While running errands, the radio was doing that thing where it's mostly playing one station, but then it briefly cuts over to a different station for just a single word like, "confusion" or "radar". I think whatever the other program may have been something like a very slow paced word based game show.
That might be long term motivation to port the whole codebase over to something that can run on my laptop.
Part of the performance surprise is that I'm prototyping the algorithms on my laptop which granted is a few years old, but was pretty high end when I got it and the math library in use sees what I'm doing and says, okay, I'll just offload all that to the GPU and I get my result instantly, then I try it on the hardware it needs to run on and all of that gets pushed back to a much slower CPU.
First test batch on the new algorithm ran into performance issues that makes it kind of painful to use, but it slashed the error in half compared to the original code. With a sample size of 1 it's not as if that's meaningful, but I've made some adjustments aimed at improving performance and I'll see how that does after staff lunches are finished.
Spent some time working on another new approach for generating measurements from the roaster cam. Results on some previously gathered test data look a lot better (25% error reduction, but I suspect I can push that farther by cranking up a parameter that I just didn't have enough data to push farther). I'll take my best guess at an implementation based on last night's exploration and work on getting a larger set of testing data for work on future refinements.
In normal times, the fire extinguisher maintenance people just come on a schedule, do their thing, and let me know if something more expensive than usual is coming up or required. If they don't show up and we don't have a fire requiring an extinguisher, the only people double checking that are the fire inspectors.
I have an initial set of test images to work with now. Sadly, the information I wanted in the file names is not what got put there (the original code was placed after the relevant variable had been re-used for something less useful and I failed to notice that when adapting it). It's not a big deal because I can just replicate the code and have it load data from a file instead of from a camera, but it's more work to get started on the analysis than I hoped.
Author of Typica software for coffee roasters.