On Wednesday, July 30, 2014 8:10:37 PM UTC-5, Donald Lacombe wrote: > > Dear Julia Users: > > I am attempting to speed up some code regarding some Bayesian spatial > econometric models and would like to translate some MATLAB code for > computing the log determinant term in the log-likelihood function. > Specifically, I'm interested in the Pace and Barry approximation to the > log-determinant term as in R. Kelley Pace and Ronald Barry. (1997) "Quick > Computation > of Spatial Autoregressive Estimators''. The term in question is det(In - > rho*W) which has to be calculated through each iteration of the > Metropolis-Hastings algorithm I'm using. > > The MATLAB code for this has the command "colamd" which MATLAB's help > defines as "P = colamd(S) returns the column approximate minimum degree > permutation vector for the sparse matrix S". >
colamd is indirectly available in the Cholmod module but I don't think there is a direct way of calling it. It is a method for determining a fill-reducing permutation for a sparse, symmetric matrix of the form X'X directly from X. There are several other approaches implemented in amd, Metis, Scotch, MUMPS, etc. for dealing with the symmetric matrix itself. All of these just use the pattern of nonzeros in the matrix as they are part of the symbolic stage of the sparse Cholesky or sparse QR factorization. Can you be more specific about how it is to be used? In particular, is it important to work with the pattern of X rather than the pattern of X'X? > Is there an equivalent Julia command? I've tried looking at "symperm" but > I do not think I am using it correctly and more importantly I have no real > clue. > > Any help would be appreciated and if I successfully code this I would be > happy to share with anyone who is interested. > > Thanks, > Don >
