Seth Axen

Thanks! That should be enough to get me started. I'll probably tackle this in a few months if no one else does before then (unless I find some time early).

Working on this now and have some follow-up questions. > Specifically, the `∂A` term is given by an integral over the Legrange multiplier term. Coincidentally, the `phiv` values used in...

Okay, I think I worked something out for forward mode at least. The pushforward of `u = expv(t, A, u_0)` is (using slide 5 of http://www1.maths.leeds.ac.uk/~jitse/scicade09.pdf): `Δu = A \phi_0(tA)...

Following up on @ChrisRackauckas's point, we can indeed compute the adjoint of `A` by solving an ODE in reverse. A working prototype here: https://gist.github.com/sethaxen/4071b401b9b4ff4f5421136cec2fa7da/dd914b79d465d8653b1674cbc466f5a29d95fbae#file-expv_chainrules-jl-L64-L77 I haven't worked out how to...

Another way to compute the adjoint of `A` comes from https://doi.org/10.1109/TAC.1978.1101743. Let `w = expv(t, A, v)`, `Δw` be the adjoint of `w`, and `∂v = expv(t, A', Δw)` be...

Here's where I landed on this. The adjoint for `A` will be computed by hand-deriving the pullback through `exp` and `arnoldi`/`lanczos`, The former will be added to ChainRules (https://github.com/JuliaDiff/ChainRules.jl/issues/331). I...

> Thanks @sethaxen . I love the idea of cutting down boilerplate. In Julia it seems like we wouldn't need separate function names, since it could just dispatch on the...

This PR should probably wait on https://github.com/arviz-devs/arviz/pull/2086 to be finalized first.

This work should be moved to https://github.com/arviz-devs/InferenceObjects.jl

Once you have an `InferenceData` with both `posterior` and `prior` groups, you can just pass this directly to `plot_dist_comparison`, so I think your questions is how to get both groups...