Stephen Martin

Probabilistic modeling and machine learning nerd

Social Psych PhD, quantitative methods

Data Scientist Manager, Comscore - Innovation Research Team.

[Stephen Martin @smartin2018]

@ajordannafa version of it, with corresponding docs.

[Stephen Martin @smartin2018]

@ajordannafa Btw, I suspect this could be a really fun basic LLM project for anyone wanting one. It would not be hard. LLMs do a decent job at converting between the two.
So, one could very much so dump out a list of target functions and ask the LLM, for each one, to write the other lang's

[Stephen Martin @smartin2018]

@SolomonKurz I did something like this too, for my dissertation. I had 3 studies or so, each one had a similar experiment. I did what I called a "full information meta analysis" where I "hierarchicalized" my logistic MLMs (including the RE correlation matrices!) to be across studies.

[Stephen Martin @smartin2018]

Ifixit is such a good company. I've now fixed two devices using ifixit tools, replacing parts from ifixit kits. Much $$$ saved. Easy guides and parts. They're well known in geek circles but if you're even remotely handy, and you break your hardware, you really should try a repair

[Stephen Martin @smartin2018]

@ajordannafa I assume it's the same reason, no? Seems like a lot of 90s songs were a half step down, so if you want to play standard it's faster to just capo.
I assume there are some noticeable differences in tone from standard, but I have to believe at one fret, it's just convenient

[Stephen Martin @smartin2018]

@tcarpenter216 Hm. I haven't seen this issue when doing something similar for my wife, but what FPS do you record SLR and iphone at? Curious if there's some weird interaction between the software you're using to stitch it together and the divisibility of the fps and sampling rate.

[Stephen Martin @smartin2018]

@just_cameron I have one blue nalgene that was swag from my visit to Valve (I used to be a Steam community global mod). I have used it every day since I got it nearly 8 years ago. I don't understand how people have so many water bottles.

[Stephen Martin @smartin2018]

@daranzolin I really don't like reading rlang esoterica. quo, ensym, etc are decent for when you need to program around the tidyverse, because they decided to use NSE for nearly everything, and the alternative base syntax for NSE is arguably worse. But to use it as a primary api? ew

[Stephen Martin @smartin2018]

@samreich Jeff Arcuri
Such a fun positive energy too.

[Stephen Martin @smartin2018]

@dustinvtran ! I didn't know you worked on that, congrats man. Bayesians do it right.

[Stephen Martin @smartin2018]

@PhDemetri I really wish the US had a mandatory paid sabbatical for everyone once every few years lasting for a few months. It does wonders for recentering yourself, getting motivated, feeling happier, recharging, etc.

[Stephen Martin @smartin2018]

@PhDemetri There was a house listed here for about $500k. One of the cheapest in our market. Modest, maybe 2-3 br, 2 bath. Not huge.
It was missing flooring in one room.
Missing. Floors. In one room.
"Fixer upper", "quaint", "character".
$500k.
It's all cursed.

[Stephen Martin @smartin2018]

@just_cameron "start" shooting each other?

[Stephen Martin @smartin2018]

@just_cameron Absolutely cursed, but I love it, not gonna lie. This is the type of "mostly pointless, but extremely clever" project that I love. Truly embracing the hacker culture.

[Stephen Martin @smartin2018]

@tcarpenter216 Buddy, our last 2 places have had 70-era brown carpets and tan walls, and we are *upgrading* to grey. Compared to brown, brown, brown, tan, brown, yellow; be happy with the white and grey.

[Stephen Martin @smartin2018]

@just_cameron And you really *cannot* do that in Python, and it's why Python has so many numeric systems and autodiff systems.
R allows some parts to be swapped out or upgraded for free, but the underlying C ultimately is the limiting factor to, say, making everything GPU-able by subtyping.

[Stephen Martin @smartin2018]

@just_cameron This is the part that I use to explain why functional style programming + lazy-expressive programming can be amazing. R has this ability too, but it's limited compared to something like Julia.
You can load packages that allow nearly everything to be autodiff'd for free.

[Stephen Martin @smartin2018]

If the lower bound is L, and length is K, and V is an unconstrained simplex

L + V*(1 - K*L)

Gives a lower bounded simplex.

Figured that out today when trying to implement a constraint in a differentiable way in torch.

[Richard McElreath 🐈‍⬛ @rlmcelreath]

This is it, what I've trained for