Apologies; I misunderstood which link you were recommending; ill have a look at your new recommendations right now! I thought you were reccomending this approach VAR-Blog/Big VAR Blog Post.ipynb at main · jessegrabowski/VAR-Blog · GitHub, which which does not use the var package from experimental
I’d like to eventually take sums of categories and draw posteriors for them, so more flexibility is always nice (Here i’d be feeding the mean for two response additive log transformed categories and then passing them to a dirichlet distribution to output probabilities)