Here are a couple of things I did yesterday! some logging improvements I was having a problem on my server where I was logging in 3 different ways: Rails was logging to a file called production.log my Go proxy was logging to Docker Compose’s default log thing (whatever that is) my Firecracker manager service was logging to journald This was very annoying because I was constantly confused about where to look for logs and it was impossible to see everything in one place.

a silly Rails login bug (or: pairing is magic) On Wednesday morning suddenly I wasn’t able to login in my dev environment, even though it was working in prod and it had worked the day before and I thought I hadn’t made any changes to my login code. The error was really weird though – it was telling me that it was expecting a provider column in my users table (for the OAuth provider), and there wasn’t one! How could this be?? I was sure hadn’t made any changes to that table.

Hello! Right now I’m back to working on neural networks with sketches of faces. current goal: cluster the faces somehow As a starting point, I thought it’d be fun to, instead of generating faces, get the neural network to do some unsupervised clustering of the faces! The idea is: Get the Machine Learning to cluster the faces into groups See if I like the faces in some groups more than others If I do, then maybe just train a model on the cluster of faces that I like how do you actually do clustering of a sequence of vectors though? The usual way to do clustering is with k-means or something, but these drawings of faces aren’t a single vector, they’re a sequence of vectors! So k-means wouldn’t make any sense.