I played more games this year, particularly puzzle games. I also just bought a Switch 2, which seems good so far (nice screen).

Puzzle games

Blue Prince

Me – and every streamer in my recommended – played this at the same time this year. You randomly draft the floorplan of a mansion with blueprints (geddit?), which resets each day, while managing limited resources. You’ll need a pad of paper. It’s definitely a puzzle game, there are puzzles, it’s also definitely a rogue-like, it resets and you do runs from scratch. Do they work together?

Mostly for personal reference for next year, here were my xmas dinner notes. I forgot about stuffing, and didn’t make yorkshire puddings properly because our main oven is broken (we have a convection oven, which was fine for everything else). We also made gravy from granules, sorry everyone.

Following these three videos mostly:

• Potatoes: https://youtu.be/xOXfA5QewH8?si=ZchkzHeKcdJG3urP
• Veg: https://www.youtube.com/watch?v=meHfrktmvIo
• Sprouts: https://www.youtube.com/watch?v=vwcG3A6WKys

Main tips:

1. Boil potates until they are falling apart, then leave to dry.
2. Parboil veg.
3. Veg and potatoes can have prep done ahead of time.

Prep (day before or morning)

Roast potatoes

I used Maris Piper potatoes.

The rest of the UK and I have been watching the Celebrity Traitors over the past few weeks, which is basically Werewolf.

The game starts with three traitors and sixteen faithful. Each night the traitors secretly choose a faithful to kill (who does not appear the next morning). There is a round table where the group as a whole vote for who they think is a traitor, and by majority vote that person is banished, whereupon they reveal whether they were a traitor or a faithful. If at the end, a single traitor remains, the traitors win.

Last week I saw John Jumper give a presentation on his team’s development of AlphaFold 1/2/3. He started explaining how AlphaFold1 worked, essentially a convolutional neural network. He spent most of the talk explaining the many changes required to produce AlphaFold2, which went from mediocre predictive accuracy to changing the whole field.

I took away two main points.

There was no single trick that lead to the big performance gain of AF2 over AF1. No single bit of intuition that lead to such a huge shift in performance. Instead, lots of good ideas each of which increased performance slightly.

NB: mcstate is being replaced by monty which will make this easier

How do you use fixed parameter values during the pMCMC with mcstate and an odin model?

I was first asked this question in 2021 and I don’t think we’ve ever managed to more clearly document it, but the relevant part of the documentation is https://mrc-ide.github.io/mcstate/reference/pmcmc_parameters.html.

To get this to work you need to make your transform function a closure. The transform function always needs to take as input a list of the pmcmc_parameters() being inferred by mcstate, and output a list of pars as used by the odin.dust model, and can’t take any other options. But, one can use a closure to bind other arguments of a more general function to fixed values.

At school/University I had vaguely heard of the Central Limit Theorem (CLT) but never properly understood it or looked it up. My understanding was pretty much along the lines of ‘most measurements are normally distributed’.

This isn’t quite right (although as we’ll see not totally wrong). The CLT actually states that sample means taken from a distribution of measurements are normally distributed, irrespective of how the full distribution of all the measurements is distributed. So if we take large enough batches of samples of a measurement repeatedly and calculate their average, those averages will be normally distributed (with the same variance as the full distribution).