Adrian Seyboldt

Maths and statistics, #PyMC core developer

[Adrian Seyboldt @aseyboldt]

@AmtrakAlerts We just called the hotline, and were told that the stop in Fraser will be skipped the second day in a row (on Thursday and Friday). Which one is it? Will the train stop here tomorrow or not?

[Adrian Seyboldt @aseyboldt]

@AmtrakAlerts We'd also appreciate an update on the situation. We'd be getting on in Winter Park, and got contradictory information. Will the train be running at all today? Will it stop in Winter Park? And if we can't leave here today, what are the chances that the train tomorrow will?

[Adrian Seyboldt @aseyboldt]

@tomaspueyo @CJHandmer I just wonder what that would look like if you add some uncertainty estimates to that graph...

[Adrian Seyboldt @aseyboldt]

Thank you for having me, it was fun!

[Pierre-Simon Laplace @LearnBayesStats]

we're kicking the New Year off with a very #Bayesian episode 🎆
#NUTS 🌰sampling, a new #ZeroSumNormal #distribution, #PyMC, the Bayesian workflow and more is all covered in episode 74 with @aseyboldt
@pymc_devs #BayesianStats #modelling

https://t.co/7IbIul8s6L

[Adrian Seyboldt @aseyboldt]

So close...

[Adrian Seyboldt @aseyboldt]

@remilouf As to the PR: I wasn't involved with that, but as far as I know the author of the PR wanted to use this in PyMC and asked for help getting it into pytensor so he can use it there.

[Adrian Seyboldt @aseyboldt]

@remilouf Can you point me to a commit that you think has incorrect author information? I couldn't find any, but I'd make sure to fix that.

[The Innovation | Medicine @Innov_Medicine]

DNA to RNA real-time speed. Gene Transcription at real-time speed. Transcription is the first step in gene expression.

[Edward Snowden @Snowden]

Genuine sympathy for Biden looking at this line on the teleprompter and being like "I'm sorry, what? 'The dark forces that thirst for power?' I thought we fired that speechwriter."

[Adrian Seyboldt @aseyboldt]

@adbreind @AustinRochford @twiecki @ferrine96 @pymc_devs We also have almost no python overhead during sampling, and because the parallelization is done in rust we also don't need to pickle anything.

[Adrian Seyboldt @aseyboldt]

@adbreind @AustinRochford @twiecki @ferrine96 @pymc_devs Also, we use a different aesara backend, which also tends to produce faster gradient evaluations (but still has a few missing features in places that we are still working on...)

[Adrian Seyboldt @aseyboldt]

@m_elantkowski Yeah, I had a few similar problems in nutpie. You can store pointers and such in numpy record arrays though, and then pass a pointer to that data as void ptr. That way you can at least avoid hardcoding the addresses.

[Alexandre Andorra @alex_andorra]

Wait, is pm.ZeroSumNormal merged?? 🤯
What a wonderful group effort involving the whole @pymc_labs team, especially @aseyboldt @lucianopaz47 and @RicardoV944 !

[WikiLeaks @wikileaks]

"This should shock and concern everyone..this decision is a grave threat to freedom of speech, not just for Julian but for every journalist and editor and media worker in this country" Assange lawyer @suigenerisjen #FreeAssangeNOW

[Adrian Seyboldt @aseyboldt]

@romanhistory1 Don't forget Herculaneum. I thought it was even more interesting in many ways.

[Adrian Seyboldt @aseyboldt]

@ChadScherrer @bvdmitri Getting good cov estimates is tricky in itself though if there are many parameters, so this didn't work all that well. If you have better ideas I'd love to hear them though.

[Adrian Seyboldt @aseyboldt]

@ChadScherrer @bvdmitri I also looked at the covariance structure by computing the eigenvalue decomp. of one cov estimate relative to the other (generalized eigenvalue decomposition). The eigenvectors of the largest and smallest eigenvalues should then indicate where things are different.