I'm excited to announce the first beta of a new Python package that I've been working on:
It's (yet another) Gaussian process library in Python, this time built on JAX. It's meant to be both performant & pedagogical with ~4x as many lines of docs as code.
Some news: in a few weeks I'll be leaving
@FlatironCCA
to join
@GoogleDeepMind
where I'll be working on the JAX core dev team.
It's bittersweet to leave astro which (for better or worse) has defined much of my identity for the past 15 years, but I'm excited about new adventures!
Today is my last day as a professional astronomer, and I already miss all the wonderful people who I have had the immense privilege of working with over the past 15 years. But, I'm excited for what comes next, and y'all know where to find me:
❤️
Approximately infinity internet years ago I went up for promotion here at
@FlatironCCA
. Good news: I got it! 🍾
We don't have tenure here, but this promotion means that my position is no longer term limited and it's expected that I'll be here for a while. My stoke level is high!
I'm going up for promotion here at
@FlatironCCA
(the closest thing we have to tenure) and I'm giving my research talk tomorrow. Naturally I'm procrastinating by making plots: here's what happens when I pit the GP libraries I've built against each other…
An awesome paper by
@rodluger
+ (incl. me) on arXiv this morning:
Even if you're not computing light curves for exoplanets, the craft of this paper is outrageous: each key equation includes a link to a sympy proof, each figure has a link to the code, …
I'm chairing the hiring committee for the
@FlatironCCA
"Flatiron Research Fellow" postdoc fellowship this year, and wanted to share some info about it here. If you're on the astro postdoc job market this year (or if you know someone who is), read on! 🧵
For the past [longer than I care to say], I've been contributing to an Annual Review titled "Gaussian Process regression for astronomical time-series" led by
@AirborneGrain
.
Well… yesterday we sprinted to finally submit it!! 🚀🎉
Coming soon to an ArXiv near you. (I hope!)
I'm excited to release some code that I've wanted for a long time. It's a browser extension that keeps track of which papers you navigate to on and then sorts the arXiv listings (in place) based on your preferences:
I'm going up for promotion here at
@FlatironCCA
(the closest thing we have to tenure) and I'm giving my research talk tomorrow. Naturally I'm procrastinating by making plots: here's what happens when I pit the GP libraries I've built against each other…
I've just been invited to give a plenary lecture at
#AAS241
with the general topic "Computational Astrophysics" - I'm so honored!! Looking forward to seeing (some of) y'all in Seattle to chat about software development in astro, among other things!
Check out my latest blog post which provides an introduction to probabilistic modeling using JAX+NumPyro. The post is aimed at astronomers, but may be of interest to others. We start from 0 & end up running MCMC on real
@ESAGaia
data with >4000 parameters!
I've used this slide in some recent talks, saying (based on my experience) that users of many scientific packages want to understand/tinker, rather than just press a button.
@chrislintott
tells me this may be more of an astro thing than a general rule. What do y'all think?
I just released the first major update to emcee in way too many years:
Even though this was a full re-write, the interface is mostly backwards compatible, but there are some changes that you should check out before you upgrade:
Today on arXiv: the first in an epic series of papers from
@rodluger
et al. where we investigates what we can (& can't) measure when mapping the surfaces of stars (or exoplanets) using time series observations:
🚨 New fellowship opportunity
@FlatironCCA
🚨
The "Flatiron Software Research Fellowship" is a 5 (2+3) year postdoc fellowship with a focus on developing & maintaining software infrastructure for astro. Due Nov 15. Get in touch with any questions!!
Well I wrote far too many words about a topic that (at best) one other person will be excited about. Is that person you?! This was meant to be a blog post but it ended up as a GitHub repo.
A tutorial: "Extending JAX with custom C++ and CUDA code"
📅 SAVE THE DATE! 📅
Who here is into (/wants to get into) writing/contributing to open source software for astro? We're organizing a workshop here at
@FlatironCCA
May 16-20, 2021, focused on learning from each other about what works/doesn't:
Hey!
Do you use and/or write scientific software?
Do you want to help scientific software get academic credit?
If so, please volunteer to review for
@JOSS_TheOJ
:
We're getting a lot of (esp. astro) submissions and the reviewers are working overtime!
