
Building ML tools @reddit. {📊, 💻, 🛠️}. Previously: @spell_ml, @quiltdata, @recursecenter, @Kaggle, @NYCAnalytics.
Brooklyn, NY
Joined February 2010
- Tweets641
- Following531
- Followers569
- Likes203
Someone made a Polandball dataset. 😍
kaggle.com/zimonitrome/polan…
It is never not time to retweet this straight 🔥.
Chuck Schumer has spent $8,600 on Junior's cheesecakes in the past decade trib.al/pfcEHGh
Aleksey Bilogur retweeted
Replying to @lindsey
I started my career in telephony. When I was an intern in the mid-80s, a lot of people (not employed at AT&T or the baby Bells) used the term "Bell-shaped head" to refer to a particular engineering mind-set...
Randomly been thinking a lot about Elfen Lied the past couple of days, a gory anime that I watched on Google Video (remember that?) when I was 12.
Which IMO missed an opportunity to tell a thoughtful character growth driven story by going for shock value instead.
Building a personal document search engine as a side project, and it's been opening a nostalgia-filled window into my past.
This NYT article on the effect Amazon is having on branding is a real trip into a part of the world I hadn't ever given that much thought to before.
nytimes.com/2020/02/11/style…
Every six months I clear out my email subscriptions...and notice how many mailing list I'm on that I never consented to. 🙄
Having finished "The Rust Book", writing toy things in the language, feels like writing Rust code requires >120% of my mental capacity.
Will get better but the time-to-acclimatize is...a lot!
In film, practical effects (effects done "IRL") are considered "more realistic" than digital effects.
I think DL applications will (eventually) extend this idea to new industries.
Started learning Rust this weekend! Been a long time since I learned a new language and I kinda missed it.
I built a thing! paintwith.spell.ml/ is a little web application that lets you draw your own landscape paintings using GauGAN hosted on @spell_ml.
A few things I learned working on this project:
* HTML canvas is neat.
* Modern DL techniques can get interesting results even with just 250 data points (albeit with not-so-great resolution).
* Wow, GANs are reaaaaally hard to fine-tune.
I uploaded the (hand-built!) Bob Ross segmentation map dataset to @kaggle, check it out if you're feeling adventurous.
kaggle.com/residentmario/seg…







