A method is proposed for constructing plausible and easily interpretab
le hypotheses about dependences in data. This method allows the use of
both numeric and symbolic attributes. The main advantage of the metho
d is the fact that the most informative combinations of attributes are
selected for descriptions of hypotheses.