The substantial formulation of the taxonomy problem may be read in the work, written as long ago as in 2nd century BC [1]. Democrit in «Letter to scientific neighbour» writes: «If you, my friend, need to investigate the complex conglomeration of facts or things, you at first distribute them into small number of heaps according to similarity. The picture will clarify and you will understand the nature of these things». 
     The taxonomy or grouping of objects (the terms «automatic classification», «self-learning», «cluster analysis» etc. are often used also) according to the similarity of their properties simplifies the solution of many practical problems of data analysis. 
     The same multitude of m objects may be divided into k taxons (k<m) in different ways. The man, performing the grouping, is guided by some criteria (let us denote them F), which help him to tell the good groups from bad ones and to select the best taxonomy variant. 
     The algorithms of FOREL family use the F criterion, based of the compactness hypothesis: one taxon must include the objects, similar in their properties to some «central» object. As the result, the sphere-shaped taxons are created. KRAB family algorithms use the hypothesis of l-compactness and the unite the objects into the taxon, according to similarity of objects to their neighbours. In that case the taxons of arbitrary shape are constructed.