Big Data with HierarchicalRegression

The same question comes up recently: ADVI Minibatch slows down with increasing size of data

According to the profiling, what slows down the sampling is the part data is being indexed (for minibatch), I wonder if there is any way to optimized that… @ferrine?