Sign in

Ehud Karavani

@ehudk.bsky.social
288 followers 237 following 541 posts

Research Staff Member at IBM Research. Causal Inference 🔴→🟠←🟡. Machine Learning 🤖🎓. Data Communication 📈. Healthcare ⚕️. Creator of 𝙲𝚊𝚞𝚜𝚊𝚕𝚕𝚒𝚋: github.com/IBM/causallib Website: ehud.co

PostsRepliesMedia
Ehud Karavani @ehudk.bsky.social · 09/10/2026
I do bitwarden too, and it's a relatively smooth experience across macos, windows, and android. though I suggest taking a look at keepass which is both open source and local so you can easily self-hosted on google drive and the like that already synced on your devices.
020
Reposted by Ehud Karavani
Emily Riederer @emilyriederer.bsky.social · 08/10/2026
In an icebreaker a few weeks ago, our team was sharing pet peeves and an early-20s said API 429 errors. Not API errors but specifically status code 429 (too many requests) and I'm still thinking about it If that somehow isn't a metaphor for early 20s exuberance.. your 30s are all 500s
0132
Ehud Karavani @ehudk.bsky.social · 28/09/2026
for good scientific reasons I presume
PubMed title screenshot: Epidemiology, Transmission Mode, and Pathogenesis of Chlamydia pecorum Infection in Koalas (Phascolarctos cinereus): An Overview
020
Ehud Karavani @ehudk.bsky.social · 28/09/2026
your scientists were so preoccupied with controlled vocabulary, that they forgot about controlling spelling 🙃
020
Ehud Karavani @ehudk.bsky.social · 27/09/2026
Unless this podium is a histogram and then Paul is right there at the population mode
010
Ehud Karavani @ehudk.bsky.social · 04/08/2026
my annoyance of bayesian modeling is mainly due to selection bias - I approach it when there's no other viable alternative, so i always encounter the painstaking experience of needing to iterate and refine with every fit taking upwards of 30 minutes
061
Ehud Karavani @ehudk.bsky.social · 30/07/2026
the body that spent months needlessly dividing cells suddenly remembers to be efficient just before the joke drops? fuck it, let them be pissed out
010
Ehud Karavani @ehudk.bsky.social · 27/07/2026
Penguin colonies poop so much that you analyze their diet from space science.nasa.gov/earth/earth-...
science.nasa.gov
Pink Penguin Guano Provides Diet Clues - NASA Science
The color of Adélie penguin droppings reveals what the birds are eating, offering scientists a way to track how sea ice conditions influence their diet.
133
Ehud Karavani @ehudk.bsky.social · 27/07/2026
What is ECE if not a version of Area Under the Calibration Curve 😅
220
Ehud Karavani @ehudk.bsky.social · 17/07/2026
another instance of tech-bros reinventing old technologies. since the majority of drug developers are already publicly traded, you can easily bet on the outcome of a clinical trial in the good ol' stock market.
010
Ehud Karavani @ehudk.bsky.social · 15/07/2026
Feels less like football and more like reenactment of The Falklands War
020
Ehud Karavani @ehudk.bsky.social · 28/05/2026
and also writing tests with print statements instead of explicit assertions 😵‍💫
030
Ehud Karavani @ehudk.bsky.social · 27/05/2026
psychologists at least have pronounceable acronyms epidemiologists can only wish for bsky.app/profile/conj...
010
Ehud Karavani @ehudk.bsky.social · 27/05/2026
i don't know if "serious" because you can always default to vscode if needed. but if you want to use other vscodium-like IDEs (Cursor? Windsurf? Positron?), then you're limited to the less protected (though still ok i guess) ssh keys.
110
Ehud Karavani @ehudk.bsky.social · 27/05/2026
ssh2 is a javascript surrogate for OpenSSH (which I guess typescript-based extensions interface more easily with?), and a lot of the Remote SSH extensions build on top of it. I tracked this to be the earliest pointer for the issue, May 2017: github.com/mscdex/ssh2/...
github.com
Feature Request: support certificate authentication. · Issue #551 · mscdex/ssh2
In OpenSSH one can use a CA-signed key and certificate pair to sign-in to an SSH server with CA authentication turned on. The server sets a public certificate as the "trusted ca" and the client the...
110
Ehud Karavani @ehudk.bsky.social · 27/05/2026
encountered a real-life example of this xkcd in the form of why no one can use the safer temporary ssh certificates in any vscode derivatives. Microsoft locked their Remote extension, leaving others to rely on ssh2, whose maintainer ignores SSH certificate support for almost a full decade now 🤷‍♂️
120
Ehud Karavani @ehudk.bsky.social · 16/04/2026
Ed Yong once wrote "science is the teamiest of team sports"
