David Widmann

I would very strongly recommend against such an approach. Removing samples/trajectories from Sobol's method will break any theoretical guarantee as these depend on the structure of the Sobol sequence, see...

I would throw a hard-error (or rather just continue returning `NaN`) for Sobol sequences. > We could just default to more robust QMC samplers that have a clearer convergence on...

> Today I tripped over this: This use of `logpdf` is deprecated: https://github.com/JuliaStats/Distributions.jl/blob/efff906e2e6aad180d0be6dcaa9c98d3c398510d/src/deprecates.jl#L39 > The same confusion of course applies to multidimensional cases like This is not deprecated (yet). The...

> Is this intended? It predates my times as a contributor, and it seems problematic in this case, but I'm quite certain that it is intentional. Having parameters of the...

> triggering a conversion and refactorization of the covariance matrix. Note though that (as intended) it doesn't actually refactorize the matrix, it only converts the underlying factor of the `Cholesky`...

What would you expect? `map(params, d.dists)`? Personally, I basically don't use `params`. But maybe someone else is interested in that functionality. We should just make sure it's a reasonable definition...

> It looks like you all defined a custom rule for tapir/etc for find_alpha for it, which I presume is a reasonable place for a custom rule for it here...

> early versions it could be loaded with Requires. Could we just make it a proper dependency on older Julia versions? It's still possible to make it a weak dependency...

It seems you also implement chainstack and bundle samples? That should rather be an extension of AbstractMCMC, I assume?

It's a weak dependency on Julia >= 1.9 if you declare it both as a strong dependency and a weak dependency. See https://pkgdocs.julialang.org/dev/creating-packages/#Transition-from-normal-dependency-to-extension.