TFW you discover that obscure old paper from a different field that holds the solution to the problem you've been working on for months because you just didn't know the right words to google…
Well I'm a huge nerd and in just over a week starts. I <3 Advent of Code and, even though I have no time, I'm going to try to do it using
@rustlang
this year because that seems pretty neat. Who else is in?
I don't write too many first author papers so I'm excited that there's one on the ArXiv today!! This has been submitted to
@JOSS_TheOJ
, and it describes the package & related libraries. Check it out here:
Do you ever use the astropy LombScargle method for computing periodograms? If so, you should check out Lehman Garrison's new nifty-ls package which provides some new faster (and hardware accelerated) backends to the astropy function:
Here are some slides that I shared yesterday at an event celebrating
@DavidSpergel
&
@FlatironCCA
's birthdays. It includes various rambling thoughts about the importance of open-source software development for astrophysics and how we can support this work:
Despite the procrastination, I did (in fact) give this talk yesterday, with topics ranging from algorithm development for Gaussian Processes, to open source scientific software musings. For anyone following along, here are the slides:
I'm going up for promotion here at
@FlatironCCA
(the closest thing we have to tenure) and I'm giving my research talk tomorrow. Naturally I'm procrastinating by making plots: here's what happens when I pit the GP libraries I've built against each other…
I just blogged a short update to my earlier post where I demonstrated how to combine
#TensorFlow
and
#PyMC3
. This time I mashed up
#PyTorch
and
#PyMC3
for all sorts of awesome! Check it out:
Alongside today's release of the new
@NASA_TESS
data, I'm excited to announce the release of a new Python library that I've written for fitting exoplanets (transits & RVs) using
#PyMC3
:
Have you ever wanted to use
#PyMC3
to sample models defined in
@TensorFlow
(or some other modeling language like
@PyTorch
)?
It's not necessarily a good idea, but I wrote a blog post about how to do just that:
I'm excited to announce that I've just joined the editorial team for
@JOSS_TheOJ
where I'll be a topic editor for astrophysics, probabilistic programming, and python. Send me all your software papers!!!
My tutorial demonstrating how to implement custom C++ and CUDA ops for JAX () was horribly out of date, but thanks to a few PRs it is now up to date with the current best practices:
I put together a little Python library that makes it possible to use units (implemented using ) in JAX. It's not perfect, & there may be cleverer approaches, but check it out/send feedback/report issues:
Thanks
@shoyer
for tips!
I don't love writing scientific papers, but I do <3 writing technical docs! I'm beginning to think I may have chosen the wrong career…
(hashtag overly honest methods or something)
It has just been officially announced that my amazing sister
@formacannie
will be representing Canada at the postponed 2020 Olympics! This year was a strange one to qualify for the Olympics, but Annie never does things the easy way. I'm so proud of her!!
13 track cyclists and 4 road cyclists will represent
#TeamCanada
🇨🇦 at
@Tokyo2020
🎉
Among these
@CyclingCanada
🚴♀️🚴♂️ athletes are:
8️⃣ First time Olympians
3️⃣ Olympic 🥉 medallists
2️⃣ Dual season Olympians
1️⃣ 4-time Olympian
See the full list 👇
We have been very lucky that folks have been generous with citations, but our "success" doesn't mean that academic credit for software is solved! There is a lot of hugely impactful software (& developers) not getting this kind of support.
How can an excellent software implementation of an MCMC method have more citations than the discovery of the freakin’ baryon acoustic feature??
Maybe software is getting new recognition!? cc
@exoplaneteer
Hey astro grads: Applications are open for the
@FlatironCCA
pre-doc program! Through this (paid) program, students visit NYC for ~6 months to collaborate with folks at
@FlatironCCA
. More info about this awesome program here:
It took forever, but applications are finally open for the "Astronomical Software Development" workshop that we're hosting at
@FlatironCCA
, May 16–20, 2022!
Check out the website for the link to the application form & all the other details:
It's not that uncommon for me to find code that I've written copy pasta'd into other astro codebases without attribution. This is compatible with the licenses I use, doesn't truly matter, & deep down I'm stoked to be contributing, but I do still feel a little sad. Am I wrong?
