Nelson Tang

Scaling Probabilistic Models with Variational Inference YouTube video by PyData

Here is the recording of my talk

PyData Berlin 2025: Introduction to Stochastic Variational Inference with NumPyro

Notebook: juanitorduz.github.io/intro_svi/

youtu.be/wG0no-mUMf0?...

#pydata #berlin #bayes

A new 246-pages book on MCMC.

"Finite Markov chains and Monte-Carlo Methods: An Undergraduate Introduction"

This is a free textbook suitable for a one-semester course on Markov chains, covering basics of finite-state chains, many classical models, asymptotic behavior and mixing times,

A Gaussian process showing that the allowed time series are forced to be compatible with data

I’m especially proud of this article I wrote about Gaussian Processes for the Recast blog! 🥳

GPs are super interesting, but it’s not easy to wrap your head around them at first 🤔

This is a medium level (more intuition than math) introduction to GPs for time series.

getrecast.com/gaussian-pro...

A new Python edition of "Forecasting: Principles and Practice" is now available online at otexts.com/fpppy/. Thanks to @azulgarza.bsky.social, Cristian Challu, Max Mergenthaler, Kin Olivares & Nixtla for making this happen. #forecasting #python

A Visual Introduction to Hierarchical Models A visual explanation of multi-level modeling

Need to explain (or understand) linear mixed effects regressions, random intercepts, and random slopes? Look no further than "A Visual Introduction to Hierarchical Models" by Michael Freeman, 2017. It's a banger!

(PDF) Stein's Paradox in Statistics PDF | On May 1, 1977, Bradley Efron and others published Stein's Paradox in Statistics | Find, read and cite all the research you need on ResearchGate

With all the bad shit going on, I'm trying to spend more time reading science and math and less doom-scrolling.

Today's diversion was an exploration of the James-Stein estimator, wherein we can get a better joint estimate of the means of three variables than by taking the mean of each variable.

This has been updated to v2.

arxiv.org/abs/2412.052...

Curious about this too, I don’t see much here but might be because of chronological feed or I’m just not following the right people

That entry level job market is going to be sobering

Couldn’t agree more - only poor options to be found here unfortunately. Hoping improvements are coming, but until then it’s a lot of long breaks away from the phone

Have you considered the OnlyPosts feed?

Yeah I think pyenv did that for me when I used it, it lets you pick a python version and set a global default (or manually activate a different one). If you haven’t nuked your computer yet, you can at least use ‘uv python list’ to see what’s installed and do some cleanups

What are you trying to do? uv kind of assumes you want to use venv for stuff and know how to activate those or use an IDE that can detect it. If you want some global python install then you have stuff like conda and pyenv

This whole time I’ve been learning Bayesian inference to detect biased coins and it turns out you can just look at it instead

You can run R in quarto and source() what you need I think

What book is this? The only other time I've seen the 'talent/skill tree' view is in Mathematics for Machine Learning (mml-book.github.io) and I wish we saw more of that view!

I guess I don't have a specific text I can share - you probably already have this, but the chapter in probml2 (Probabilistic ML - Advanced Topics) by Kevin Murphy that covers them as an intro to SSM is my other go-to probml.github.io/pml-book/boo...

And I like this paper, discusses HMM and state space models and Bayesian networks - mlg.eng.cam.ac.uk/zoubin/paper...

Here’s an interactive intro: nipunbatra.github.io/hmm/

Time to reinstall R

It’s important to budget time for yak shaving

Now you too can refine macrodata! lumon-industries.com

The quiet posters feed seems to be the best bet, otherwise everything just gets drowned out with people reacting to news

Friendly reminder, don't forget to prune your uv cache

Maybe an initial assignment could be around critiquing the LLM output based on something the student knows well or is an expert in. Like how you listen to certain podcast hosts and once you hear them talking about your field of expertise you realize they have no idea what they’re talking about

I wonder if, in addition to learning about a specific domain and their problems, you could teach students to critique a LLM’s output?

The analysts are motivated because they have immediate problems and these tools provide solutions and they learn skills at the same time. I don’t think it absolves them entirely of having to learn a little coding but LLM as coding mentor saves a ton of time for instructors

I don’t know what it would look like prior to entering the workforce, but arming working traditional excel-based analysts with the ability to solve their everyday problems with these new tools (in this case, writing Python with LLM help) works well.

time for a hierarchical model?