I truly hate dealing with dependencies. Trying to get to grips with some neural net techniques for machine learning this week. The most frustrating part has just been checking you have the right library installed, or then you think you do, the code seems to run but hangs endlessly, so you do some googling and find out you have the wrong version installed. Then I threw in trying to do this on the university’s GPU compute cloud with Docker images in the mix, and I can’t keep track of the images. It’s a mess, I think I’m mostly there now.

For our group reinforcement learning project we’re aiming to apply PPO to Super Mario Bros for the old NES. This is a pretty feasible problem with plenty of demonstrations around, just steadily working through it at the moment! The PPO paper is one of the most confusing pieces I’ve read. It makes it so hard to understand how the algorithm actually functions, while somehow giving enough info to recreate it. I’m very grateful to Phil Tabor for giving an implemented demonstration. There is also the Stable-baselines version, but that’s significantly harder to read.

Alongside working on that I’ve been steadily progressing with my lit review, mostly making my daily word count. The Easter break comes to an end tomorrow, so it’s a 3 week slog to the end now. I have 4 pieces of coursework due between 26th April and 5th May, and I’m currently 0/4. Maybe 0.5/4. Everything is in motion, I just need to pick the speed up a little. It feels like my skills are really improving at the moment, which is the benefit of the work.

My life is still pretty much just the MSc for the moment. Lockdown’s easing another stage tomorrow, so I might be able to be a little less hermetic following that. I’m finally getting insured on my Dad’s car so I can get some practice in prior to my test at the end of May. Getting through the days.

Reading list