010
Ehud Karavani @ehudk.bsky.social · 31/03/2026
we already have estimands and predictimands, so I guess "descriptimands" is just citations on the floor at this point
010
Ehud Karavani @ehudk.bsky.social · 31/03/2026
the estimands for MBA curriculum looks lit 🔥
010
Ehud Karavani @ehudk.bsky.social · 25/03/2026
No book recs but just good luck and swift recovery
010
Ehud Karavani @ehudk.bsky.social · 25/03/2026
For what it's worth, here we usually pronounce it with a proper, yet wrong, German "ch" (we don't have that soft ch-sh-like sound but our local "ch" sounds similar to the harder German "ch").
020
Ehud Karavani @ehudk.bsky.social · 25/03/2026
A frame from there movie Inglorious Bastards, where a German officer sits at a bar and the man impersonating a German officer in front of him signs "3" with his hand using the 3 middle fingers.
120
Ehud Karavani @ehudk.bsky.social · 23/03/2026
at last!
A tweet from May 2024 (part of a discussion about academic uses of quarto): 
Another hurdle for writing big papers will require to ease collaborations by creating good real-time browser-based online collab environments. The current solution Posit suggests is (not well known and) a bit insufficient - not as easy as just opening an Overleaf or google docs.
020
Ehud Karavani @ehudk.bsky.social · 23/03/2026
one inescapable truth of modern science conduct is that if you're a researcher then you're a software engineer. but if you make the most out of it, it can make you a better researcher. just had to get some stuff out of the system and into yours 🫶 ehud.co/blog/2026/01...
ehud.co
Research ❤️ Software Development | Ehud Karavani
In a world where science becomes inseparable from coding, code becomes king.
0182
Ehud Karavani @ehudk.bsky.social · 23/03/2026
the upside is that the amount of information you can encode depends on the number of dimensions you have, so this near-orthogonality allows you to store more information in practice with fewer actual dimensions. you win some you lose some 💅
000
Ehud Karavani @ehudk.bsky.social · 22/03/2026
bayesian pooling >> bayesian updating
020
Ehud Karavani @ehudk.bsky.social · 20/03/2026
I second this. Isn't this an equivalent discussion of intention-to-treat effects for policy makers vs. per-protocol effects for the people?
120
Ehud Karavani @ehudk.bsky.social · 20/03/2026
similar to overlap weighting (or trimming), sometimes the data forces your hand. but if you're clear about whom your conclusions should apply to, then it is likely to be better than knowing nothing.
140
Ehud Karavani @ehudk.bsky.social · 19/03/2026
Not to mention that a lot of prediction tasks are actually anticausal in nature: disease (Y) produces the symptoms (X), objects cause image pixels, intent causes words...
000
Ehud Karavani @ehudk.bsky.social · 19/03/2026
do many hundreds of pull request reviews count as professional critic? 😅
000
Ehud Karavani @ehudk.bsky.social · 18/03/2026
In my experience, many of the culprits are due to data entry (and organization) directly in excel. In participant-facing studies, I used to advocate research assistants to enter data through a shared Google form (or equivalent). Makes entry much more structured for relatively little work.
000
Ehud Karavani @ehudk.bsky.social · 13/03/2026
I think the more important benchmark in such long hauls is how well your battery synchronizes with your bladder 🙃
001
Ehud Karavani @ehudk.bsky.social · 09/03/2026
i'm a simple man: some 6-length codes are best remembered in 3 pairs and others are better remembered as 2 triplets. which type comes more often is left for astrology.
010
Ehud Karavani @ehudk.bsky.social · 07/03/2026
STAT saw and raised
010
Ehud Karavani @ehudk.bsky.social · 07/03/2026
always use a Mondrian theme in your ggplots @statsepi.bsky.social
081
Ehud Karavani @ehudk.bsky.social · 18/02/2026
To his credit, he states as much as the very beginning of the post. The type of software this approach aims for is rarely seen nowadays, and (as someone who once thought 𝘵𝘩𝘪𝘴 𝘪𝘴 𝘵𝘩𝘦 𝘸𝘢𝘺) it should be judged accordingly. no need to be too harsh on it.
Note of reflection (March 5, 2020) 
This model was conceived in 2010, now more than 10 years ago, and not very long after Git itself came into being. In those 10 years, git-flow (the branching model laid out in this article) has become hugely popular in many a software team to the point where people have started treating it like a standard of sorts — but unfortunately also as a dogma or panacea.

