mikedecr.computer
Quant finance researcher/developer in Chicago.
Former political scientist, sometimes bike rider.
Longer form blog stuff at mikedecr.computer
260 posts
198 followers
228 following
Regular Contributor
Active Commenter
comment in response to
post
you mentioned docker elsewhere, maybe that helps a lot; you have the feeling of the monolith env inside the VM
comment in response to
post
I found CRAN very convenient in school when I just kinda worked out of a monolith env, but I think if I had to pin backdated pkg versions in multiple parallel environments I'd struggle. But I migrated to Python before I strengthened those muscles post grad-schol so maybe it isn't so bad? You tell me
comment in response to
post
i can agree on the wackness of pandas though.. feels awful
comment in response to
post
UV is certainly easy for Python-only stuff but for cross-language things (like C-universe dependencies) conda works fine. I use use conda (micromamba) to manage R projects even. I think it's totally reasonable and more predictable than things like renv tbh
comment in response to
post
when you're doing stats in python this just really isn't a mental hurdle IME. users are very accustomed to bringing in certain packages or namespaces (even in std lib) for specific things. I suspect it's less Rational forces like dependencies and more miserly micro-econ/social things that explain it
comment in response to
post
I guess that's a little goofy but it at least feels binary in nature. The thing that's got me weirded out is that max/min could be n-ary arguments right so why not something prefix style like \wedge uv which would allow \wedge uvwxyz etc
comment in response to
post
the extent of my knowledge about this notation is this post only, but it’s kinda bothering me that it’s written infix-like? If I knew more would that still bother me?
comment in response to
post
one cool thing: in bsky settings/accessibility you flip a switch to REQUIRE alt text on any image that you post, so you *can't* forget to add it
comment in response to
post
dang are we neighbors? I know exactly where left alley is, it hit me like a bolt of lightning
comment in response to
post
i installed this colorscheme forthwith
comment in response to
post
"so we agree, popularism works"
comment in response to
post
He who would silence pundits should take care lest he himself do punditry. When you gaze long into the timeline, the timeline gazes also into you.
comment in response to
post
when I type x[i] the AI tells computer where to find data
comment in response to
post
The theoretical model under this is like, someone's idea of "best practice" might actually be a combination of N separate things but k<N of them are the things that have the biggest impact, so just do those things
comment in response to
post
it's been working pretty well tbh. it lets me stand up new features pretty quickly, I won't immediately regret the *structure*, but I won't waste time on making the hidden code that nice. not perfect but an optimal balance so far between speed and maintainability.
comment in response to
post
cookin em
comment in response to
post
The way I often say it to friends is like “R has many lessons for other languages but I don’t want to use R”. For example, handling expressions as their own thing in an environmental evaluation model has been huge for data frames. The main reason why pandas feels like prison and polars feels better
comment in response to
post
Learning more about computing broadly at work has increased my ability to understand and appreciate and leverage base R for sure
comment in response to
post
tldr won’t someone think of the held out data
comment in response to
post
The world of toy models is one thing. Outside the toy model, the model is useful it helps us forecast the effects of new interventions. I think explosive coefs from high correlation is only benign if you KNOW that new data follow the exact same DGP as training data. But.. reality has a way of biting
comment in response to
post
when the completion menu appears and it steals scope from the tab and directional keys? sinful. what if I want to ignore the completion and do a normal tab? or a normal down key? let me do what I want to do.
comment in response to
post
idk... banning trading has the smack of cheap talk to me. it isn't really a high salience issue on which dems are notably disadvantaged, so it not much distance from a bad party reputation. and it has the vibe of "low subjective probability that anything will happen about it" even if the cand. wins
comment in response to
post
at a high level it's basically this
comment in response to
post
who was the economist that said that if you make communication costless then almost all communication you receive will be worthless (original context was about email)
comment in response to
post
this gif apparently don't animate I guess