On the representation of training tables in a K-valued code and the construction of empirical regularities

The problem for the transforming of information given in a training table in an arbitrary code into a k-valued code, without violating the classification, is discussed. The necessary and sufficient condition for the existence of such a transformation in the pattern recognition problems is the nonintersection of classes. Algorithms for effective coding have been suggested. To the k-valued training tables, constructed in this way, it is convenient to apply combinatorial logic approach for pattern recognition problems. This method may be used also in expert systems for the construction of empirical regularities in the properties of the objects