Move SFGS, CSG and Elliptical Slice samplers to pymc3-experiments

[twiecki]

These three samplers are not well supported and do not appear to really be used. They are mostly experimental and should be in the main software. Thus, they should be moved to https://github.com/pymc-devs/pymc3-experimental alongside any tests and docs (tests and docs do not need to compile/function). If someone picks them up and demonstrates their usefulness they can certainly make their way back, but they have lied dormant for too long.

[Akamoha]

So this would mean removing all the work done in #2544?

[twiecki]

@Akamoha Yes, alongside the SFGS.

[eigenfoo]

#3233 and #3357 are both problems to do with the CSG sampler. Is the plan to simply remove CSG altogether? If so, then perhaps we should close those issues.

[junpenglao]

Could we move to a new repo maybe? (pymc-devs/experiment?) Along with Elliptical Slice Sampling.

[twiecki]

@junpenglao I like that.

[twiecki]

I created https://github.com/pymc-devs/pymc3-experimental.

[junpenglao]

Awesome ;-)

[junpenglao]

I can finally dump all my codes on Laplace approximation etc there :-|||||

[twiecki]

@Akamoha Do you want to take this one on?

[harshitbansal05]

hey, @twiecki can I work on this one?

[twiecki]

Yes! thank you for volunteering.

…

On Fri, Feb 1, 2019, 06:17 Harshit Bansal ***@***.*** wrote: hey, @twiecki <https://github.com/twiecki> can I work on this one? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3353 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AApJmLghZNDBYGWve3msOZGPvrIQ4KqRks5vJCJcgaJpZM4aQrlC> .

[fonnesbeck]

ESS should not be experimental, and is useful for GPs. One of the demo notebooks uses it.

[ferrine]

I remember @ColCarroll was working on this https://github.com/ColCarroll/l2hmc_pymc3
Its worth potring as well

[junpenglao]

@fonnesbeck In that case maybe we should have some more example of using ESS - eg how it is used with other Samplers

[ColCarroll]

@ferrine just added a license so it can be copied/moved with or without my permission (whoever wants to move it has my permission, but prominent acknowledgments to the authors should stay close to the code!) I can also move it... soon.

[vbsinha]

Hi @twiecki, Can I work on this issue?
From the previous comments, it is not clear to me if I should move Elliptical Slice as well. Could you please clarify that? I did a quick grep for EllipticalSlice and found that only one notebook uses it in the entire repo. It is nowhere else mentioned apart from its definition and tests.
Moreover, should I simply move the classes SFGS, CSG and Elliptical Slice to pymc3-experimental? I mean that standalone classes in the experimental repo won't make much sense. So do I move all the other python files involved with these samplers to pymc3-experimental repo?

[twiecki]

@vbsinha Definitely. You can leave Elliptical Slice here.

To the extend that whole files can be moved that's what I'd do. Ideally we can still retain the functionality in that repo, it should just require redirecting imports to pymc3 instead of relative imports.

[vbsinha]

@twiecki It seems that these notebooks (this and this) are using SGFS while this one is using CSG. Could you please suggest how should I proceed? Should we move these notebooks to pymc3-experimental repo as well or leave SGFS and CSG intact in this repo?

[twiecki]

The NBs should be ported too. Basically the idea is that there is no mention of SGFS or CSG in pymc3, but if someone wants to try them, they can install pymc3-experimental.

[junpenglao]

Would there be a way to index the doc on the website still? Maybe under a tab call Experimental?
Basically, it would show how to install the experimental feature, which NB example of how to use some of them.

[canyon289]

@junpenglao
I think(?) all you need to do is add an "Experimental api" header to the api.rst file and then add the relevant modules below

https://github.com/pymc-devs/pymc3/blob/master/docs/source/api.rst

[twiecki]

Closed by #3391.