Monday: It's like madness designed this thing.

There are two main classes in the script, and they do similar things.  They both encode the model as a series of polynomials.  The first one uses a dict of arrays, and packs the two polynomials is uses as [a0, a1, a2, b0, b1, b2].  Fine.  The second one packs the two polynomials into a dict of array arrays, such that there is [ [a7, a6, a5, a4, a3, a2, a1, a0], [b7, b6, b5, b4, b3, b2, b1, b0] ].  First up, I'm pretty sure that you could constrain this model with a far lower order curve.  Second, why use such different formats?  Third, the second model uses millimeters for the first polynomial, and micrometers for the second.  Why be so difficult?

It was also super rainy this evening, and this is the sky lit up by a lightning bolt out of frame.

• This was a good episode of Supergirl, and I liked Supergirl driving in to the white martian base in a car while doing the Bugs Bunny thing.
• This is kind of a cool thing, and I might steal this implementation (without the shiny interface, because I don't need that).  The syuzhet library it uses for sentiment analysis looks interesting.
• I had the exact same thought watching the first new Star Wars movie.  These are big complicated machines, and even though they're designed for "wars around stars", all the things they have are important parts of those machines.  You can't just have something not work.  If you don't need that thing to work, then it probably isn't going to be in the design.