Konsta Happonen

Dr. Object of type 'closure' is not subsettable

Mostly sound principles! Especially the one about opinion pieces. If I want to read a random person express unfounded opinions about things, I come to Bluesky.

I still ocassionally think about the first Twitch Plays Pokémon stream and how much fun it was.

NO is the correct answer to the lure of the dome fossil. PRAISE HELIX.

I wish I had an excuse to write something in Julia.

My experience with Bayesian latent factor models is that guaranteeing convergence across multiple chains is hard due to multimodality, but can be facilitated by smart priors and setting of initial values.

The blavaan basics and examples sections are good.

blavaan.org/index.html

After reading, I first laughed and then wanted to cry.

The linked SO post exemplifies why social media can both increase and decrease your wellbeing.

True! But there's just so much cool stuff in the world and so little time!

insert we heard you like meme here

If I had to do factor analysis, I would definitely use the opportunity to go Bayes and use a specific model comparison technique just so I'd get to write "We used Bayes factors to compare the Bayesian factors" in the methods section.

A colourful 3D pie chart.

Always a joy to spot 3D pie charts in the wild.

Maybe on Sunday.

Confused why I can't seem to find time to do cool new side projects this week until I realized I have a full-time job, a part-time PhD project, courses on opera singing and plainchant, choir practice, and a ballet to go see. Why am I like this.

If the spline is not wiggly enough, you can increase the prior on its variance by letting it sit in water overnight.

Can't talk, fitting splines to data.

Trademarking "natural learning": all of our trees are hand-grown and carefully pruned by our natural learning experts 🌱

It's actually what got me opening a tst.jl script on my computer.

Creating variables that have greek letters as names.

I want to be able to do stuff like this, but I'm told it's not a good idea in #rstats >:|

If I had time to learn a new language, it'd probably be Julia. Looks cool, syntax makes intuitive sense, doesn't have weird quirks "for compatibility with S", seems to have some powerful packages for e.g. probabilistic stuff, agentic modelling etc. And unicode variable names are fully supported!

Also, editors who merely forward (often contradictory) reviewer comments to authors without any advice on how they expect the authors to interpret them are IMHO not doing their (admittedly, unpaid) job.

I've always felt that its weird how many editors seem to outsource editorial work to reviewers. Like, I've always felt uncomfortable recommending rejection, revisions or acceptance. Isn't it the job of the editor to decide where the bar for their journal lies?

Thing that goes up;
Thing that goes down;
Thing that stays about the same;
Secret fourth thing;

This is why we have terminals and Rscript!

I have a theory that the reason Finland is often ranked the happiest country in the world is that our most common grading scale runs from 4 to 10, which effectively truncates the lowest values.

What if the thing I like is hitting myself in the face with a hammer

the assignment operator is a function and you can overwrite it in the global scope with '<<-' if you need to brick a session:

> unlockBinding('<-', baseenv())
> '<-' <<- function(x) 5
> a = '<-'()
> a
[1] 5

Next time I see R installed on a public computer, I know what I'm putting in its .Rprofile