Fostering Stan user communities through domain-specific resource pages

Hi all, my name is Vianey Leos Barajas and I’m a postdoctoral researcher at North Carolina State University, with a primary research interest in topics related to statistical ecology. I’ve used Stan for a few years and have noticed that there’s quite a bit of interest in learning Stan in the community in general (statistical ecology/general ecology).

In order to foster a community of Stan users in the (stat) ecology world, what would you all recommend? Is there the possibility of having a domain-specific Stan resource page? Having a centralized location to submit small case studies or Stan code for models typically used in ecology, along with data examples, would be really beneficial. I know that case studies already exist, but having something more ecology-focused would be great. I can’t speak for others, but I would certainly be willing to spend time forming this community (and I imagine there are others that would do this as well). Having a basic template of how to put together case studies might help others understand how to contribute as well (just a thought).

Other software has made quite a bit of inroads into this community by giving workshops/tutorials at ecology events, the latest being at the wildlife society meeting. One of the ecology PhD students here that attended the conference returned and proclaimed that Nimble was the way to go and much faster than Stan, from what he had learned at a workshop.

From what I’ve seen, there’s a lot of interest and a group of people that are already using Stan heavily. I’d love to be able to foster that community somehow, especially in the months before our major conference next summer (at which Dan Simpson and Andrew MacDonald will be giving a short course): http://www.isec2020.org/training-program#ShortCourses.

Gracias!

17 Likes

Thanks for posting this Vianey!

And taught Stan!

Definitely! There’s actually a spot on the Stan website for just this sort of thing, although to be honest I had forgotten about it until @Bob_Carpenter reminded me recently. It’s the “Specialized Field Guides” section at Stan - Documentation, but the problem is that you wouldn’t know it exists unless you scroll all the way down the page (we definitely need to improve the website at a minimum) and there’s also just one of them. So I think a revitalized effort at this would be great!

5 Likes

While going through the central Stan web/docs as recommended by @jonah is definitely worthwhile, there is a risk we won’t be very flexible with updates as many other things are going on. But if you just go ahead and register www.staninecology.org / ecology-stan.github.io or whatever and maintain a list there (setting up a simple page with blogdown/hugo + GitHub + possibly netlify is a breeze), it would help and linking your list from the Stan page could be done quickly. We would probably eventually want to copy content to official site but such duplication won’t IMHO be a problem - Google will sort it out.

As the community manager here I am definitely open to provide some support for your endeavor.

Best of luck with your effort - communities are hard (and important)!

5 Likes

Excellent idea!

Based on the Nimble marketing material the speed difference is due to the compilation time in Stan. I haven’t seen proper benchmarks for bigger problems where the sampling time dominates. If that PhD student knows some more recent benchmarks results, I would be happy know.

1 Like

It also occurs to me that @lizzieinvancouver has a group (http://temporalecology.org/) that does great work in ecology and uses Stan. Lizzie, do you think anyone from your lab or other people you know in ecology might want to collaborate with @vianeylb on organizing this and/or contribute to an ecology page for Stan?

4 Likes

If that PhD student knows some more recent benchmarks results, I would be happy know.

I’ll ask him and get back to you. I don’t recall he provided specifics about the speed difference.

This is definitely something that we would be interested in collaborating on / contributing to as well! Thanks for raising the topic, @vianeylb .

3 Likes

I would like to see this hapen and help out. Our group uses Stan for our riparian ecology questions here in NM.

3 Likes

Hi!

As @nerutenbeck indicated, we at SilviaTerra use Stan quite a bit for our work in forestry, and would be interested in helping this sort of effort along.

Any thought to a short course or workshop at ESA? I’ve considered the idea in the past, but have felt too intimidated to take it on as a solo endeavor. I’d definitely be interested if it was a group effort, though! The deadline for 2020 meeting proposals is Nov 21st:
https://www.esa.org/saltlake/short-courses/

7 Likes

How great! I’ll keep you all posted as to how this develops. I’m quite excited to learn about these applications as well.

ESA overlaps almost exactly with the Joint Statistical Meetings next year: https://ww2.amstat.org/meetings/jsm/2020/. Perhaps there are others on here that’d be interested, someone from the Stan crew?

I think these efforts are very helpful toward building a solid community of Stan users in ecology.

3 Likes

I would be curious to know if the application where Nimble was faster was state-space related. Nimble has some samplers that are specialized for this class of models. There are a number of things you can do computationally in state-space models that take advantage of the recursive structure. I haven’t looked in any detail at their samplers. but I know we have used variants of the forward filter - backward sampling method that uses a lot of what is known computationally about calculating state-space models. I think Nimble has some variant or particle filters and/or SMC and its variants.

Also I am on the board for WNAR and we are looking to expand our ecological offerings the next meeting (in Alaska). Perhaps we can gin something up (not me to do - but to help organize).

5 Likes

I would also very much like to contribute to a kind of ‘specialized field guide for ecology’! A group of us at MIT, UCSC, and Dalhousie University are preparing a workshop for marine ecological/oceanographic modeling in Stan and are producing a number of case studies and background materials that could go into such a project. Please keep me posted! gbritten@mit.edu

7 Likes

That sounds incredible!! (I have a slight preference for all things marine)

Will definitely keep you posted on any progress. I have a few ideas about how we could move forward on this effort.

Thank you so much for posting and reaching out!

2 Likes

Hi!

Me and @avehtari are working in a spatio-temporal model applied in a specific problem in fisheries sciences. We are not working directly with Stan, we are using the library \texttt{tmbstan} where we can do all the inference as if we had built the model in Stan.

This is a great idea and I would like to participate!

5 Likes

How did I not know tmbstan existed?! This would be fantastic to include. I’ll keep you posted and let you know how things move forward! It’d be great to include your work.

On a slight aside, one of my goals over the next 5-10 years is to try to figure out how to get involved in the statistical ecology community in Latinoamerica (or get it going if it doesn’t really exist). I might reach out to you for this too, if that’s alright! Algo como http://www.isec2020.org/ pero en LA sería espectacular. Se puede soñar… A ver si un día @jonah también se anima a dar una charla de Stan en español.

3 Likes

@cavieresJJ thanks! (And I still want to come back to Chile, so we should talk again soon!)

2 Likes

I’m happy to contribute too – I have used Stan for a variety of occupancy modeling and capture-recapture applications.

In terms of resources that are already available, the translation of the Kery/Schaub BPA models (link to Bob’s blogpost) to Stan could be a nice focal point for a workshop aimed at applied ecologists: https://github.com/stan-dev/example-models/tree/master/BPA

Longer term, if anyone is looking for a research project, and has time to do it, it might also be nice to compare the JAGS and Stan (and NIMBLE?) implementations of the BPA models using simulation based calibration. Ideally they are all providing valid inference…

4 Likes

Unfortunate scheduling there.

If anyone is interested in helping to pull something together, feel free to post here or email me at brian@silviaterra.com. I’m happy to lead the charge.

Cool to see so many different use cases in ecology/natural resources represented in this thread!

2 Likes

@vianeylb \texttt{tmbstan} is a library relatively new but with another platform of modeling (Template Model Builder (TMB)). Our work is in final phase so we hope colaborate soon with this.

with @jonah already discuss the posibility to do a course of Stan in Latinoamerica EN ESPAÑOL … We hope he have time for that… (:

2 Likes

Only a slight preference?!
Let’s stay in touch. We will make the workshop materials public after it’s held in January so I’ll be able to pass you the link. It may be a nice starting place for a broader ‘ecological field guide’ (what a perfect name too… ).

2 Likes