nscancor: Non-Negative and Sparse CCA

This package implements two algorithms for canonical correlation analysis (CCA) that are based on iterated regression steps. By choosing the appropriate regression algorithm for each data modality, it is possible to enforce sparsity, non-negativity or other kinds of constraints on the projection vectors. Multiple canonical variables are computed sequentially using a generalized deflation scheme, where the additional correlation not explained by previous variables is maximized. 'nscancor' is used to analyze paired data from two domains, and has the same interface as the 'cancor' function from the 'stats' package (plus some extra parameters). 'mcancor' is appropriate for analyzing data from three or more domains.

Downloads: