Abstract:
To solve machine learning problems, we have developed a method to identify closed sets of common features of objects (patterns) of the training sample. The novelty of the method lies in the fact that it is implemented within the concept of constraint programming and uses a new type of table constraints — compressed tables of the $D $-type — for internal representation and processing of the training sample. Search reduction is achieved by applying the proposed method of branching the search tree and using partial order relations on sets of objects (features) to prune unpromising branches. The method has a computational complexity estimate that for some types of input data is better than the estimates obtained for the studied prototypes.