ksachdeva

@docmilanfar For me first understanding SSM (the model) and then Bayesian view on filtering and seeing Kalman filter an “algo” for LGSSM (model) was useful. I explain it here - youtu.be/-DiZGpAh7T4?si…

The Kalman Filter was once a core topic in EECS curricula. Given it's relevance to ML, RL, Ctrl/Robotics, I'm surprised that most researchers don't know much about it, and many papers just rediscover it. KF seems messy & complicated, but the intuition behind it is invaluable 1/4

@EikoFried @rlmcelreath Indeed, initially this single term created lot of challenges for me towardsdatascience.com/but-what-is-a-…

@rlmcelreath I still struggle w “random variable”. It’s such a stupid term.

There is no scientific context in which the term “random” helps to clarify. It is a devil word.

An in-depth explanation of challenges and one potential remedy for gradient estimation in stochastic computational graphs youtu.be/nKM9875PVtU

Variational Autoencoder, Latent Variable Model & Amortized Inference... youtu.be/h9kWaQQloPk

What is Probabilistic Programming? + A quick & opinionated review of @mcmc_stan @pymc_devs @PyroAi @TuringLang @TensorFlow Probability (TFP) youtu.be/-C8PGgnF1yg using an example from @rlmcelreath's Statistical Rethinking book

The Metropolis Algorithm. Watch the full tutorial (The Metropolis-Hastings Algorithm) on youtube (youtube.com/watch?v=oX2wIG…)

@BecomingDataSci A series of YouTube tutorials on Bayesian regression & inference. So far 9 videos and many more to come. youtube.com/playlist?list=… . An attempt to cover both the intuition and underlying mathematics. 🙏

@rlmcelreath .. students admire & love you back for all the wisdom you grant.

Things I do not do often enough: 1. inflate my bicycle tires 2. descale my de'longhi 3. call my mom 4. remind you that I made 20 hours of free bayes stats (really anti-stats) lectures because i love you #calendar--topical-outline" target="_blank" rel="nofollow noopener">github.com/rmcelreath/sta…

Published a tutorial on Importance Sampling youtu.be/ivBtpzHcvpg

A visual introduction to Rejection Sampling! Full tutorial on YouTube youtu.be/si76S7QqxTU

Published a video explaining "MADE: Masked Autoencoder for Distribution Estimation" youtu.be/7q4ueFiJjAY MADE is an essential component in many modern Normalizing Flows algorithms

Published a comprehensive tutorial on Normalizing Flows - Motivations, The Big Idea, & Essential Foundations youtu.be/IuXU2dBOJyw

@jeremyphoward The complexity/Simplicity of a language is a function of how it is taught and the domain for which it is designed! E.g. Your fastai effort makes the complex subject of ML relatively simpler. So what is the problem - language or the teachers :)

Are there any universities that do their intro coding courses in #golang? I know that Python is very popular, but it's a huge, complex, and (delightfully!) idiosyncratic language. Go, OTOH, is small, simple, and (delightfully!) boring.

@willkurt IMHO reason great libraries got written in python are because of domain expertise & not because language is superior. Refactoring/updates/performance/lack of tests are real issues in any industrial code & python makes my heart stop because of lack of strong typing and tools.

@ksachdeva My concern is precisely with the idea dynamic typing is a flaw that "needs bandaging". It is an important language feature. You make not care about the language, but it is because of those merits that so many great scientific libraries are written in dynamically typed languages.

I will always feel ambivalent about types in Python. I do think they are a useful documentation tool, but dynamic typing makes a language fundamentally different than static typing. If you want strong, static typing choose one of the many languages that support this by design.

@smdiehl @willkurt If you take inspiration from JavaScript community then it can be done. Many JavaScript libraries did manage to have type definitions. It is about the community effort... seems most of the python developers still do not appreciate the importance of type annotations.

@willkurt Not surprising that bolting a very limited type system on top of a overtly dynamic language doesn't really pay dividends in terms of developer productivity. I don't even think it's possible to add types to a library like Django or Pandas without crazy research.

But what is Entropy? by @ksachdeva buff.ly/2DxeHYK

If you are hearing about #covid19 “reinfections” in Asia, I can offer you my take as a virologist. The best explanation for what we’re seeing is likely due to three things...

@ozataman @rlmcelreath @fehiepsi @junpenglao Yes there is one for pyro (1st Ed) and numpyro (2nd Ed) by @fehiepsi ..see the links to all the ports here - ksachdeva.github.io/rethinking-ten…

@ksachdeva @rlmcelreath @fehiepsi @junpenglao Is there a similar port for pytorch or pyro? Just curious.

Statistical Rethinking (2nd Ed) Port to Tensorflow Probability github.com/ksachdeva/reth… Thanks @rlmcelreath @fehiepsi @junpenglao