Dear Allstaters,
I am trying to randomly generate a set of sparse large-scale graphical
Gaussian models. I've got a large number of nodes (>100), and in the
corresponding precision matrix most (95% ) off-diagonal elements are zero.
The remaining non-zero entries take positive as well as negative
values (this is important).
Is there an algorithm to randomly construct a valid sparse precision
matrix, i.e. one that assures that the resulting covariance matrix (i.e.
the inverse of the precision matrix) is always positive semi-definite?
Thanks in advance,
Juliane Schäfer
__
Juliane Schäfer
Institute of Statistics
University of Munich
Ludwigstr. 33
80539 Munich
Phone: +49 (0)89/ 2180-6408
Fax: +49 (0)89/ 2180-5041
Email: [log in to unmask]
|