Hey folks, I’m looking to use the “new” (I know, it’s been around for ages now; I just haven’t started using it until now) offset & multiplier feature to non-center-parameterize a hierarchical model. I have a version working whereby I do (excuse my idiosyncratic use of whitespace):
...
parameters{
real mean_subj_mean ;
real<lower=0> sd_subj_mean ;
vector<
offset = mean_subj_mean
, multiplier = sd_subj_mean
>[num_subjects] subj_mean ;
}
...
Then I attempt to expand my model to a scenario where subj_mean is a matrix:
...
parameters{
vector[num_locs] mean_subj_mean ;
vector<lower=0>[num_locs] sd_subj_mean ;
matrix<
offset = rep_matrix(mean_subj_mean,num_subjects)
, multiplier = rep_matrix(sd_subj_mean,num_subjects)
>[num_locs,num_subjects] subj_mean ;
}
...
Which passes the syntax checker, but yields a runtime error:
Exception: constraint: subj_mean ((3, 10)10, 3) must match in size
Am I doing something wrong, or is a matrix-valued offset & multiplier not possible?