<p>The problem of searching for hidden managing patterns in complicated data is studied. A new approach to the problem is presented in cases where the desired pattern cannot be described directly in terms of the recorded features and is only an assumed rule for combining the effects of a priory known particular derivatives of this quantities in values of integral observables. This is of the most importance when the available sample is small, and the knowledge concentrated in such derivatives is based on more representative statistics. A variant of the approach based on generalized precedents is presented as a multidimensional analogue of the Hough transform, when the fact of detection in the secondary distribution and the exact form of hidden patterns are estimated based on the proximity of the forecast to the actual values of integral observables. Promising options for solving applied problems from these positions are proposed, in particular, options for arrangement of friendly interface to applied researchers.</p>