Show more

Offspring woke up and the first thing he says is a verse from our new favorite Japanese rap song 😍 and I‘m like “😂 there’s people who don’t like weird people like you and me but!!!, you find the other people who do like us and stick with them!”

“Don’t worry I’ll do that.”

Show thread

The inestimable on “one of my most ridiculable ideas”, that 95th-percentile isn’t hard:

“If you follow these people who ask for help, you'll often see them ask for feedback at a much higher rank (e.g., moving from 10%-ile to 40%-ile) shortly afterwards. It's nice to see that the advice works, but it's unfortunate that so many players don't realize that watching their own recordings or posting recordings for feedback could have saved 1198 hours of frustration.”

danluu.com/p95-skill/

“Why can’t you just be normal? Please? One day?” —coworker😍

kVin of : “Ten years since the Disappearance of : an eternal legacy”

blog.sakugabooru.com/2020/02/0

“The Disappearance of Haruhi Suzumiya was released 10 years ago today, on February 6, 2010. A decade later and despite the tragic losses, its legacy at and the industry altogether still endures.”

Ten years! Feels like six weeks ago. Rewatching tonight.

This project is following a similar trajectory as . Design the probabilistic model in PyMC, fail to get any traction with MCMC, then write out the explicit likelihoods and posteriors and integrate the shit out of them on 🐺🐏α. I don’t have a posterior on a scalar random variable here, I have a machine learning dataset, so experimenting with straight maximum likelihood Via optimization, but will probably have to go (stochastic) gradient descent. Excited to see if it works.

Show thread

And only three of five slots have been taken: mobile.twitter.com/dhh/status/

Ruby and EU are well-represented in that sample of three 🤣! But I really like the clarity DHH brought to this. I look forward to telling my grandchildren how primitive and backwards-looking companies were in my time.

Show thread

’s & DHH has a series of fire threads:

“We ended up with over 1,200 applications for the programmer opening at Basecamp. If you’re having trouble attracting talent, feel free to copy our playbook: Hire remote, be explicit about pay, detail job exhaustively”:
mobile.twitter.com/dhh/status/

Then he’s like, people say we get such responses because I have a big twitter following, so will boost five job openings that meet his standards:
mobile.twitter.com/dhh/status/

Finished 2020 dev survey—

stackoverflow.blog/2020/02/05/

“we want our survey to represent everyone who codes” — even if you don’t identify as a developer, if you write any code and take the survey, I’d appreciate your thoughts on it.

My favorite part of the analysis is the most loved and most hated languages and tech stacks! And even though it’s very inaccurate I like seeing the pay ranges in different parts of the country and the world.

Hot on the heels of a solid model with a (maybe) reasonable fitting scheme! More details to come, but let it be known now—

I followed a nice answerer on Stack Overflow and named a variable κ and all I can think about every time I see "kappa", "logitKappa", and "kappas", is 🤩 .

Show thread

🤞 discourse.pymc.io/t/recommende

Many people have been incredibly kind to me over the years, answering my complicated questions online, on any number of ridiculous things, from generating maps (like, geography) in JavaScript to heavy statistics. I'm so grateful I can't really express it, can just try to pay it forward.

Still, in case this is the end of the line, I'm looking at TensorFlow as an alternative, and manually doing Gibbs sampling. Although there's likely a direct approach.

Show thread

😭 my chains aren’t mixing at all let alone converging, I must have messed something up royally while implementing a new distribution but don’t know what. Will ask the gurus tonight 😢.

Show thread
22 boosted

Actual sign at a candy shop in Hakone, Kanagawa, Japan 

Show thread
22 boosted

Actual sign at a candy shop in Hakone, Kanagawa, Japan 

22 boosted

For those of you(us) considering academia for your future careers, I have some bad news for you:

biorxiv.org/content/10.1101/79

> Our findings suggest that there is no single clear path to a faculty job offer and that metrics such as career transition awards and publications in high impact factor journals were neither necessary nor sufficient for landing a faculty position.

22 boosted

Terrible confession.

As much as I go on about how self-teaching and apprenticeship/mentorship is all you need for learning to code and such, there’s probably no way in hell I’d have ever gotten any good at and without spending at least two years in school immersed in it.

Doing projects (in grad school, at work, and for fun) is crucial to cementing skill, ya, but the preliminary material was so hard for me at least that, sigh, I’m glad I went to school for it…

Show thread

It's been more than six years since my last question on Stack Overflow (when it wasn't !), but it's gotten really good.

stats.stackexchange.com/q/4473

Question comes from experimenting with Duolingo's half-life regression (flashcard scheduler), which seems to be broken but has a couple of salvageable ideas, including

- quiz the same item multiple times per quiz session? Bernoulli→Binomial.
- can you collapse quiz history to just "# times flashcard seen vs correct"?

Show thread

‘Did you know there were other worthwhile anime airing this season besides Keep Your Hands of ? It’s true!’

🤣 continues to drop 🔥 animefeminist.com/2020-winter- fixing to get a trial for Funimation to try , ‘“The stealth best comedy of the season.”’

Show thread
22 boosted

Isn't it unnerving that an event with 3% probability is 26% likely to happen at least one time out of ten?

Do something that rare—and 3% seems very rare right??—ten times a day and you're encountering it twice a week.

Kind of freaky.

Bernoulli magic.

Show thread
Show more
Octodon

The social network of the future: No ads, no corporate surveillance, ethical design, and decentralization! Own your data with Mastodon!