WIP: Enhancing inference docs

[springcoil]

Based on #1968 I updated some of the ADVI docs.

Some more work needs done.

[aseyboldt]

I think it makes sense to split this into two pages: One about MCMC, and one about variational methods. We could then start both pages with a few sentences about what they do. I don't know that much about variational methods, but how about something like this (suggestions welcome):

"Variational methods approximate the posterior by looking at a finite dimensional subspaces of the set of possible posterior distributions. They try to find an element of this subspace that is similar to the actual posterior. While they tend to be much faster than Monte Carlo sampling, they are limited by the choice of subspace. Posteriors that are very different from all posteriors in the subset can not be approximated reasonably."

After than a table of the most important functions (not sure myself...):

.. currentmodule:: pymc3.variational

.. autosummary::
   
   fit
   ADVI
   FullRankADVI
   advi 

After that the api of the individual modules.

[aseyboldt]

They aren't in pymc3.variational.advi, but in pymc3.variational.inference, so this doesn't work.

[aseyboldt]

None of the methods here have a docstring, so nothing shows up in the doc.

[twiecki]

I think @ferrine has re-implemented these in OPVI, so we can probably drop SVGD for now.

[springcoil]

Hi, That makes sense - well tomorrow evening when I get time I'll try to add in docstrings. Maybe call this a 'WIP' instead.

…

On Thu, Mar 30, 2017 at 12:08 AM, aseyboldt ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In docs/source/api/inference.rst <#1969 (comment)>: > + +.. currentmodule:: pymc3.variational.operators + +.. automodule:: pymc3.variational.operators + :members: + +OPVI +^^^^ + +.. currentmodule:: pymc3.variational.opvi + +.. automodule:: pymc3.variational.opvi + :members: + +SVGD +^^^^ None of the methods here have a docstring, so nothing shows up in the doc. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1969 (review)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA8DiJNdFaGD908OefFjhLYx5qN64lZiks5rquR5gaJpZM4Mtkfm> .

-- Peadar Coyle Skype: springcoilarch www.twitter.com/springcoil peadarcoyle.wordpress.com

[springcoil]

I'll try to get to this again this weekend. I'll agree that there's some changes and this has to be changed. I was hoping to just get the placeholders in there so docstrings can be added in later.

[junpenglao]

Close in favor of the new VI docs. Thanks of the effort @springcoil

[springcoil]

No problem. I often start stuff and never finish it.

…

On 10 Jul 2017 09:25, "Junpeng Lao" ***@***.***> wrote: Closed #1969 <#1969>. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1969 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA8DiM2naa0kj_GJ-Hfy8ZxmEW12-vNaks5sMd_qgaJpZM4Mtkfm> .

[junpenglao]

I know that feel - There are just too many fun stuff we want to do but also many (less fun) stuff we had to do ;-)