Some very exciting news... after many, many, winter beards, I'm overjoyed to be joining sunny
@UCSDPsychology
as an Asst Prof in Jan 2025!
Looking to recruit folks interested in doing some fun science, while building some cool (and useful) toys! Details:
Ever wonder how diff analysis strategies for repeated-measures data compare (e.g multi-level models, cluster-robust errors, two-stage-least-squares)? I wrote a thing! A lot of this was to build some intuitions for myself, but hopefully others enjoy it too:
1/3 A bit late, but proud to share my work on the social value+function of gossip is out in Current Bio . This is the first serious project that
@lukejchang
and I did together when I joined his lab and it's been a tough one to get published making…
It's been a long, frigid journey, but the paperwork and signatures are in...I've officially completed my evolution from bleary-eyed eskimo to doctor of cognitive neuroscience with stops along the way to collect a variety of beards. Thanks all for the love+support over the years.
After an awesome review experience w/ great feedback, happy to see my first solo Python project has been accepted at
@JOSS_TheOJ
. Hoping it's helpful to scientific Python folks craving some native-feeling lme4 mixed-modeling (plus other goodies!)
Thrilled to finally share a preprint of some of the work I recently presented at
@SANS_news
and discussed previously in my
@InnovatorsTalk
(
@jeremyRmanning
was so excited he beat me to the punch!)
3/3 So I wanted to take the opportunity to share a bit about the journey rather than the science. If that sounds at all interesting to you, I've written that up as post here:
With all the challenges and stochasticity that make “succeeding” in academia feel impossibly difficult, it can be really fun, intellectually stimulating, (and rewarding) to try on some other hats every once in a while esp. Good pill for that imposter syndrome.
Team of psychology graduate students was unanimously declared first place by judges from
@Google
and
@McKinsey
for the Dartmouth Consulting Case Competition. Good reminder of how communication and analytical skills taught in academia can be useful in solving real world problems!
Excited our preprint is finally out! Massive effort combining naturalistic paradigm with fMRI, behavioral, and facial expression data. And a few open source toolboxes to boot :)
Bit of happy news: my talk on thoughtful design abstraction about
@cosanlab
's nltools library and a digital poster for pymer4 were accepted to my first ever submission to
#Scipy2020
@SciPyConf
. Going virtual (pre-recorded talk) = I get to embarrass myself with permanence.
Been attending my first
#SciPy2020
this week. Such a warm, fun, welcoming, helpful community. In case it's of interest, here's my talk discussing software design abstraction from a more psych+learning perspective, guiding our dev decisions at
@cosanlab
Absolute honor and totally trippy to hear my voice on the same programming (Vermont Public Radio
@vprnet
) that I myself have been waking up and listening to since I moved to the area nearly a decade ago 😅. Thanks so much to the wonderful
@mwertlieb
!
Been looking forward to this for a while! I highly recommend fellow psych researchers check this out, especially if you use MTurk/crowdsourced data collection. It's a serious reminder that crowd-workers are HUMANS, not cheap data points, and they deserve to be respected as such.
Ghost Work (w/
@marylgray
) is out today! “A startling exposé of the invisible human workforce that powers the web - and how to bring it out of the shadows.”
My nearly all Python lifestyle means I can never remember how to R. But I still need some lme4 glmm goodness, so I wrote this get my fill (plus some extras) in the pydata ecosystem. Hope it's useful for some. Bug finds + suggestions welcome!
#LanguageWars
(1/12) Some new work to share from
@cosanlab
in which we examine the efficacy of custom headcases for reducing head motion in "naturalistic" fMRI experiments. Big shoutout to
@sadhukhas
without whom this would never have gotten done:
Excited to announce a new pymer4 release! Tons of bug fixes+hopefully easier installs, shiny new docs with tutorial notebooks, two-stage regression models, bootstrapping+permuting arbitrary functions, model i/o compatible with R's readRDS and more:
(1/12) Some new work to share from
@cosanlab
in which we examine the efficacy of custom headcases for reducing head motion in "naturalistic" fMRI experiments. Big shoutout to
@sadhukhas
without whom this would never have gotten done:
Excited to share a project I've been working on with
@jcheong0428
@DALI_Lab
@MagnusonCenter
, using ML to track real-time wait at our local
@dartmouth
cafe. Mobile app will have historic trends, forecasts, and crowdsourced inventory tracking. Our blog post:
I can't overstate the incredible support from my post-doc mentors
@lukejchang
&
@esfinn
and my faculty mentors
@ThaliaWheatley
&
@jeremyRmanning
. It's not possible to do justice to the full journey + all the people in a thread, so look out for a longer post later this year 🥲.
In academia, I often wonder if anything we do has **real measurable** societal impact. I'm lucky to know these women right here that took matters into their own hands and answered that question with a resounding **fuck yes**.
2/3 …the rounds for ~5 years at a bunch of journals. While it certainly feels good to have one's work "out there," this project in particular has a had so many unexpected effects on my personal journey in research and my thinking more broadly.
This was an awesome experience. Big believer that committing to a short time window of focused project collab and learning from others is the fastest, most fun way to level up your skills and maybe more importantly, your confidence. Come to learn, leave hungrier to learn more.
Using a naturalistic (movie-watching) approach, we consistently find that people spontaneously represent and remember other people as a function of their social relationships, rather than person-specific features like traits.
Another fantastic explorable by
@ncasenmare
. Play around to really build an intuition for how ideas spread between people and how small world networks can strike the optimal balance between insular tribalism and collective madness.
Excited for folks to check out the preprint! I think our findings complement many recent papers studying how people acquire and use relationship information and social cognitive maps with task-paradigms by folks like
@psychNerdJae
,
@SeongminAPark
, and
@parkinsoncm
among others!
