@rosezeemore I came out 3.5 months after cracking. I started HRT, exactly 1 year after cracking.
One of the wild things is looking at my selfies from first time in a dress to the day before I started HRT.
It was as big a shift as the first year on HRT after.
I got good at presenting.
@rosezeemore It was *unfun* but still better than living as a dude for me.
I definitely didn't pass and the UK is not kind to clocky trans girls
And like i proved i could.
And I could again if I had to (goddd don't make me have to).
They can take my hormones but they can't take my identity
we did use it when I was at Invenia for test data.
And I did (mostly very shortly after starting) make some minor changes to it for Invenia, mostly just to make it support auth'd AWS3 via S3Paths.jl
(or anything else that overloads `download`)
I have for new job been using DataDeps.jl
it is nice to just pick up a package I wrote during grad school and have barely touched since,
and it still
1) works
2) is completely fit for purpose.
It's like I want someone else's data on my computer and I don't want to think about it
@jamimmunology@wanderer_jasnah@o_guest mediocre white men at scale!
They have all the over-confidence of a mediocre white man, and air of acceptability, while stile being mediocre.
However, society has never been bottlenecked on mediocre white men. This is now something we needed to scale
Fundamentals of Systems Biology (2015) by Markus Covert
Equation 6.16 is an implementation of sampling from a categorical variable.
Its a medium smart one.
But its not the smartest (alias tables are the smartest).
I feel like the implementation should have been a sidebar
@Spiindoctor I would lean more towards saying things from the Decision tree family (e.g. Gradient Boosted Trees, random forests) (optionally with PCA) outperform MLP when the amount of data is relatively small vs the complexity of the error surface, for similar engineer effort
After many years I am back in classical ML land.
I return to a question i had back then.
Is it a standard technique to orthogonalize features before applying Decision Tree type methods?
(e.g. run PCA first)
It makes excellent sense to me and seems to work.
Does it have a name?
So Runge-Kutta methods are just midpoint methods with more sampling points? and slopes caculated from other samples to workout where next samples go?
And usually, adaptive time stepping?
@evalparse right, but the particular combination of PCA (or other orthanonalization) with decision tree like approches, for axis alignment, rather than for dimensionality reduction,
should have a name
me trying to do things in python is very funny.
Its like "How do i install mamba? why are all the instructions deprecated? Did i install mamba wrong? oh no i did"
urggg breaking changes in python 3.12, distutils standard library was removed.
This language is so unstable, people shouldn't use it.
It has broken one of the tools i install through apt on ubuntu's package manager.
Why ubuntu not bundle compatible versions?
I was a good girl and generated new SSH key pairs for new laptop rather than copying from old.
And I will starting next week void my old key pairs wherever i find them.
@just_cameron Its legit hard to. Like I been there babe.
androgens increase sweat and oil production, and add there own distinct musk to scent.
So men's deodorant actually legit works harder.
It's why my wife has used it always.
There is a lesser and different musk from estrogens, also.