Seeing as N is much bigger than the other count variables, you can probably get speed-ups from computing only the unique combinations then indexing into that to define theta (see example here), but before you get too worried about performance you should pin down why this model is not recovering the data-generating parameters first. The fact that it takes so long suggests to me that something is awry in the first place, as this isn’t a particularly complicated model nor large data. First add an intercept (and a prior on said intercept) and see how that fares. Then if you’re still not getting good recovery, post your data-generating code here too as possibly the model is recovering the data-generating parameters accurately but you’re not generating the data in the way you think you are.
1 Like