During those 10 years, Git itself has taken the world by a storm, and the most popular type of software that is being developed with Git is shifting more towards web apps — at least in my filter bubble. Web apps are typically continuously delivered, not rolled back, and you don't have to support multiple versions of the software running in the wild.

This is not the class of software that I had in mind when I wrote the blog post 10 years ago. If your team is doing continuous delivery of software, I would suggest to adopt a much simpler workflow (like GitHub flow) instead of trying to shoehorn git-flow into your team.

If, however, you are building software that is explicitly versioned, or if you need to support multiple versions of your software in the wild, then git-flow may still be as good of a fit to your team as it has been to people in the last 10 years. In that case, please read on.

To conclude, always remember that panaceas don't exist. Consider your own context. Don't be hating. Decide for yourself.
020
Reposted by Ehud Karavani
Nadieh Bremer @nadiehbremer.com · 12/02/2026
📣 NEW! I’ve just released the BIGGEST and perhaps most creative project I’ve ever worked on! “Searching for Birds” searchingforbirds.visualcinnamon.com 🐤 A project, an article, an exploration that dives into the data that connects humans with birds, by looking at how we search for birds.
26503182
Ehud Karavani @ehudk.bsky.social · 06/02/2026
of course, this meme is magnetic 🧲 🧭
000
Ehud Karavani @ehudk.bsky.social · 04/02/2026
Robocalypse: The Revival of the Mechanical Turk
010
Ehud Karavani @ehudk.bsky.social · 04/02/2026
it's a good view, don't get me wrong, but I also find it a bit limiting - because you can benefit from using random effects to pool estimates / shrink effects of finite factors, too, especially in imbalanced settings
020
Ehud Karavani @ehudk.bsky.social · 04/02/2026
that whammy bar on the guitar is excellent detail 👌
110
Ehud Karavani @ehudk.bsky.social · 30/01/2026
funny, I wasn't aware he had other fame aside from his spanning trees fame
010
Reposted by Ehud Karavani
Dr. Headbutt 🌈🧪 @nicoleackermans.com · 28/01/2026
If you haven't watched the video abstract about Veronika the tool-using cow, you really should, especially the message from her owner at the end! It cleanses the timeline a bit. 🧪 www.cell.com/current-biol...
cell.com
Flexible use of a multi-purpose tool by a cow
Osuna-Mascaró and Auersperg report flexible, multipurpose tool use in a cow, expanding the known range of mammalian tool users and underscoring overlooked cognitive capacities in livestock.
05414
Ehud Karavani @ehudk.bsky.social · 28/01/2026
Quarto's yaml autocomplete was my sole inspiration for coding a json schema for some internal package we developed that has a yaml/hydra interface. was a fun Pydantic exercise but I wouldn't recommend to a friend...
110
Ehud Karavani @ehudk.bsky.social · 28/01/2026
*doesn't have to necessarily write all of the code, but thoughtful use of llms to contemplate different approaches rather than blindly accepting code shows care and more detail-oriented mentality and critical thinking, which are broader traits of interest in an era where everyone can now code.
000
Ehud Karavani @ehudk.bsky.social · 28/01/2026
this. "why" questions are the key. someone writing their own code* will have in depth considerations of choices made because they struggled with deciding on different implementations. but delegating coding to an llm makes them only a reader of the code, not so different than you reviewing it.
100
Ehud Karavani @ehudk.bsky.social · 23/01/2026
Always happy when you chime in. Interesting idea. Do you have any examples? I read "error bars" in a frequentist sense, so you'll only have mean and two interval edges (rather than continuous throughout the interval?), and then dithering of three colors would seem too irregular?
110
Ehud Karavani @ehudk.bsky.social · 23/01/2026
If you can map your error bars to a 0-1 confidence there are ways to use color transparency and/or hex size (softening the edges) to make more uncertain hexes less prominent
110
Ehud Karavani @ehudk.bsky.social · 22/01/2026
happy b-day wikipedia wikipedia25.org/en
000
Ehud Karavani @ehudk.bsky.social · 20/01/2026
Both are "AI" in general (at least colloquially), but are different in how we create them, validate them, and deploy then to interact with the public
000