Stan General Meeting, Thursday April 16, 2020 11am EDT

[wds15]

For discussion: should we do a quick change to the user function given to reduce sum before the release?

See the thread on un normalized densities for motivation and rationale.

Can we have unnormalised densities in user functions?

IMO propto=true is fine. It has no effect except when using these undocumented _propto_ functions. Also, an _lpmf function in reduce_sum is obviously a weird hack because the signature doesn’t make sense. It has to be mydensity_lpmf(start| end, slice, ...); but this implies it’s a probability distribution for the start index. For the function to make sense the slice should be the first argument and the indices must come after it. (We could allow both orderings; there’s no ambiguity and the m…