Woah this is awesome thanks much
@shotgunosine
! I have a ton of updates, code refactors, issue resolves, and installation eases (some visible in dev branch) planned for late fall after I defend my dissertation 🙃. Happy to take suggestions+contributions about improvements, etc!
I hacked together a sklearn wrapper for
@Eshjolly
's pymer4 wrapping of lme4 in a few hours then spent a day or so getting it close to inline with the sklearn-contrib guide. Check it out if you need mixed effects models in an sklearn pipeline:
Highly recommend psych/neuro folks try to engage with communities like
@SciPyConf
if you get the chance. Lots of amazing work and opportunities to contribute and give back to a lot of the (free and open) tooling we heavily rely upon.
@fablab
@kristinarapuano
is a postdoc in the FAB lab studying neurobiological vulnerabilities to health risk behaviors—such as obesity and substance use. Kristina is both on the job market as well as the dog market. Hire her, or at least let her pet your dog. 🧠👩🏼💻🐶
Hi friends, if you/anyone you know has benefitted from using my pymer4 package I'd love to solicit some contributions to help maintain it. Between various cross-lang dependency updates it's become really tricky for me alone. Please see and RT if you can!
Three rules for building good software:
1. build on existing tools wherever you can
2. the development team should be the first users
3. iterate until onboarding new users is painless
@DrAnneCarpenter
at
#SciPy2020
"...it is classical economics-the *theoretical* subfield-that has been the dominant influence on U.S. public policy for a century or more...basing policy on classical economic theory is like basing a contract on philosophy."
Yael Niv + Alison Adcock FTW
@samnastase
When our teeth and finances finally erode away, at least we'll be shielded from the embarrassment of reading our own work for lack of access
@DrDamienFair
@ndosenbach
@cosanlab
@sadhukhas
While the mean differences early on (< 5m) appear the favor headcases, we didn't find this difference to be reliable (shaded region = 95% bootstrapped CIs) (Fig S6)
🎥 How the event loop works.
➡️ How stuff gets from other threads into JS.
➡️ Why infinite loops block rendering.
➡️ Why setTimeout loops don't.
➡️ rAF vs setTimeout.
➡️ Is rAF before or after paint?
➡️ When microtasks happen.
@EikoFried
It’s possible for this to occur if you don’t “freeze” the value of regularization hyperparam to what was estimated with original data, ie every bootstrap should use same regularization if ci’s are used to draw inferences about regularized model params. Not sure if that’s helpful!
@rxxqx
This is a solid reference: but doesn't have minimum recommendations if memory serves. Other things to consider are the ratio between N of samples and N of features (how "wide/tall" is your data) and how to cross-validate (eg leave-one-out, k-fold...).
@Leesplez
Always like this paper that suggests there isn’t much of a diff between methods but what matters more is if they’re applied to a model fit via REML (yay) or ML (uh oh):
@talyarkoni
@psforscher
@StuartJRitchie
Aren't they not *completely ignoring school variance* due the use of robust SEs? I've gotten the impression that it's common in some disciplines to favor OLS + robust SEs vs mixed models, when making pop level inferences and w/ adequately sampled data, result is often similar.
I really love this. For me it connects naturally with asymmetry of risk as discussed by folks like
@nntaleb
in
#skininthegame
. Our scientific theories/models can be valuable, but when taken out of the lab always worth asking who bears the brunt of the consequences if we're wrong?
@talyarkoni
If you like Vue checkout Svelte. Just came across it recently. Highly similar but ingeniously “compiles” to vanilla js dramatically dec memory footprint, no depends and inc speed. Plus you have to do basically nothing to get benefits:
@criticalneuro
Also for any packages I work on I always use the pip install -e flag for dev mode so package code is auto updated (technically symlinked) on every save of source files. Makes side by side packaging testing and developing a breeze.
@Mark_A_Thornton
@sbrietz
@kenam15
As staff, I haven’t gotten any such email or been asked to sign anything. I also don’t recall signing anything ever that asks me to trade my health for a pay check. Will update if that changes. Regardless, this some serious bs.
Erratum: Due to naming confusion on my part, two-stage-ls (a technique for instrumental variable estimation) is *not* what I mean here, but rather two-stage-regression/summary statistics approach. I've updated the post accordingly. Who said twitter comments can't be helpful?
@elvandy1515
A novel approach to establish construct validity via predictive brain models applied to a wide variety of task domains made possible through open and shared data sets. Plus a new tool for folks to try this approach themselves:
#dataparasite
(4/12) Unlike Power et al who found headcases to be helpful, depending on the metric, we found headcases were similar in efficacy to foam pillows (Sherlock/grey) or foam pillows+medical tape (FNL/blue) (left column in the figure):
@C_Shamballa
@DanielaJPalombo
In general, it's highly dependent on how clearly people speak, with faster speech being transcribed more poorly. Most transcriptions we've done require editing a few words or a phrase here and there (not too bad).
@jscofield24
Thanks! The latest version facilitates conda installs to iron-out a lot of issues users were having. Currently on pre-release, but stable is coming very very shortly!
@C_Shamballa
@DanielaJPalombo
There's also a way to get out confidence values from the api with alternative transcription possibilities to potentially making fixing the transcriptions easier. Haven't personally played with this option however.
@pepmaio
@Hesketh_GD
Thanks for moving the discussion to GH. See my reply there. Link for anyone else who finds this twitter question and has experienced this issue:
@criticalneuro
Atom for package development/maintenance, analysis scripts. Jupyter notebooks for interactive analyses. Remote script work through atom as well which has a nice plugin for that with benefit that any syntax checking/linting now works in remote files.