Imagination is the only weapon against reality. @petrisphilter - Tumblr Blog

Thanks for tagging me @do-androids-dream-ao3acc - I’ll take that as a fair excuse for a study break :D

17 questions 17 people

Nicknames: Ironically I don’t like nicknames that much, but my friends and family have quite a few for me. But then again, I have a pretty long name, so it’s not surprising. My favorite one is probably Murka, which has nothing to do with my name, but it’s the name of a pretty famous song (and I kinda like the tune). I mean, it’s actually the name of a criminal, but nvm I guess :D

Zodiac: Virgo

Height: 5′3 (1,60 m)

Last thing I googled: “Dark” in medieval Occitan (for writing and procrastination purposes)

Song stuck in my head: You’re gonna go far kid, The Offspring

Number of followers: yeah, about that... we don’t do that here :D

Amount of sleep: precious little (I’m a student shortly before their next exam. We don’t sleep. We live off caffeine and panic)

Lucky number: 8

Favorite song: per genre, right? Nah, srsly, way too many. Some of my favs are Zombie (The Pretty Reckless, tho I like Zombie by The Cranberries too ofc), Rusted from the Rain (Billy Talent), Es könnte schöner sein (Faber), Es wird ganz groß (Faber - actually if the question was who’s your fav singer the answer would be Faber).

Favorite instrument: violin and piano. I played piano when I was a kid but I was lazy af, so I quit, sadly

Dream job: Data science/market intelligence

Aesthetic: depends on my mood, I guess?

Favorite author: Can we make it favorite book? The answer would be so much easier (Le Rouge et le Noir by Stendhal). Idk... I’ll go for Heinrich Heine

Favorite animal noise: a Great Dane’s bark (yeah, I’m just a wee bit weird)

Random: When’s normal English become so friggin hard? :’D

How it started: what exactly?

How it’s going: Could be better, could be worse. In my dialect we’d say passt scho :)

I really don’t know who else to tag, so yeah, whoever reads this, feel free to join in :D

#random #games #personal

stannisislovestannisislife

calyxestra

hexersource

NILFGAARD’S ASSASSINS

We do what we must. I am not ashamed of that.

elisabethwheatley

yennderburg

LAMBert

ghostcupdraws

«As far as the foppish Dandelion was concerned, he was regularly mistaken for an elf or half-elf, particularly since he had begun wearing his hair shoulder-length and taken up the habit of occasionally curling it with tongs» (Baptism of Fire)

I’ve been dying to draw book!Jaskier. This idiot man.

Replaying the Witcher. Me:

Must safe Ciri

Must fulfill Olgierd's 3rd wish and safe his ass from Gaunter O'Dimm

Must find the Beast of Beauclair aka help Regis with his emotionally unstable friend Dettlaff

Also me:

Runs around Toussaint like a mad man to get all the Skellige Gwent cards.

#witcher #witcher 3 #gwent #gwent is life #priorities #random weirdness

Mysterious hat man strikes again, this time in Vizima

His mission: annoy the crap out of Emhyr avenge the North by going full Vesemir on Emhyr

#witcher 3 #witcher #witcher 3 mods #screenshot #lambert #lambert lambert what a prick #emhyr #the emperor's not amused #it was geralt's idea

ask-the-witcher

Does Regis name his ravens?

*ugly crying*

spiritcaesar

Glad to see I'm not the only one bothered by this

visualizing-archive-data

The One-Shot: Can Popularity be Predicted? (AtLA Edition)

Let’s find out! First, I scraped metadata from all one-shots in AO3′s Avatar: The Last Airbender tag, excluding crossovers, to get about 10,000 examples. Here’s the breakdown of what some of the parameters look like, and how they correlate with one another (bright colors mean they’re co-dependent)

Of course, a lot more parameters than these went in to the model! Every character that was tagged more than 15 times (excluding only ‘mentioned’ characters), every ship (romantic and platonic) that occurred more than 15 times, and every tag that occurred more than 100 times went in to the parameters.

The biggest problem is actually what to fit. Hits? Kudos? Bookmarks? Kudos is the most consistent AO3 metric these days, now that they did away with multiples, but the way hits have been counted has changed over the years. Bookmarks is usually solid, but what I don’t like about that is that some people use it like the ‘Mark For Later’ button, and private bookmarks can’t be counted. After a few tries and not very good models, I settled on fitting for the kudos-to-hits ratio. In this dataset, the average value was 1:10. I excluded stories less than a month old and more than ten years old, and those with more than 20k hits (sometimes this many hits results from bots, but it also serves the purpose of removing those uber-popular fics that skew the data) just to try for a bit more consistency.

Then I fit a multiple linear regression model to part of the data (the “training set”) before seeing how that prediction did on the rest of it (the “test set”). The model’s coefficients can give an idea of what parameters influence the fit to the data:

Hmmm. The first variable, author_idx, was constructed by sorting the data by number of bookmarks and then numbering the authors in the order that they appeared, so that authors who wrote fics that got a lot of bookmarks would have a higher number. Previously I did this in a random way – I didn’t want the model to depend on any external metrics like bookmarks, hits, or kudos, but doing it like this really did make the model better. So how good is the model anyway?

Well… not that great. It only got 25% of the predictions on the test set right, with a standard deviation of ~ 0.5. So it’s usually correct to within a factor of 2, but that’s pretty big – getting 1 kudo for every 5 hits is much more than getting 1 every 20!

Let’s look at some of the outliers to see if this tells us anything.

It definitely couldn’t account for the crowd appeal of some of these stories! But given that the distribution is normal with a tail, I think this is excusable.

So what makes a one-shot popular in the AtLA fandom? It looks like being a popular author with a good bibliography of one-shots helps your chances a lot… which definitely feeds my theory that popularity breeds popularity, unfortunately for the rest of us. But we’ll see what a neural network has to say next!