jamieballingall / lambdapca Goto Github PK

Currently, PCA decomposes the covariance matrix using the Jacobi eigenvalue algorithm. A singular value decomposition (SVD) would probably be preferable. It would likely be more numerically stable as it avoids computing the covariance matrix. It would also likely be in two phases: bidiagonalization and full diagonalization. This may help with the recursion depth issue.

Combine PCA and PCA_INTERNAL into a single function so that users need only copy a single named range. This will require defining a recursive function within a LET clause. See the blog post on SumProduct.