Dan Tasse
@dantasse
Followers
394
Following
758
Media
30
Statuses
892
Software engineer @LanceDB. Previously: data science @StitchFix, HCI, and geotagged social media data. Still into bikes and coffee.
Pittsburgh, PA
Joined April 2009
oh neat! this paper, more than any other, inspired my phd thesis. congrats Justin Raz Jason and Norman!
Congratulations also to our ICWSM 2022 Test of Time Award winners, @justin_cranshaw, @razsc, @jas0nh0ng, and @NormSadeh for their work, "The Livehoods Project: Utilizing Social Media to Understand the Dynamics of a City" 🏆🏆
0
0
2
neural networks are terrible houses of cards, but until today I'd never just thrown twice as much data and training time at a model and actually had it significantly improve
0
0
1
yes this thread! I've gotten annoyed with "tech debt" over the years. feels like "national debt" - we all shrug, maybe it's not even bad. but connect "maintenance thing X" to "... and then the toilet will overflow at midnight and person Y will have to fix it."
0
0
0
The (deceptively difficult) question of "what color is this item?" Now with neat algorithmic details! @stitchfix_algo
https://t.co/7igoUtVwTr
0
3
6
hey, my @stitchfix_algo post! one simple attribute for clothes is "what color is it?" but different people care about color at different granularities. how did we solve this? https://t.co/2dxJtWK4Zm plus sweet interactive explore graphics by @iPancreas
multithreaded.stitchfix.com
We need to know what colors our merch is. But because downstream users include many different people and algorithms, we need to describe colors as a hierarch...
1
1
4
TIL Fréchet is "fre-sheh" not "fre-shet" I think long ago I must have overcorrected on "not wanting to assume something's French and make a fool of myself", gotta fix that anyway I can't wait to see the new Nolan movie, Tenay
0
0
0
get data, preprocess, train and test models, output metrics and confusion matrix: 4 hours figure out how to tell matplotlib to make that confusion matrix a little bigger so I can read the labels on it: also 4 hours
0
0
4
man like 6mo ago I started using "indiana jones" as a verb, meaning "to do this with database tables", and now I use it all the time; I'm sorry database gods
1
1
4
if you run chaos monkey, but then you start depending on your jobs restarting every N hours... do you then need "order monkey" to sometimes make them run longer?
1
0
0
TIL parquet is not in fact pronounced "par-KET" but "PAR-kay." I definitely too-clevered myself out of this one for a while :P
2
0
2
I hope this email finds you living in a shotgun shack I hope this email finds you in another part of the world I hope this email finds you behind the wheel of a large automobile I hope this email finds you in a beautiful house, with a beautiful wife
343
8K
41K
this is a cool thing! sometimes, in addition to metrics, you need to just look at the thing. (often, I'd say.)
The latest blog post from the Stitch Fix Algorithms team: "A picture is worth 1,000 false-positive bug reports" by @iPancreas. https://t.co/m0uZZmEgnZ
0
0
1
if I had a dollar for every time "statistically significant" just meant "we should care about it", then I'd have a statistically significant amount of money
0
0
4
one thing I learned from studying neighborhoods is, lol good luck posting a "correct" neighborhoods map:
0
0
2