Re: Marginal maximum likelihood in Stan

[Andrew Gelman]

Hi, Avi. We are planning to have this in Stan but it’s waiting on the current refactor which will include higher-order autodiffs. For now the best option is full Bayes.
A

> On Aug 4, 2015, at 5:25 PM, Avi Feller <avi.f...@gmail.com> wrote:
>
> Hey Andy,
>
> A friend of mine was trying to work through a problem that's essentially marginal maximum likelihood in a multilevel model. I know this was something that you were eventually planning to implement in Stan. Any idea what the timing is like for that?
>
> Thanks,
> Avi

[Bob Carpenter]

The higher-order autodiffs are not actually part of the
command-structure refactor that we've been talking about.
The point of this particular refactoring is to simplify, make
more consistent, and make more reliable, the interfaces for
RStan and PyStan and CmdStan by bringing functionality
into Stan C++ that can be reused. The reason we don't want
to pile in changes to commands before the refactor is that
we know we'll have to immediately rewrite anything we do now
and that it'll just push back the completion date.

My understanding is that Rob's finished the pull requests
for higher-order autodiff so we should have at least an experimental
second-order soon (it still needs way more testing). Then all
we need is someone with time to implement MML.

- Bob

> --
> You received this message because you are subscribed to the Google Groups "Stan users mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-users+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.
>
>