- Date
2 June 2009
- Author
Anand Patil
- Contact
- Web site
github.com/onyin/covariance-prior
- Copyright
Anand Patil, 2009.
- License
MIT License, see LICENSE
This package provides a convenient prior for covariance matrices in PyMC. To use it, do the following:
c,o = covariance(name, v)where v is any vector-valued variable whose elements will always be positive. o will be an OrthogonalBasis object and c will be a deterministic returning a covariance matrix whose eigenvalues are v and whose eigenvectors are c.
OrthogonalBasis objects are matrix-valued stochastics whose columns form an orthonormal basis, but which are otherwise indifferent to their values. They are handled by GivensStepper step methods, which propose Givens rotations in randomly-selected planes.
By default, OrthogonalBasis objects' logp functions enforce orthogonality. You can skip this check for speed if you like by setting constrain=False.