JAGS 3.2.0 requires conjugacy to sample a symmetric, positive-definite (SPD) matrix. It is common, however, to compose a covariance matrix in a sparse fashion through taking the kronecker product of lower-dimensional SPD objects. Each lower-dimensional SPD term in the product are conjugate under a matrix normal distribution. Is it possible to make an update to JAGS to allow a take advantage of the matrix normal conjugacy of a covariance matrix constructed from outer products? Such an enhancement would open up a large class of models for JAGS. Even better, of course, would be to permit non-conjugate updates of covariance matrices.
We already discussed kronecker products in this thread. I'm happy to consider it as a feature request, but I need a complete example to test the code.
Sign up for the SourceForge newsletter:
You seem to have CSS turned off.
Please don't fill out this field.