Friday Blast #45
Digg’s v4 launch: an optimism born of necessity (2018) - I love war stories like these. Digg was big in the day, before something happened and it wasn’t. The massive and borked v3.5->v4 migration was widely blamed for it’s downfall, though the story isn’t always as simple. An interesting view into why things broke. And Python’s stupid default parameters behavior.
Vue.js: the good, the meh, and the ugly (2018) - Nice overview of where Vue.js is right now. I do think that you’re basically OK with any of the big choices nowadays - Angular, React or Vue. In RetroFeed, I’ll actually be experimenting with doing less SPA magic, and more good-old-fashioned rendering, cause I think there’s a pretty good sweet spot to be had with React in this fashion.
H3: Uber’s hexagonal hierarchival spacial index (2018) - an overview of Uber’s H3 library, whose name is an homage to Google’s S2. Both are geo libraries which offer various forms of geo-hashing - mapping coordinates to strings with nice properties and back again. H3 seems to be the more advanced version now - it uses hexes and has all sorts of nice error properties, whereas S2 uses quads.
Least squares solutions to over-or underdetermined systems (2018) - one of my biggest “aha” moments was when studying applied linear algebra and approximate solutions to linear equations and seeing that sometimes you don’t need the exact solution. And getting one which is close to the one sometimes is good enough. And many times is the only thing you can get, as there’s no real solution to speak of. There’s a lot of power in this idea of not wanting the best solution, but still finding a good one - from approximate methods to NP problems to Baeysian optimization etc. This particular case is just for linear systems, but the same thing appears a lot in statistics, ML etc.