# Stein approximations and MCMC

**URL:** <https://discourse.mc-stan.org/t/stein-approximations-and-mcmc/11844>\
**Category:** Developers\
**Tags:** features\
**Created:** [November 10, 2019, 5:09pm UTC](https://discourse.mc-stan.org/t/stein-approximations-and-mcmc/11844 "2019-11-10T17:09:08Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![bgoodri](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/bgoodri/32/4451_2.png) [@bgoodri](https://discourse.mc-stan.org/u/bgoodri)\
**Post date:** [November 10, 2019, 5:09pm UTC](https://discourse.mc-stan.org/t/stein-approximations-and-mcmc/11844/1 "2019-11-10T17:09:08Z")

</div>

Has anyone looked into

> **[Stein Point Markov Chain Monte Carlo](http://proceedings.mlr.press/v97/chen19b.html)**
>
> An important task in machine learning and statistics is the approximation of a probability measure by an empirical measure supported on a discrete point set. Stein Points are a class of algorithms ...

  

> **[gsd\_ksd-slides.pdf](https://web.stanford.edu/~lmackey/papers/gsd_ksd-slides.pdf)**
>
> 4.19 MB

It seems as if the authors are at least worried about the right things and the method utilizes gradients, so it would be well-suited for Stan. But they weirdly don’t compare their experimental results with Stan’s MCMC or look at the sorts of models where Stan does really well.

---

<div class="post-metadata">

**Author:** ![maxbiostat](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/maxbiostat/32/1326_2.png) [@maxbiostat](https://discourse.mc-stan.org/u/maxbiostat)\
**Post date:** [November 10, 2019, 6:21pm UTC](https://discourse.mc-stan.org/t/stein-approximations-and-mcmc/11844/2 "2019-11-10T18:21:46Z")

</div>

> [@bgoodri](#):
>
> But they weirdly don’t compare their experimental results with Stan’s MCMC or look at the sorts of models where Stan does really well.

In my experience some of the (MCMC) community hasn’t yet caught up with the fact that Stan/dynamic HMC is the gold standard when the answer is not known in closed form – excluding discrete parameter spaces in this discussion.

---

<div class="post-metadata">

**Author:** ![bgoodri](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/bgoodri/32/4451_2.png) [@bgoodri](https://discourse.mc-stan.org/u/bgoodri)\
**Post date:** [November 10, 2019, 6:28pm UTC](https://discourse.mc-stan.org/t/stein-approximations-and-mcmc/11844/3 "2019-11-10T18:28:47Z")

</div>

Yeah, but Mark Girolami knows. Not sure about the other co-authors.

---

<div class="post-metadata">

**Author:** ![anon75146577](https://avatars.discourse-cdn.com/v4/letter/a/f07891/32.png) [@anon75146577](https://discourse.mc-stan.org/u/anon75146577)\
**Post date:** [November 11, 2019, 2:20am UTC](https://discourse.mc-stan.org/t/stein-approximations-and-mcmc/11844/4 "2019-11-11T02:20:35Z")

</div>

I was going to say the code exists but it’s all matlab so it might as well be punch cards buried in a time capsule. I’ll put the paper on my “to read properly” list

---

<div class="post-metadata">

**Author:** ![andrewgelman](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/andrewgelman/32/1577_2.png) [@andrewgelman](https://discourse.mc-stan.org/u/andrewgelman)\
**Post date:** [November 12, 2019, 12:33am UTC](https://discourse.mc-stan.org/t/stein-approximations-and-mcmc/11844/5 "2019-11-12T00:33:20Z")

</div>

I used to work in a lab where they had old data stored on paper tape. At some point lots of the data got lost because the tape got chewed on by mice.