rcar_normal for sample generation

[Hanah]

Hi all,

I am trying rcar_normal function to get random generation for ICAR distribution. 

I follow the example in the manual documentation:

num <- c(1, 2, 2, 1)
adj <- c(2, 1,3, 2,4, 3)
weights <- c(1, 1, 1, 1, 1, 1)

and then generate a sample by

rcar_normal(n = 1,  adj,  weights = adj/adj,  num)

but I got this error "Error in eval(quote(model[[nodes]]), parent.frame(3)) : object 'model' not found"

Could someone advise how to fix this?

Thank you very much.

Best regards,

Hanah.

[Daniel Turek]

Hanah, I'm sorry for the brief response.  But the dcar_normal distribution (intrinsic, or improper CAR distribution) is not a proper distribution, and cannot be used for random generation of values.  See the comments available in help(rcar_normal) for details and more reading.

The example that I think you're looking at, also from this documentation, is used to calculate the density, not for random generation, as:

x <- c(1, 3, 3, 4)

num <- c(1, 2, 2, 1)
adj <- c(2, 1,3, 2,4, 3)
weights <- c(1, 1, 1, 1, 1, 1)

lp <- dcar_normal(x, adj, weights, num, tau = 1)

An implementation of the "random generation" function rcar_normal is only in nimble, because in some of the model initialization routines used internally by nimble, this function could be called (from within a nimble model object).  So such a function must exist, although it does not generate values from the intrinsic CAR process (which cannot be done). This also explains why it's looking for a "model" object, since this routine expects to be called from usage of a dcar_normal() prior distribution, within a model object.

Cheers,

Daniel

--
You received this message because you are subscribed to the Google Groups "nimble-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nimble-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/ad8bc5b9-49a9-49a1-8c41-4543d63f8f18n%40googlegroups.com.

[Chris Paciorek]

Please also see this discussion in a GitHub issue on the nimble GitHub repository, which provides a bit of guidance on how one could write one's own code to simulate from a constrained

version of the ICAR prior.

To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/CAKbe0hrG%2BqiTwSX4tzT_vH6VxhnUVgpfZWvDbBBpPbvA5xR%2B4g%40mail.gmail.com.