Nonlinear transformations of different type features and the choice of latent space based on them

The problem of forming a latent feature space through nonlinear transformations of different type features is considered. Two types of transformations are used: the replacement of gradations of nominal features by the values of the function of objects belonging to classes and the combination of features according to the rules of hierarchical agglomerative grouping. The dimension of the new latent space is less than the original one and it is determined by the grouping algorithm. The ordering of latent features in relation to informativeness allows solving the problem of the curse of dimensionality and visualizing data taking into account the description of class objects.A comparative analysis of linear and nonlinear methods for reducing the dimension of space is given. The division of methods using the division of objects into classes and without such division is given. Without division into classes, the PCA and T-SNE methods are implemented on data in interval measurement scales.Using the method of calculating generalized estimates of the objects it is doing their visualization according to a certain set of different type features.