A python implementation of Knol, D. L., & ten Berge, J. M. (1989). This algorithm converts a non-positive definite matrix into a positive definite one. In addition, it allows constraints to fix some correlation factors and force the smallest eigenvalue to have a certain value. Note: the second constraint related to eigenvalues has not been implemented as it does not relate to my current use case.
For information about the usage of this code, please view the example in the repository.
The ktb1989 script imports numpy and scipy (code written using numpy==1.19.4 and scipy==1.5.4)
Depending on my future use of this code, there are the following changes that I would make:
- additional checks/functionality to make sure constraints (such as matrix being symmetric) are adhered to
- improved documentation
- support for additional input types (like pandas DataFrames)
- additional utility functions for re-ordering matrices
- implement the remaining pieces of the algorithm