Kaan Öcal

There's a subpackage called `Adaptation` in `AdvancedHMC` - for now we could probably borrow code from there (this should be permitted by the BSD-3 and MIT Licences). So we could...

> How about moving the AdvancedHMC mass matrix adaption code into `AbstractMCMC.Adaption` submodule? It might make sense to define an abstract `Adaptation` interface in `AbstractMCMC`. As far as I know...

I created a new branch `adaptive` (available on my GitHub) which implements a more generic interface (I don't want to spam PRs). Main changes: - `MetropolisHastings` has a second field...

Not really to be honest! I tried to see if it would be cleaner but at least my new "solution" is more complicated than what we have already. Unless you...

I have uploaded a new version that uses Welford's algorithm. The code now more or less covers the `AdaptiveMvNormal` features in #39 . The current adaptation interface is via `trackstep!`...

> Cool. I like it. I've turned on the test suite so we can see how the tests run. There was an error due to my using a different version...

@arzwa We should also incorporate your univariate adaptive proposal. Any suggestions for a name? We can also change `AMProposal`.

One issue with this is that `propose` returns a `Transition` object but acceptance is determined only in `AbstractMCMC.step`. While I am not sure about backwards compatibility one could rename this...

Additionally one could allow for user-defined Transition typed by templating some functions (eg. `propose`), but I'm not sure if this would be worth the effort.

It does, indeed! Thanks for the hint.