juanitorduz

I got mail! I can’t not wait @vincentab.bsky.social I’ll try to do many of these examples by “hand” (learning by doing).

💯

7 reasons to use Bayesian inference!
statmodeling.stat.columbia.edu/2025/10/11/7...

IMO 80% data science problem in the industry can be solved with a (good!) linear regression (I also consider GLM as just regressions with a link function)

Exited about notebooks in 2026 🚀

I have not tried this myself but this great blog (and the corresponding GitHub repository) might be helpful florianwilhelm.info/2020/10/baye...

Thank you @patrickdoupe.bsky.social

It was fun (painful 😅) to implement VAR(p) models from scratch juanitorduz.github.io/var_numpyro/

Festival der Riesendrachen
#Berlin

I'm learning causal inference “on the street” 😄

Reproducing Uber's Alternating Direction Method of Multipliers (ADMM) based automated budget allocation system in JAX

gstechschulte.github.io/posts/2025-0...

Amazing! Thank you !!!

Hehe classic.

💯

I went into the notebooks in detail and looks super interesting! I have never tried myself but now I’m very curious 🧐

Yay!

Kudos to @sethaxen.com for implementing the Pyro wrapper that makes this possible (shipped in sbi v0.25)!

And thanks to @juanitorduz.bsky.social sharing the cookie factory example—it's a great accessible example for hierarchical inference.

Everything runs in Colab 📊

Cool! Any plans to add NumPyro support? 😄

A nice primer on normalizing flows by PyMC/PyTensor devs Ricardo and Jesse. pytensor.readthedocs.io/en/latest/ga...

Here are the materials for the PyData Berlin 2025 talk on Stochastic Variational Inference with NumPyro:

- Slides: juanitorduz.github.io/html/intro_s...
- Notebook; juanitorduz.github.io/intro_svi/

Creating a LogStudentT distribution Hi everyone, What is the suggested way to use a LogStudentT variable, a la LogNormal? Is there a more straightforward way than implementing a new Distribution? The use-case is the following. I’d lik...

I think you can try with CustomDist. Here is an example discourse.pymc.io/t/creating-a...

If you need support from the (JAX) community I suggest having some “good first issues” on the repo :)

Great! There is some “related” work with JAX

- github.com/py-econometr...
- github.com/py-econometr...

Yes 100% (easy to iterate with LLMs)! I found many bugs in dat pre-processing steps.

Thanks for the resources! I had the same wish :)

Nice! I’ll join :)

Way (way) faster! Conda is, for all practical applications, infeasible. Try micromamba for example. Very similar api (you can keep the environment.yml) and faster :)

Or even micromamba mamba.readthedocs.io/en/latest/us...

I feel you, man, I really do.

This goes to this week’s reading list 💪