If you're looking for some light Sunday morning reading, I wrote a tutorial for fitting
@NASA_TESS
light curves using exoplanet (the code) and
#PyMC3
:
For those of us who are interested in open & reproducible scientific publication practices, check out these icons throughout this paper. The key equations are unit tested & the *exact* code+environment that was used to generate the figures is available:
Today on arXiv: the first in an epic series of papers from
@rodluger
et al. where we investigates what we can (& can't) measure when mapping the surfaces of stars (or exoplanets) using time series observations:
I've started using for a paper that I'm writing and I want to emphasize that this is SO AWESOME! If you're in the early stages of writing a paper, you should definitely check this out.
@rodluger
is taking open/reproducible science to the next level!
When it rains it pours. Second tweetstorm of the day, this time on a new open source scientific article workflow, showyourwork:
Make your scientific articles transparent, reproducible, and truly open source.
🧵 (1/6)
Hey there! Are you considering using the word “optimal” in your astronomy paper (or just about any other scientific discipline)? It’s probably not what you actually mean!
Also: “nearly optimal” means nothing.
That’s all. Thanks for listening!
I'm giving my
#AAS241
plenary talk in just under half an hour!
If you're still here (in-person or online), come hear me blather, or (maybe better yet) just check out the
@astrobites
summary afterwards… :D
I just published a new release of the "corner" package with support for plotting
@arviz_devs
datasets, making it compatible with most Python inference libraries out of the box:
pip install -U corner
Let me know if you run into any issues!
Here I am, overcommitted and behind on everything, and what is my brain doing? Completely fixating on working out an information theory description of this wordle game to the detriment of all other responsibilities…
Got some really nice feedback in an issue for one of my open source projects this morning:
"This is beyond my understanding that how this simple problem still could not be solved […] Sorry to say this..CORNER SUCKS!!!"
🤔
Well I just had ice cream for breakfast & I'm ok with that because I'm a grown adult & the country I've adopted continues to be on fire around me & there's a global pandemic. How's your day going?
Do you ever get sick of clicking "Reload?" and waiting for
@github
to render
@ProjectJupyter
notebooks?
I certainly do. So I made this:
You give it the URL for a Jupyter notebook and it will (hopefully) render quickly right in your browser.
Applications are open (deadline: Feb 28) for this year's
@FlatironCCA
summer school! This time the topic is fluid hydrodynamics and you can find more details (and the application) here:
I was pretty pleased with my tutorial for today's LSSTC-DSFP. We coded up three parameter estimation examples in emcee, dynesty, and pymc3. I think that this could be a useful resource. Check it out!
Worksheet:
Solutions:
Starting this week at
@FlatironInst
we have a bot that suggests random coffee meetings between members of the Institute. There's been some writing about how this can be useful for companies, but I think academic institutions stand to benefit a lot. We'll see how it goes!
Check out this big news!! I'm very excited that
@dalcantonJD
is coming to be my* boss and lead the next chapter of
@FlatironCCA
.
* well not *just* mine…
If you or someone you know is on the astro job market this year, the
@FlatironCCA
Postdoc Fellowship applications are due super early this year (Oct 15, in ~3 weeks!). This is an awesome job & a deadline you don't want to miss! Get in touch w questions.
I just pushed another blogpost in [what is turning out to be] a series about learning how to use
#pymc3
. This time: how to learn the off-diagonal elements of the mass matrix. Comments welcome!
Have you ever wished that papers on the arXiv weren't sorted in chronological order?
Do you hate racing to submit your paper to arXiv within the first second?
Now, you can get a randomized listing at: (made at
#astrohackweek
by
@adrianprw
& me)
Today at
@FlatironCCA
,
@j_bellovary
updated us about the new
@GC_CUNY
Astrophysics Master's program. I am SO STOKED about this! Free for students, PhD-level stipend, formal research collaborations across NYC, so much more.
This year's deadline is Mar 30:
🚀 I just released tinygp v0.2.0 🚀
This is a not tiny, backwards-incompatible release, that I've been working on for a while. I'm very pleased with how it turned out!
Check out the release notes () and/or read on for more details!
