Non-linear canonical correlations with a M sets of variables.

Non-linear canonical correlation analysis, OVERALS, is introduced as a generalization of both linear canonical correlation analysis and homogeneity analysis. The basic notion of OVERALS is minimizing a loss function under certain conditions. In the OVERALS program the parameters are solved for by alternating least squares. OVERALS is invariant under certain transformations of the data. There are multiple and single transformations. The latter ones consist of three types: nominal, ordinal and numerical. An application of OVERALS is treated in the last section. (A)