Hi friends!
Happy lunar new year!
Jamie Brandon and I spent a few months in 2018 collaborating on Reltron, a relational database GUI.
What if you could just open up a new database, immediately see the data, sort/filter, and join, without having to muck around with SQL or fully internalize the schema?
Check out a 10m demo video and my notes on what worked and what didn’t: http://kevinlynagh.com/reltron/
Most of my career has been on the web, which is a gazillion layers of abstraction away from what computer hardware is actually doing.
When I came across a perplexing Rust pull request — changing a simple match statement to a lookup table improved performance by 20% — I thought it’d be a great excuse to explore “under the hood”. I ended up looking at both the assembly code (CPU instructions) as well as the LLVM intermediate representation (which is nicer to read, but also sort of mysterious and low-level).
I mean, how complex can a function like:
fn match4(x: u8) -> u64 {
match x {
b'A' => 1,
b'B' => 2,
b'C' => 3,
b'D' => 4,
_ => 0,
}
}
really be?
Well, the article is, uh, 5000 words and ends with a bunch of open questions: https://kevinlynagh.com/notes/match-vs-lookup/
Neat things from other people that I’ve found this past month:
Noise2Self: A principled framework for removing noise from high dimensional data without a ground truth reference. I am a little baby when it comes to understanding statistics, but the gist of this approach is pretty sweet and seems widely applicable.
Comprehensive human cell-type methylation atlas reveals origins of circulating cell-free DNA in health and disease: Looking at methylation patterns of free-floating bits DNA can tell you which part of your body they came from (i.e., which of your cells are, uh, dying/exploding)
Gradient boosting machine learning. Deep neural networks are on trend, but a friend of mine in the ML space mentioned that gradient boosted machines are pretty sweet and easy to use for lots of problems. How to explain gradient boosting is a wonderful written intro, and I also enjoyed this Trevor Hastie talk on the subject.
The problem with price controls. I’m not an economist, but as someone who writes Clojure I’m always sympathetic to an argument of the form: “What if we took these things apart?”
Colliding blocks compute π. Math puzzles aren’t normally my thing, but I decided to take this on to stay sharp. Two Saturdays later, I have a few dozen notebook pages covered in algebra scribbles and misshapen circles. Next week I plan to expand into push pins on a corkboard covered with twine and photographs of dead presidents. So, uh, be careful.
Best,
Kevin