I thought about that because of papers of Ovaskainen team (2016, 2017), which implement two very interesting features :
- Sparse infinite bayesian factor models. This approach cannot be directly implemented in stan, but could potentially be approached by shrinkage or marginalization, I think, even if I do not undestand the subtility of these subjects
- Determine underlying structure of factors (spatial in 2016 paper) or loadings (dependant of environmental variables in 2017 paper).
I would love to be able to implement that in stan language and exploit the powerful sampler it represents.