The numerical restrictions of Bayesian method of pattern recognition are investigated. The probability maximum of correspondence of local state to one of the basis patterns is defined through the expansion of examined vector over the patterns. The practical example of recognition without errors is presented in the frame of nearest neighbor method for the case, when the probability interpretation of the expansion coefficients is not valid. The spectral portraits of solving matrices are constructed for the literature texts author identification problem.
non-stationary time series, probabilities of letters combination, basis patterns, pattern recognition
Mathematical modelling in actual problems of science and technics