There has been developed a large collection of algorithms to carry out different types of analysis of data. Common for these algorithms is the H-method, which is an implementation of the H-principle. From the modelling point of view the basic idea is to identify an appropriate covariance. According to the H-principle the covariance is the measure that gives a balance between the fit and prediction aspects of the modelling task. It is suggested to select weight vectors that extract as large covariance as possible.