The conclusions about the strata of society, various parties are supported by, have been made. The method of information-extreme machine learning of the system of functional diagnosis of the technical state of a complex machine with the optimization of the hierarchical data structure is considered. It is shown that the functional efficiency of machine learning of the system of functional diagnosis is significantly influenced by the location in the hierarchical structure of the recognition classes characterizing the technical state of the machine and its nodes. At the same time, for each level of the hierarchical structure under consideration, a restriction on the number of recognition classes is imposed, which makes it possible to reduce the degree of their intersection in the space of diagnostic features. Optimization of the hierarchical structure was carried out in the process of information-extreme machine learning of the system of functional diagnosis, which allows to maximize the information capacity of the system. As a criterion for optimizing the parameters of machine learning, we considered a modi fied information measure of Kulbak, which is a functional of the accurate characteristics of diagnostic solutions. In this case, the algorithm of machine learning represented a multi-cycle iterative procedure of finding the maximum global value of the information criterion for optimizing learning parameters in the working (permissible) domain of determining its function. Based on the optimal geometric parameters of recognition class containers obtained in the course of machine learning, decision rules have been constructed that allow making diagnostic decisions in a real time. As an example of the implementation of the method of optimization the structure of input data, the machine learning of the system for the functional diagnosis of a mine hoist was considered. As a result, alphabets of recognition classes have been created for strata of all tiers of the hierarchical structure, providing the maximum functional efficiency of machine learning.