🎃 Pumpkin Spice Latent Variable 🎃

@SolomonKurz @twarrestrial @ChelseaParlett Maybe another alternative for "the model" is just P(D); θ is kind of just a figment of the model's imagination, an implementation detail: P(D) = \int P(D|θ) p(θ) dθ. E.g. if you have two different parameterizations θ and θ', but always P(D) = P'(D), I guess same model?

@twarrestrial @ChelseaParlett So you're saying the model is defined, in part, by the posterior?

Conceptual question: In Bayesian analyses, we call the p(D|theta) the likelihood and p(theta) the prior. Would it be the case then that you would call p(D|theta) p(theta) the "model"?

@mitsuhiko Fwiw, one thing you can do here is use `value: object` instead of `value: Any` in Aggregator; mypy will then complain about subsequently trying to use `value: float` and `value: str` in the subclasses (breaks liskov substitution).

On the topic of Python typing: One of my biggest qualms with the system is that once you are past having everything typed, it lets you do stuff like this and does not even complain.

@sharoz @AliGunerMD @lakens Fwiw, I wish they were called "frequency intervals" (and, modest proposal, I idly wish frequentists would just not use the word "probability" and instead say "frequency"—fairly or not, the ship has sailed and people are going to use "probability" in a subjective sense)

@sharoz @AliGunerMD @lakens E.g. imagine having a goofy/dumb CI procedure that some percentage of the time outputs an obviously, glaringly incorrect interval—you know for sure it *can't* contain the param. If you get such a dud interval, would you still say you're "95% confident" it includes the param?

If many people each ran an experiment and got a 95% confidence interval, which of these are true? 1. 95% of the CIs contain the true mean. 2. There is a 95% chance that my CI is one of the CIs that contains the true mean. 3. There is a 95% chance that my CI contains true mean.

The new @polyphia album is so sick. So happy it's finally out 😭

@dj_d_sol Whoa!

as of a few weeks ago mypy has support for recursive types (why wasn't this bigger news?) finally python can do JSON

@TkDodo Disagree in spirit. I wish more languages made it easy to write in-file tests (e.g. Rust, or Pyret). Tying the ability to check that something is working to its public visibility has never made sense to me. In-file tests can also really help reading the code!

Today's 🌶 take: Don't export functions just to write a unit test for them. If the function isn't used outside of the file in your app / library - don't test it. It's literally an implementation detail. Test the public interface only!

@akbirthko @typedfemale Lol twitter.com/lightspringfox…

@typedfemale avoiding the urge to mention Paul Erdos

amphetamines are the real "food for thought"

If you're interested in learning Russian, this is an amazing course: russianthroughpropaganda.com

@armaniferrante Have you tried post-merge code review before? Devs can just go ahead and merge their PR and then review happens async. It's the default at @.WaveSenegal (you can obvs request pre-merge review if you want, it's just not the default)—has tradeoffs but it's definitely fast 😛

Of course, it helps when you work with truly amazing people.

Easy trick for maximizing value creation for devs at early stage startups. Minimize time in code review. Even better don't bother with code review!

@dancherp Always makes me think of the scene in Die Hard 3 where one of the bad guys, a sneaky East German, gives himself away to Bruce Willis by saying "feels like it's gonna rain like dogs and cats later" 😆 youtube.com/watch?v=d6lNIp…

Funny because no one (no native speakers around me) has really ever used this idiom to describe rainfall, it seems archaic

Shower thought: regarding idioms, did other English speakers grow up learning “it’s raining cats and dogs” as one of the first examples taught in class?

@steveluscher Really enjoy threads like this (have been bit by this specific issue in the past too)—keep them coming! 🙏

@Rina30857169 @BenHambrecht arxiv.org/pdf/1505.07835…

@BenHambrecht Interesting analogy. Could you share the paper link? Thank you!

Way to start a paper 🧙‍♀️

@crispheaney @simplyianm Also applicative functors ❤️

@simplyianm If i love options/results is that enough 😅😂

What's a concept in computer science/software engineering that not enough devs understand?

@thal0x @buffalu__ No-context drive-by, but matklad.github.io/2022/06/11/cac… might be a fun read

@buffalu__ i genuinely have no idea how to make the compiler happy when it comes to this lol. but i have to move on to other things, hoping to take another stab at it tonight

realizing the reason i’m making it so far in this attempt at converting my arb bot to rust is because i saved the part i was struggling with (caching) for later now i’m at that part 🫡

I tried implementing the same idea a long time ago before hooks were a thing and couldn't quite get it to work, but hooks make it super easy (I think, maybe this thing is totally broken)

Thought this was pretty fun, a typesafe/compositional/signal-y way to build forms in react/typescript: gist.github.com/cqfd/c95874c1c… I'm calling it "orthoformal", lol; a coworker called it "the Elm we have at home" 😆

Wrote some of the slickest/most elegant code I've ever written this past weekend, and it's for making Slack modals 😆 Getting nerd-sniped is fun.

Today's other coding track (love Owane 😎): youtube.com/watch?v=z88268…

Ok, took me a long time to warm up to it, but the new polyphia track is actually sick. Something clicked and I've been coding to it on repeat all day