I am looking for a geometric interpretation of CCA. Especially one that relies on the fact we are doing singular value decomposition, which has the geometric interpretation of a rotation, scaling and another rotation. Why taking these two rotations from the SVD of the cross-correlation matrix gives the most correlated linear projections?
1
-
2$\begingroup$ Can this help? stats.stackexchange.com/q/65692/3277 $\endgroup$– ttnphnsSep 21, 2016 at 11:12
Add a comment
|