Machine learning

  • Tried out vgg16 as a feature finder for the birdhouse camera, and xgboost as a classifier; 86% accuracy, which isn’t bad.

  • Set up BirdNET-Pi at home – very interesting project

Hardware hacking

  • Try setting up an MQ135 sensor prototype board and hooking it up to a Pi. Mixed results; seemed to show 403PPM, which is at least in the general neighbourhood. But it seems fussy, and takes a lot of warmup time.

  • Ordered some ESP32 chips to play with – Lolin32 Lite from Universal Solder (I’m a very happy customer of theirs). Took a while to get going, but this was mainly because I didn’t realize the USB cables I was trying it with were charging-only – or even that that was a thing. Got MicroPythong going, and wow – wifi set up right away, and with a decent range on it too. Amazing. Played with MicroDot, a web framework in MicroPython that works on ESP32. 🤯🤯🤯

Natural history