intro to Stan for mathy programmers?

[David]

Hi everyone,

I'll be speaking with a group of people at work about Stan in a week,
and I want to suggest some reading for them (say, somewhere from 5 to
40 pages). Can anyone help me out?

All of these people are software engineers who are also strong in
machine learning. They have some basic stats knowledge but e.g. are
likely vaguely (but not deeply) familiar with MCMC.

The ideal reading would give them a sense for why and how someone
would use Stan, and also enough of a sense for the algorithms involved
to feel like they basically know what's going on. I'd also love to
give them (and get for myself) a sense of the various software design
designs involved in building Stan.

At this point, I'm thinking of:

1. Some basic intro to MCMC
2. The NUTS paper:
http://www.stat.columbia.edu/~gelman/research/published/nuts.pdf.
3. Some selections from the STAN manual.

Any suggestions?

Thanks!
David

--
David J. Chudzicki
http://davidchudzicki.com
dch...@gmail.com
(518) 366-7303

[Ben Goodrich]

On Sunday, April 10, 2016 at 8:59:05 AM UTC-4, David Chudzicki (dch...@gmail.com) wrote:

At this point, I'm thinking of:

1. Some basic intro to MCMC
2. The NUTS paper:
http://www.stat.columbia.edu/~gelman/research/published/nuts.pdf.
3. Some selections from the STAN manual.

Any suggestions?

http://web.archive.org/web/20151112194115/http://www3.nd.edu/~mclark19/learn/IntroBayes.pdf
http://utstat.toronto.edu/craiu/Papers/final_ARS.pdf

[Andrew Gelman]

Hi, I’ll channel Bob and remind you to call it Stan rather than STAN.
Beyond this, the Nuts paper is fine but I think it has more detail than is really relevant for most purposes. On the other hand, it’s written in a CS style so that could be good for your audience, I dunno.

We also have 2 short papers on Stan that I highly recommend:
http://www.stat.columbia.edu/~gelman/research/published/stan_jebs_2.pdf
http://www.stat.columbia.edu/~gelman/research/published/Stan-paper-aug-2015.pdf
Despite their nearly identical titles, the two papers have almost zero overlap, as one was written by me and the other was written by Bob.

See you
A

> --
> 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.
> To post to this group, send email to stan-...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Bob Carpenter]

That's a lot to introduce!

I have slides that took about six hours in lecture
that cover most of that:

http://mc-stan.org/workshops/nyc/

The manual has an intro to MCMC with most of the math
you need. Then there's the Stan paper and intro(s) to the
manual that give a high-level view on how Stan works.
All are cited in our doc:

http://mc-stan.org/documentation/

You really want to read Michael Betancourt's arXiv papers
for info on HMC itself. There's also Radford Neal's
intro in the MCMC Handbook (it's a free chapter, as is
Geyer's intro to MCMC).

- Bob

> On Apr 10, 2016, at 8:58 AM, David <dch...@gmail.com> wrote:
>

[James Savage]

On the "why would you use Stan" side, the case studies are excellent. Interesting problems, lots of hand-holding, good discussion of the statistics, etc. 

http://mc-stan.org/documentation/case-studies.html

On HMC, the Stan manual and the Betancourt and Girolami paper are very straightforward. Michael's video (from 57') made HMC extremely clear for me https://www.youtube.com/watch?v=pHsuIaPbNbY

A bit self-promoty, but I've got a short tutorial on what I call "Bayesian Workflow" (writing out the probability model, simulating with known parameters, estimating the model on simulated data, model comparison, inference/posterior prediction etc.). It doesn't get into the technical detail of how the model is fit, but the idea is to get folks excited about why they should learn that stuff. 

http://nbviewer.jupyter.org/github/khakieconomics/intro_to_stan_for_econs/blob/master/IntroToStan_basics_workflow.ipynb

Hope this helps,

Jim

[Bob Carpenter]

I'll put in a second vote for Jim's Jupyter notebook.

- Bob