BurgerQuest posted:

BurgerQuest posted:

BurgerQuest posted:

kicking it up a notch, how many hours have they actually handed out this year? sorry burt, you ain't even close.

LITERALLY A BIRD posted:

I don't know if I ought to be pleased to see I don't make the top thirty, or a little disappointed. I do like stats.

Wes Warhammer posted:

It's really cool to see all of these stats in an easy to understand table like this. I like the one FYAD (i presume) IK who gave two 100k-hour probations, four sixers, and nothing else

In Training posted:

What is this

sourdough posted:

Not op, but what it should be is:

Each user is a point in n dimensions, where n is the number of different probation lengths (and bans? Dunno) and each user's value is the number they've given out. So you could be like 100 in the 6 hour, 1 in the 24 hour, 10 in the ban, etc. PCA tries to find common patterns in those dimensions. Like, maybe IKs tend to give a ton of 6ers and bans and nothing in between, old admins usually gave week or month or ban and not much shorter, something like that. The pattern that explains the most variance in the data will be principal component 1, the next principal component 2, and now you can take the n original dimensions and visualize them in just those 2 PC dimensions. Let's say "IKs" and "old admins" were the patterns that were most dominant, now each user being plotted would be how similar they were to the IK style of probes (lots of 6ers, lots of bans, not as much else) on the x axis and how similar they were to the old admin style on the y axis. Points close to each other are users that have similar probe patterns, and that fall in a particular spot on the spectrum of "like IKs" and "like old admins."

Uh, those lines shooting off labeled with the probe durations were probably him trying to show how the principal components related to the actual probe durations, but they're pretty unreadable. So it's tough to tell from that figure what the actual principal components mean in terms of patterns of probations.