Advice on dimensionality reduction?

How could my models’ priors be quickly sampled, in their model and data context?