Lightnews — Scholar-powered news

Sean Pinkney

@spinkney.bsky.social

930 followers 590 following 80 posts

I mostly post about probabilistic programming stuff, statistics, and R/Julia/Python (in that order). I'm a volunteer Stan developer and citizen scientist (papers on arxiv). Currently my day job is doing marketing analytics.

Posts Media Videos Starter Packs

Sean Pinkney @spinkney.bsky.social · Jun 23

I'll take one if you have any left

Sean Pinkney @spinkney.bsky.social · Jun 4

Which base measure are we in?

Sean Pinkney @spinkney.bsky.social · Jun 4

So go re-run your models with simplex and report back with your results!

Sean Pinkney @spinkney.bsky.social · Jun 4

and to be comparable, if not a bit better, in all other cases.

What has changed is that we use the sum-to-zero vector transform and then softmax it. There are still n-1 free variables but the over/underflow for finite precision computing is less likely. 2/

1 2

Sean Pinkney @spinkney.bsky.social · Jun 4

Stan 2.37 will be out sometime next week. I helped get the updated simplex parameterization and sum-to-zero matrix (both rows and columns sum to zero).

We found the new simplex parameterization to be much better in the extremes of the simplex and across very small alpha values of the dirichlet 1/

1 7

Sean Pinkney @spinkney.bsky.social · Jun 2

Ask me any questions you may have about this! #stats

Pierre-Simon Laplace @learnbayesstats.bsky.social · May 30

🎙️ Ep. 133 is out now!

@alex-andorra.bsky.social chats with ‪ @spinkney.bsky.social
& Adrian Seyboldt about making Bayesian models more efficient without losing rigor — zero-sum constraints, Cholesky tricks, practical wins & more

🎧 learnbayesstats.com/episode/133-...

#Bayesianstats #podcast #LBS

Learning Bayesian Statistics – Laplace to be for new & veteran Bayesians alike!

Laplace to be for new & veteran Bayesians alike!

learnbayesstats.com

7 30

Sean Pinkney @spinkney.bsky.social · Jun 1

Sean Pinkney @spinkney.bsky.social · May 24

This is the pre-pre-print of the meta-meta-analysis

Sean Pinkney @spinkney.bsky.social · May 7

I've come on here after months away to let everyone know that I apparently really like trance music

Sean Pinkney @spinkney.bsky.social · Mar 29

Oh I first read this as the water cycle is changing as a good thing rather than a permanent thing

1 6

Sean Pinkney @spinkney.bsky.social · Mar 27

Hi!

Sean Pinkney @spinkney.bsky.social · Mar 27

Are there any other statistics nerds in the NoVA area that want to get together?

1 3

Sean Pinkney @spinkney.bsky.social · Feb 28

😈

Sean Pinkney @spinkney.bsky.social · Feb 28

Promising results from the Pymc folks! #bayesian #pymc #stats #stan

discourse.mc-stan.org/t/updated-ch...

Updated cholesky corr parameterization testing

Promising results from the Pymc folks Confirmation that the det Jacobian is correct and early tests show improved sampling!

discourse.mc-stan.org

Sean Pinkney @spinkney.bsky.social · Feb 23

Hi I don't know you, you don't know me but I have the same feels. Come say hi sometime to me

Sean Pinkney @spinkney.bsky.social · Feb 20

Please let me know! I'd love to see timing and ESS comparisons

Sean Pinkney @spinkney.bsky.social · Feb 20

It's similar to what Adam Haber came up with for tfp but it seems to be slightly better in small dimensions and much faster in large (100+) though this could be because of Stan's implementation vs tfp.

Sean Pinkney @spinkney.bsky.social · Feb 20

Fellow probabilistic programmers I came up with a parameterization for a Cholesky factor of correlation matrices that looks good. I'm asking to test and see if it works well for you discourse.mc-stan.org/t/updated-ch...

#rstats #statistics #bayesian

Updated cholesky corr parameterization testing

Hi everyone, I’m asking if people could test out this parameterization of the Cholesky factor of correlation matrices? From my tests on a macpro m1, it looks to be much faster and I can easily sample...

discourse.mc-stan.org

2 6 27

Sean Pinkney @spinkney.bsky.social · Feb 17

I don't know wtf this is but I'm super into matrices so the non-matrix theology is a non-starter for me

Sean Pinkney @spinkney.bsky.social · Feb 3

I will only be impressed by you if you solve a long standing mathematical conjecture and win the Tour de France within 2 years of each other. Otherwise, what are you even doing?

Sean Pinkney @spinkney.bsky.social · Jan 31

Basically he Colin from What We Do in The Shadows

Jason Koebler @jasonkoebler.bsky.social · Jan 29

some really great tips here:

“Insist on doing everything through ‘channels.’ Never permit short-cuts to be taken in order to expedite decisions.”
“Make ‘speeches.’ Talk as frequently as possible and at great length. Illustrate your ‘points’ by long anecdotes and accounts of personal experiences. Never hesitate to make a few appropriate ‘patriotic’ comments.”
“Bring up irrelevant issues as frequently as possible.”
“Haggle over precise wordings of communications, minutes, resolutions.”
“‘Misunderstand’ orders. Ask endless questions or engage in long correspondence about such orders. Quibble over them when you can.”
“In making work assignments, always sign out the unimportant jobs first. See that the important jobs are assigned to inefficient workers of poor machines.”
“To lower morale and with it, production, be pleasant to inefficient workers; give them undeserved promotions. Discriminate against efficient workers; complain unjustly about their work.”

Sean Pinkney @spinkney.bsky.social · Jan 13

Congrats!

Sean Pinkney @spinkney.bsky.social · Jan 11

Functions are encouraged! You can even write functions in a file with .stanfunctions ending and include it mc-stan.org/docs/stan-us...

Stan Program Style Guide

mc-stan.org

1 3

Sean Pinkney @spinkney.bsky.social · Jan 2

"Painting to be stepped on." Do not step on the painting!

Sean Pinkney @spinkney.bsky.social · Jan 2

Reminds me of when I tried salvia in college. I immediately ran into the bathroom and started splashing water on my face to stop feeling like I was going to die. Is that stuff still legal? I do not recommend