im just gonna go ahead and give you all 7s cause your renoise module is 700 bpm.

My current predictions have this track at #34 out of 39 tracks.

So if you actually did that, it's be cool to be bumped up a place or too, but I'd also lose the satisfaction of correctly predicting my track ranking.

(I know you're joking)

im very curious how you're calculating that stuff. do you use favs/comments amount in relation to rank on previous major battles?

am i fucking up your calculation if i write a lot of comments?

is this what your neural network doesnt like?

if entry["name"] == "pointy peek": entry["comments"] -= 4.

:P

But in any case, a neural network would be super overkill and overfit, it's just a linear regression over plays, favs, downloads, number of votes, number of comments.

The factors make intuitive sene to me: faves is the biggest, and they're all positive (except for plays, which is counterintuitive and maybe I should take it out).

I estimate that the error for a given rank I predict is typically +/- 30%, but it gets the very top and the very bottom pretty much correctly and
consistently.

Of course now that I've explained how it works, someone could try to abuse it, but it's not worth the effort and I can correct for it anyway :P

By the way, in my calculations, yours comes out in the middle of the pack, so I can't predict its rank very accurately (except to say that it's in the middle of the pack)

-= 5

ah! yeah that makes sense. we'll see how accurate it is c:
interesting would be a prediction regarding score as well - as in: who's over 20 and who's lower than badge progress

Relative predictions are good: I'm fairly sure the top 5 is going to
(roughly, and if people don't vote a lot at the end of the compo) be:

#1: Thomas long name
#2: Winter walk
#3: Kirby Air Walk
#4: Lazer City Aerobics
#5: Fluidman

But the absolute scores are completely messed up

https://youtu.be/1K5VntfTl9k

I'm confused - which I guess is your point. :P