Seth Axen

> @sethaxen Is this now fixed? No. Distributions now has `LKJCholesky`, but it's not yet usable inside Turing. My understanding from previous Slack discussions is that the reasons are: -...

Thanks! I did some reformatting to add checkboxes for tracking the progress.

This is currently how I'm doing it. Images work fine, but I haven't checked that everything else works correctly. https://github.com/sethaxen/sethaxen.com/blob/main/_layout/meta.html

> Do you use or want to use non-Euclidean KE's with DynamicHMC? Yes, I'm writing a package for HMC on manifolds that uses most of DynamicHMC's building blocks as mentioned...

Thanks! Sorry I hadn't replied yet. Shortly after my last comment, I realized that the simple expression of the generalized NUTS criterion on Riemannian manifolds in Betancourt's paper presumes one...

> 1. One option would be to have an optional first argument for the "outer type", the constructor for the type to be returned. One issue here is that if...

> For scalar return values, we probably need to use 0-dimensional arrays or `Ref`s; or if we have a way of detecting measures whose points are scalars, we can implement...

> I feel strongly about [jrevels/YASGuide#13 (comment)](https://github.com/jrevels/YASGuide/issues/13#issue-484616702), otherwise I feel flexible Hmm, I definitely prefer spaces around `*`. Less whitespace makes it much harder for me to read code. Somehow...

In that case, I think it makes sense to have an un-implemented `logtotalmass`, and a default `totalmass(x) = Exp(logtotalmass(x))`. Then, in order to normalize a measure, one would need to...

> For scalar distributions, there are big advantages in MeasureTheory over Distributions for log-density computations. But for random sampling, this is less clear. For now, a lot of this goes...