Seth Axen

Some things I don't like about the current interface: - To customize the adaptor, if I end up creating the `StepSizeAdaptor` myself, then I need to also create the integrator...

Thanks, @baggepinnen!

> I'm tempted to just do this. But if you don't mind, I'd _really_ like to improve our bus factor. What to take a shot at it? I can jump...

> It just hit me that we can just hard-code this. I'm treating leading underscores as "private" variables, so we can probably just make the small change > `:($(st.x) =...

I think this is somewhat more involved than the suggestion in https://github.com/TuringLang/Bijectors.jl/issues/134#issuecomment-679096131. I do not think it can be resolved only by changing/adding a bijector. It seems we are missing...

Okay, well in the meantime, I'll try to push things along on the Distributions side so when the DynamicPPL/Bijectors is finished, it's ready to use.

In https://github.com/JuliaStats/Distributions.jl/issues/1336 we're discussing possibly defining an `LKJCholesky` distribution whose support are objects of type `Cholesky`. Would this be problematic for Turing? I'm wondering e.g. how samples would be stored...

I created https://github.com/JuliaStats/PDMats.jl/issues/132 to discuss modifications to PDMats for easy creation of a `PDMat` from a Cholesky factor of a correlation matrix and a vector of standard deviations.

@mgmverburg was able to get something working in https://github.com/mgmverburg/Turing_examples/blob/f639302c5b28ecc30f6b05e90b9f95adf97be027/Utilities/lkjcholesky.jl#L277-L385. Is that too hackish to adapt for this specific distribution? I agree, ideally we have a general solution for distributions whose...

Perhaps for the short term, Turing could include non-exported functions for 1) defining the number of unconstrained variables in an `nxn` correlation matrix 2) mapping a vector of unconstrained variables...