The problem of selection of an adequate, in some cases, and a target model (?, ?, P(?)), in other cases, for a statistically steady experiment E from some parametric family {(?, ?, Pu(?)): u ? U} has been consistently solved in [1-4]. Preliminary results on this problem were first presented in 1986 at the Steklov Mathematical Institute of the Russian Academy of Sciences and at the Tenth Prague Conference on information theory, statistical decision functions and random processes. The approach proposed in [1-4] is based on the general idea of the experiment (control system) in the form of a scheme, information, coordinates and function [5-9]. This allows one to select m ? 2 meters ?1, ?2, ..., ?m [1] of elementary outcomes {?} ? ? of the experiment and then build its probabilistic model (?, ?, P(?)) with the use of the information on a single trial of the experiment. A simpler problem is considered in the classical statistical decision theory when it is necessary to choose the distribution of a random element ? ? (?1, ?2, …, ?m) from some parametric family {Pu(?): u ??U}. In this case, the strategy of choice of the distribution does not depend on the parameter u ??U, and depends only on observations of the element ?. In this work, which completes the studies in [1-4], the problem is solved on the reduction of the selection error probability of the model (?, ?, P(?)) from the family {(?, ?, Pu(?)): u ? U.
|
