I’m writing “toy” Gibbs sampler and my idea was to use LKJCholeskyCov for covariance matrix prior in
“stand alone mode” (without pymc model). But seems it has no “random” method like others distributions. So my question is why and is it possible to use LKJ distribution outside of a model.
yep LKJCholeskyCov does not have a random method. I think you can try modifying the LKJCorr random method : generate a random correlation matrix using LKJCorr random method, and transform the sample, then combine with the random generated diagonal matrix using sd_dist.