Bob Carpenter

Sorry I missed this the first time around. The reason we've never done this before is because it's not clear with @SteveBronder's example whether it's the columns that have a...

> presumably the internals only care that the number of elements in y and mu vectors/row_vectors match, That's right. I still don't like any of the new proposed signatures because...

> Presumably there's no performance benefits to putting the re-formatting inside multi_normal_*() There's a slight performance benefit to coding directly in `multi_normal` because we could use views (expression templates) rather...

@SteveBronder: Yes, that makes sense in a type-theoretic sense. We (meaning everyone on the project at the time coming to the general Stan meetings) discussed all these signatures at length...

@nhuure: There's already an issue to create a matrix normal (and matrix Student-t) distribution. The idea is to do it with a Cholesky factor parameterization of the covariance matrices (or...

I'm still opposed. See https://github.com/stan-dev/math/issues/2532#issuecomment-1042322064

> @mike-lawrence Actually more user-convenience I'd say. That was my understanding. Performance gains would be minimal given that there's a quadratic form in each multivariate normal eval. > @steve-bronder I...

Thanks for posting an issue, @JackCaster. It looks like there are four distributions called "generalized Pareto": * [Wikipedia link](https://en.wikipedia.org/wiki/Pareto_distribution#Generalized_Pareto_distributions) The type 2 version is implemented in Stan already: * Language...

I'm definitely confused on nomenclature. Any of those densities would be easy to implement in Stan as they all have simple pdfs. I'll edit the issue to reflect the specific...

@JackCaster: the most useful thing for us would be a definitive pointer to the function that you want to be implemented. For instance, is it the definition given in the...