BRMS tutorial: add rstan example

[falkmielke]

Description

This PR adds an example of rstan usage with a raw Stan model to the existing brms tutorial.
I also rearranged that tutorial a bit (separate section for data).

I was able to test the Rmd code, all runs as intended. However, I did not knit or publish any of this to html, the structure and publication procedure of this repo seem too sophisticated for me (though great that it is based on hugo ssg!). I hope it will not crash GHA.

Thank you in advance for review and suggestions!

Task list

• My tutorial or article is placed in a subfolder of tutorials/content
• The filename of my tutorial or article is index.md. In case of an Rmarkdown tutorial I have knitted my index.Rmd to index.md (both files are pushed to the repo).
• I have included tags in the YAML header (see the tags listed in the tutorials website side bar for tags that have been used before)
• I have added categories to the YAML header and my category tags are from the list of category tags
• I was not able to compile the tutorial into html files, and would appreciate if you original authors and tutorial maintainers could take over.

[wlangera]

Hi Falk

Very nice addition.

I checked the Dutch version. I will assume the same comments apply for the English version.
General remarks:

• Could you keep the term burnin instead of warmup. I like burnin more.
• Any particular reason you provide the poisson_model.stan file in brms_eng folder and not in the brms_nl folder?
• I think I would like to put your section ("Deep Dive: rstan") in a separate section between "Vergelijken met frequentist statistics" and "Referenties". I think the use of rstan is quite advanced for the general users in INBO. Therefore, I think it is better that it is not in the middle of our tutorial, but behind it. People get a view of the whole process of bayesian statistisc (in brms), and if they want (or if they heard about stan somewhere and want to look into it), they can progress further to your section. What do you think?

[wlangera]

wat bedoel je met "een model beveiligen"?

[falkmielke]

This was a translation fault of a faulty English sentence, which on top of that I overlooked in proofreading. My apologies! It is corrected.

[wlangera]

Ik zou hier ook zelf een burn-in gebruiken zoals we in de brms-voorbeelden hebben gedaan. Dan is het nog meer gelijkaardig als hierboven.

[falkmielke]

Thank you for pointing this out, @wlangera !
I briefly asked @hansvancalster and googled "burn-in tuning warmup Bayesian stats". There is a to-the-point discussion here. My notion of changing came from the fact that we don't physically "burn" or "burn in" anything; thus I was surprised that apparently the term was considered appropriate for some MCMC algorithms. "Tuning", my actual favorite 🤘, is reserved for a certain sub-part of the warm-up phase (Merci, Hans!).

Anyways, the experts seem to prefer "warm-up", and if you agree and confirm, I will in turn adjust this throughout the tutorial.

[wlangera]

Ok that sounds good. Thanks

[wlangera]

From looking at the content/tutorials/r_brms/index.md file, I think that the html's should be rendered locally and commited to the repo in this folder.

[falkmielke]

From looking at the content/tutorials/r_brms/index.md file, I think that the html's should be rendered locally and commited to the repo in this folder.

Thanks, I will try that, and we see where checks lead us.

[falkmielke]

• Any particular reason you provide the poisson_model.stan file in brms_eng folder and not in the brms_nl folder?

Good point; I was thinking about git / data redundancy.
Thinking one step further:

• should they rather be in the [...]/html folder, close to what people actually read?
• ... or shouldn't they best be in a top level /data folder, or in /inst, to be generally accessible?
  Thank you for advice @wlangera @hansvancalster ! I will move the file and adjust links.

[hansvancalster]

• Any particular reason you provide the poisson_model.stan file in brms_eng folder and not in the brms_nl folder?

Good point; I was thinking about git / data redundancy. Thinking one step further:

• should they rather be in the [...]/html folder, close to what people actually read?
• ... or shouldn't they best be in a top level /data folder, or in /inst, to be generally accessible?
  Thank you for advice @wlangera @hansvancalster ! I will move the file and adjust links.

I would put it in the base folder of your tutorial: content/tutorials/r_brms/ or equally fine in a subfolder of that: content/tutorials/r_brms/<subfolder>. The name for the <subfolder> perhaps model, model_files, stan, stan_files or ...?

[falkmielke]

I incorporated the change suggestions and re-knitted the html reports. This also yielded two image-related issues, here and here.

• the cmdstanr package is explicitly mentioned (suggestion by @hansvancalster )
• general code files are now in a general folder
• knit with low res images

Thank you for re-checking and publishing!

[wlangera]

Hi Falk, I have send you the images via mail.

Thanks for the changes