Bayesian Prediction for Winds of Winter

[Symon]

Abstract: Predictions are made for the number of chapters told from the point of view of each character

in the next two novels in George R. R. Martins A Song of Ice and Fire series by fitting a random effects

model to a matrix of point-of-view chapters in the earlier novels using Bayesian methods. SPOILER

WARNING: readers who have not read all five existing novels in the series should not read

further, as major plot points will be spoiled, starting with Table 1.

http://arxiv.org/pdf/1409.5830v1.pdf

[m'lady Melisandre]

:bowdown:

[Civility Reigns]

An interesting read. Thank you for posting this.

[cxvb]

Awesome!

[Lady Blizzardborn]

Yep. We need the next book, pronto!

Also :o math!

Makes no sense to include Ned in this. He hasn't had a POV since AGoT. Also...he's one of the few characters who it is fairly safe to say is going to stay dead.

[Rhaenys' kitten]

Interesting, but it doesn't take into account that books 4 and 5 take place concurrently and that's why some POVs have 0 chapters in one or the other.

[ogi]

Interesting, but it doesn't take into account that books 4 and 5 take place concurrently and that's why some POVs have 0 chapters in one or the other.

Exactly. Still an interessting read though, a pity that I'm not into all that Bayesian stuff ;)

Another minor point in case the author follows here: Save the R graphics as PDF ^^

[Ser Cold Fingers]

My head hurts now.

[Red Wedding Cake]

Interesting, but it doesn't take into account that books 4 and 5 take place concurrently and that's why some POVs have 0 chapters in one or the other.

The data addresses it with data smoothing (fudging) for books 4 and 5 happening concurrently. Poisson distributions and probabilistic frequency is great for, say, how many customers you can expect in a grocery store in a certain hour on a typical day of the week. But this is a book series with considerably less data, albeit a book series that has more data than most.

I'd like to see how often non-POV characters are brought up to predict their importance, but going through the chapters to find them is considerably more difficult. It would give a larger sample size than simply POV chapters.

Makes no sense to include Ned in this. He hasn't had a POV since AGoT. Also...he's one of the few characters who it is fairly safe to say is going to stay dead.

Ned and Catelyn are important for fitting since we're sure we won't see any more of them.

[jaimereborn]

Abstract: Predictions are made for the number of chapters told from the point of view of each character

in the next two novels in George R. R. Martins A Song of Ice and Fire series by fitting a random effects

model to a matrix of point-of-view chapters in the earlier novels using Bayesian methods. SPOILER

WARNING: readers who have not read all five existing novels in the series should not read

further, as major plot points will be spoiled, starting with Table 1.

http://arxiv.org/pdf/1409.5830v1.pdf

no offense intended but i think this was a slight waste of time... it really doesn't reveal anything illuminating at all, and is hugely overcomplicated

[Ser Cold Fingers]

no offense intended but i think this was a slight waste of time... it really doesn't reveal anything illuminating at all, and is hugely overcomplicated

If you can't dazzle them with brilliance, baffle them with bullshit! ~ W.C. Fields

[Equilibrium]

Know I know how Oberyn felt, my head just exploded

[Minstral]

So what rocket ship was being built in that link, I don't see how that pertains to the bo....

*Goes back for a closer look*

Kidding aside, I took a brief look at it and then stopped. Trying to quantify... whatever the guy is trying to quantify/predict seems counterproductive to me. GRRM has mentioned that he has tried to cram as much content into some of the books he has written to the point that it is still possible to bind the book together, but in the end had to axe some of the content in ADWD and put it into TWOW. Far more plausible to me that he is at the mercy of his editor's/publisher abaility to produce his books in readable formats then him spending his time forming some strict criteria as to how many chapters a character will still get.

[Cthulhu the Drowned God]

So assuming I read that correctly, Arys Oakheart has roughly a 50% chance of getting a POV in TWOW?

[Red Wedding Cake]

It would be nice if there were more fans willing to dive into mathematical models. It invites criticism, and if there were enough math nerds around we could come up with more parameters like geographic location and how many other characters are involved with the POV. We might come together as a community to decide a metric to determine just how much this character's death foreshadowed in the next book, then apply that percentage as a probability, but there would be a lot of arguing and uncertainty. This is a fine first stab at a model using parameters that are uncontroversial: how many books have appeared (time), how often have we seen this POV (frequency), and when was their first POV (initiation). Mr Vale uses this to make a crude Poisson distribution to predict if the POV character will have no chapters (OMG!DEATH!!) and if they do have POV chapters how many can we reasonably expect based on prior experience? Take the WoW sample chapter characters for instance

Since we know they have at least one chapter already, based on Table2 this model predicts

Arya is likely to have 5 chapters.

Tyrion is likely to have 8 chapters.

Theon will have between 1 and 4 chapters.

Barristan should only have 2 or 3.

Aeron should only have 2 or 3.

Meanwhile, the model says Arianne should only have 1 or 2 chapters, but we know she has at least 2 chapters in TWOW. Why? Clearly this model didn't expect we'd get as many sample chapters from her as we did in the entirety of AFFC+ADWD. It invites discussion for a better prediction.

But I don't think a lot of people on here have that kind of patience. More people will appreciate Walt Hickey's linear extrapolation of when to expect the Winds of Winter release based on time, word count, and average final edit quality words typed a day (with a playful look at livejournal moods and Jets/Giants NFL records).

http://fivethirtyeight.com/features/predicting-when-game-of-thrones-author-george-r-r-martin-will-come-out-with-his-next-book/

TWOW in November 2016 if he stays on Dance with Dragons pace.

[Red Wedding Cake]

So assuming I read that correctly, Arys Oakheart has roughly a 50% chance of getting a POV in TWOW?

Yes. And Melisandre has a 50% chance of not having a POV. The model only goes on how many POV chapters they've had, and they've had them too recently for the model to show what will happen.

[JonCon's Red Beard]

Mathness!!! <O>

[Minstral]

<snip>

Models of prediction would be more reliable if we could reliably predict the work ethic of a human being at all intervals. Perhaps we can think to ourselves that the while GRRM has failed to predict in the past the progress of his work, we can do it for him. Prediction assumes a steady pace of content produced that does not take into account error or downtime. The author has run into both, when he split his fourth book in two as well to the other side projects and commitments he sets for himself (events he attends and other work).

Models that you propose just breed expectations that usually do not follow through, or at least in my opinion. When planning my own workload or seeing proposed outputs in production I don't always see predicable results, I see other factors being entered into the process that were not accounted for originally, or one aspect was underestimated and must be made up.

To use a parallel example, no general has ever written the textbook definition on how to win a battle/campaign. If they did, then they wouldn't need to consider any other factors, but plans rarely survive contact with the enemy.

[Minstral]

<snip>

And I thought there was something on my screen for a second. Congratulations, your avatar managed to set me off balance there

[Rhaenys' kitten]

So assuming I read that correctly, Arys Oakheart has roughly a 50% chance of getting a POV in TWOW?

Arys is definitely an anomaly, I wouldn't expect him to fit any model. Probably his chapter should have been an Arianne chapter, but it didn't work that way for Martin, so he made it an Arys chapter.

Which makes me wonder what was in Arys's chapter that Arianne's POV couldn't accomplish - I may need to re-read that again.