winning or losing game, i.e., a classification task based on real results and examples. As a learning algorithm, decision trees generated by the program C5/See5, also with boosting, were used. This algorithm does not employ chess rules or calculations of positions, it just learns from a selected set of 450 training positive and negative examples with 8 different representations of real positions played by human players. The most accurate classification reaches 92.98% for the combination of automatically generated trivial descriptions of positions (64 attributes) with expert descriptions suggested by humans (92 attributes).
Related projects:

You are running an old browser version. We recommend updating your browser to its latest version.

More info