Starting a Python project in 2022 is easy - just use the arrokuda cookiecutter and execute the cresselia script inside it but dont forget to install graveler before you do or lillipup will fail when it tries to access your nuzleaf config
I would pay $20/month for AI that doesn’t sound like a fake flirty Californian but instead sounds like a guy from New England who can’t believe what you’re asking of him
Are you a junior data scientist? Double your salary with a few simple word replacements:
Instead of "data points" say "manifold"
Instead of "my beliefs" say "priors"
Instead of "I don't have enough data" say "I'm a Bayesian"
The same people who made a pile of money five years ago telling your executives that ML is the Future are about to make a new pile of money telling them that it was overhyped and that they should invest in something called "statistics"
I love teaching backend/data engineers about machine learning because there's always a point at which they go "so its just kinda wrong all the time" and I have to go "uh, yes"
“Now you can run Python and R in the same docker image” I can also order a Big Mac and chili cheese fries and eat them over my sink like a gremlin but that wouldn’t be sensible would it
A friend asked "so you're a math guy, do you have like a favorite equation?" and at first I laughed at the idea then I realized I do and it's Euler's formula
New in Python 3.10:
Every package must have a utils․py file containing all important logic and any bugs. Bugs outside of utils․py are considered unpythonic.
Every industrial recommender system I've seen or worked on was a lumbering amalgam of recommendation techniques, cultural data, product beliefs, etc
But every writeup or critique I see about industrial recommenders assumes "lol it's just knn retrieval obviously"
my favorite part of advising on recommender systems is how often I have a conversation that goes "how do you look at your recommendations?" and they go "why would we look at our recommendations?" and I say "to know if they are any good" and they say "oh, good idea"
"What is the best tool for managing python dependencies?" idk man. I mean I know but I wish I didn't. I wish I didn't know. I wish I didn't have to know. I wish you didn't ask me. I wish I were outside on the grass right now.
Today is my last day at Spotify, and I'm grateful to so many people, but most of all I'm grateful to the person who responded to my goodbye e-mail blast with just the word "bummer"
Big tech cos have such well-developed MLOps that it takes quarters-to-years to roll out ML products while national retailers' tiny tech teams have 3 people putting SoTA in prod in a week
It’s serverless!
Just make sure you leave one server on, to reduce cold-start latencies. And minimize tasks that might slow down server boot-up. Actually, leave a few servers running, for spikes. And don’t forget to allocate instances to serve your db. Also,
List of professional networking platforms in order of efficiency:
Twitter
Instagram
Standing on the Kendall/MIT train platform screaming your startup idea at the top of your lungs
LinkedIn
Early-career people will get lucky by solving problems with a certain skill. They then take the wrong lesson from this and assume every problem requires that skill. You can avoid this by never solving any problems.
I am weak for 1:1s. I will accept a 1:1 invitation with anyone. Calendar already full? No problem. I'm two weeks behind on my work? Accepted. I'm totally unqualified to help with your problems? Better make it recurring.
The best thing about
@normconf
is that it feels more like hanging out and talking shop over dinner with friends than a high-falutin high-pressure Data/ML conference
If you're a parent WFH and you have to blend childcare with meeting time then I just want you to know that yes I DO want to see your baby and your baby IS ADORABLE and can you please hold your baby up AGAIN thank you
StitchFix’s Style Algorithm uses principles from quantum mechanics and state of the art machine learning to ensure you are still dressing like it’s 2015
Hell yeah I'm a "Full Stack ML Engineer" -- I'm
confused by web engineering
intimidated by data engineering
frustrated by backend engineering
totally unaware of design
and ready to rock this linear regression 😎
I am delighted to announce that I have answered the age-old question of "Build or Buy?" and the answer is: Neither, just ruminate on it for a while, then forget
Buying an M1 MBP so I can have the exquisite experience of remoting into my M1 Mac Mini from which I can SSH in to the Raspberry Pi where Python actually works
After years of "oh you work at Spotify, you should change <thing>" someone finally told me about something that was actually my fault and that I could fix so congratulations
@jhforster
@sarahcat21
Rings true - any time a group is organized by skillset, rather than by goal/product, they'll get slammed with decontextualized requests -- and the more flexible the skillset (like broad "data skills") the more groups can make these requests
just got my emcee instructions for
@normconf
and it says I'm supposed to introduce every speaker as "the inventor of the data mesh"? that can't be right, I invented it
StackOverflow is a powerful tool because you can encounter a scary looking Warning blurb from your code and someone will condescendingly explain how to turn off Warnings.
In 2018 I met someone from Facebook's VR team and they described their cool new vision of the future and I said "oh neat, like Second Life? I love Second Life" and they abruptly ended the conversation and walked away
"Yeah, okay, so Bitcoin is money but it's on a Blockchain... Yeah, so, it's like, a digital solution to the Byzantine General's Problem... Oh, okay, so when Constantine moved the seat of the empire from Rome in 330AD..."
First presentation for my new boss today and I can't decide whether to wear Business Casual (dungeons and dragons t-shirt) or Business Formal (dungeons and dragons polo)