Too bad they stopped making rollerblades in 2006, they made good stuff

I've heard the t in "Chat" the not being silent in that context, so really it's "chate j'ai pété" ...

👀 en.wiktionary.org/wiki/queef

If it was a nullable context and you cleared it in Activity.onDestroy(), I'd call it a dirty hack, pretend to be mad then wink and let you get away with it.

But, alas, it's a lateinit, which means I have to send a murder of crow-canaries hunt you for sports

For me this happens when the network is spotty while I foreground the app and some initial request fails.

I never relogin, I just kill & restart until I have good network and then I'm right back in the feeem

💯 I recently saw a solid senior new hire impacted by layoffs because they spent too much time doing this vs "executing".. 😞

On bugs: trying to teach folks that the jobs is not fixing, it's root causing. Obvious in hindsight but not common industry behavior..

bsky.app/profile/p-y....

Besides my deep technical work & impact, my biggest career boosts came from being straight with people, especially leads. The higher in the hierarchy, the less "real talk" they get, the more they tend to value directness (good leaders anyway)

Yep and that wasn't a dig at you, I know what you mean. But too often folks are afraid of breaking the mold of being nice, won't speak up and that leads to bad decisions.

It's pretty wild that my biggest career advice, especially for American coworkers, is the opposite: don't be nice.

Be fair, be kind, but never hesitate to question authority, colleagues making vague statements ("for some reason"), call out errors, etc

US org dysfunctions stem from ppl being nice

Ooooh the headaches of having a phone like UI in portrait and tablet like in landscape.. good times

Something like this?

bsky.app/profile/p-y....

ugh I meant @jakevdp.bsky.social not typing a URL sorry

E.g. you have 20 "fast" frames, 10 "not fast" and 2 "kinda slow" => use bootstrapping on each cluster, then merge back into a distribution with the same ratio?

Not sure it's worth going further down that rabbit hole..

That's cool.. but we just said frames aren't independent! We can't randomly pull from the entire source distribution to bootstrap.

So it seems like we'd need to split the source data into clusters, bootstrap each cluster separately, then merge & diff?

Bootstrapping would mean using the frames from the "before" & "after" benchmark to generate synthetic "before + after" distributions of frame durations, diffing their P50s, and getting a distribution of diff from which we can get a 95% CI

When I started looking into this, I was working of benchmark results from a benchmark that only had a single iteration with N frames.

ChatGPT suggested using bootstrapping, and happily wrote the code for me.

(thanks @rahulrav.com for the pointer to jakevdp.bsky.social's great talk!)

That's where ChatGTP pointed me to the Hodges–Lehmann estimator. It seems to be a fairly simple concept (see screenshots)

It's tempting to diff the medians from the 1st iteration of "before" & "after", then 2nd iteration, and build a distribution of diffs that way.

But.. that would assume there's pairing of iterations between "before" and "after". Can't be, we said iterations are independent!