The invention allows modeling and searching patterns in a data sequence. A data sequence is analyzed to determine at least two data elements that, when replaced with one replacement data element, will cause a metric of irregularity of the data sequence being analyzed to be maximized. The replacement data element is not included in the data sequence being analyzed. The determined at least two data elements are replaced with the replacement data element thereby generating a residual data sequence. The steps of analyzing and replacing are iterated until reaching a predetermined value of an iteration threshold, wherein with each iteration the analyzing and replacing are applied to the residual data sequence generated in the preceding iteration.