👇🧵
This week at
@FlatironCCA
we're hosting the workshop & it is so awesome! These are my people (although there are many I didn't know before) & I'm learning so much!! <3
[over dinner this evening]
Ruth: [our friend who lives elsewhere] just bought a house in cash.
me (who regularly looks at Zillow in NYC): Where's Cash? I've never heard of that town…?
Tired of *thinking* about the format of your code in
@ProjectJupyter
notebooks? Me too.
So: I hacked together "black" () and "nbconvert" () to automatically format and check Python code in notebooks:
Check it out! The latest
@astropy
paper is on ArXiv today:
Of note:
1. I'm a co-author this time (+ >100 other excellent folks; ref: )
2. it was written using
@rodluger
's
#showyourwork
so here's the source:
👋 Astro grad students: There's one week left before the application deadline for the
@FlatironCCA
pre-doc program! Through this program, current grad students spend 5 months in NYC collaborating with CCA staff. It's an awesome opportunity for you and us:
As part of the
@JOSS_TheOJ
review for the paper, I completely revamped the docs and case studies with several new tutorials and resources. Check them out:
docs:
case studies:
paper:
Want to build your own objective scale for judging people based on the opinions of your own superb panel of experts? Now you can!
[hint: try changing the "what" parameter in the URL…]
Today and tomorrow I'm attending the "Open Digital Infrastructure in Astrophysics" meeting
@KITP_UCSB
().
Here are the slides for my talk about open source software and tools for probabilistic data analysis:
We (
@AirborneGrain
& me) have updated our Gaussian Process review in response to (awesome!) community feedback, and v2 is available on arXiv:
All lines are now a reasonable thickness…
After a few years of not traveling, I (for some reason) decided to book 3 back-to-back weeks of work travel (3 talks, a conference, a collaboration hack week, & 8 time zones). I'm only a week in & it's been wonderful, but I have no idea how I once thought this was sustainable!
starts on Friday! 🚀
Advent of Code is a daily coding challenge Dec 1-25 that is fun in ways that I can't totally put my finger on. I normally use it as a way to learn a new language, but with a 5-month-old, I'm sticking to Rust again this time around…
I'm sure something like this exists elsewhere, but here's a nearly trivial function that plots what I wish matplotlib's step command did:
I hope it helps others too!
I've spent the majority of my research time for the past month or so (where do the days even go?!) on maintenance and infrastructure updates for my open source projects. Now I get a week to work on some new things before it's time for me to start again from the beginning!
I just published a new release for Ganymede (v0.1), a desktop app for running
@ProjectJupyter
JupyterLab in a self-contained environment. I've been using it for all my work for the past few months and it seems to be pretty stable!
Big news over here:
@DavidSpergel
is becoming the president of
@SimonsFdn
!
This is already cool, but it also means that
@FlatironCCA
is going to be looking for a new Director in an open search that (I think) will be announced early next year!
When you play video games with your friends who you also work with…
[This pull request adds a 45 minute "update" every time you import. Seems compelling.]
The "Astronomical Data" group at
@FlatironCCA
has just started a blog! We're planning on posting an update about something that we're working on every Friday. The URL is (of course). Stay tuned!
I've just had my first pull request accepted by
@astropy
(it was my first contribution to any big project, in fact)!
If you're looking for transiting exoplanets, try out the BoxLeastSquares implementation that is now included in the dev branch:
I took this week off from [the things I should be doing] to hack on JWST spectral time series of exoplanets & OMG these data are _unreal_!! What is data vs. model…?! It's been a *long* time since I've had a full week of research & dang I missed it!
@jiayin_dong
@SoichiroHattori
I know that Wordle was totally last week, but I forgot to mention that I followed through on this (spoiler warning):
Here's how the machine solved it today:
⬜🟨⬜🟩⬜
🟨⬜⬜⬜⬜
🟩🟩🟩🟩🟩
Here I am, overcommitted and behind on everything, and what is my brain doing? Completely fixating on working out an information theory description of this wordle game to the detriment of all other responsibilities…
On this week's
@FlatironCCA
Astro Data blog, I share a project led by James Hitchcock (in collab with me, et al.) to accelerate astronomical difference imaging using
@PyTorch
. (If you read to the end you're rewarded with a cool thing from
@katestoreyfish
!)