Details
-
Improvement
-
Status: Closed
-
Minor
-
Resolution: Fixed
-
2.2
-
None
-
None
Description
The CorrelatedRandomVectorGenerator class uses a kind of rectangular Cholesky-like transform M = B.Bt where B is a rectangular matrix. The difference with respect to a regular Cholesky decomposition is that rows/columns may be permuted (hence the rectangular shape instead of the traditional triangular shape) and there is a threshold to ignore small diagonal elements. This is used for example to generate correlated random n-dimensions vectors in a p-dimension subspace (p < n). In other words, it allows generating random vectors from a covariance matrix that is only positive semidefinite, and not positive definite.
It would be nice to have this decomposition available as a stand-alone class outside of the CorrelatedRandomVectorGenerator.