Using Math to Predict the Lineup

[ARRRRR]

Ok, so this is something I have been playing around with for the better part of a year. I figured there has to be some way to get a pretty accurate prediction model for the lineups. So, I am going to build a program and spend some time tuning it to hopefully get the most accurate results. 

My original thought process was to utilize API's to get tour info, and then just analyze trends from past years from that. This seemed sound and all, but clearly something was missing. It wasn't until I was a big part of a music curation app that I realized that it was the Sasquatch culture....the reason we can expect who we do, and not Metallica or some shit. 

So I have looked at options, and I have decided to heavily rely on the Echonest. The Echonest will allow me to seed a prediction model with past lineups and get a feel for the type of music. I am going to build the program to analyze trends, such as how often an artist from x years ago revisits, or how far away an artist might be on the weekend before or after Sasquatch. The cool part of this is the fact that I can seed with other festivals as well, and potentially predict other lineups (or use the information to cross reference and determine the likelihood of a Sasquatch appearance). 

I am going to build all of these variables to be reactive and defined by the program, rather than by me. The unfortunate (but necessary) part of the Echonest will be a sort of randomness. This is the key element, but it also means mistrials/outliers are possible. So I will run trials x times before shifting variables y amount.

I might open source the thing and throw it up on GitHub. Let me know if you guys have any suggestions or ideas for moving forward. 

My end goal is to correctly guess 30% or more of the Sasquatch 2016 lineup.

[TBRenegade]

I like where this is going

[emptyfox]

This is like when pea asked snacks to tell us about crossovers last year, except it's nothing like that. I hope this is awesome.

[ARRRRR]

Ok so a bit of research today and I have settled on the following APIs:

1) EchoNest. Hands down this will be the heart of the human aspect. Also, its ambiguous "artist_hotttness" will allow me to gauge how relevant an artist is publicly. (From the echonest website, "Returns our numerical description of how hottt an artist currently is"). In a perfect world, I could have access to artist_hotttness as a graph, and analyze how it is before Sasquatch (ie, is Sasquatch getting bands before they get big? What is the range of artist_hotttness before they play sasquatch? What about ROC leading up to Sasquatch?) but I don't think this is the case. Going to save this data monthly once the program is running so that we get more accurate with time.

2) Eventful. Echonest and Eventful play nicely together because of project rosetta stone. My few tests of Eventful seemed to have accurate and updated tour information. I will be able to use Eventful to gauge where they are touring and when. This could be the most obvious give away to whether or not an artist is going to appear at Sasquatch, so it is important this is accurate.

3) Google APIs, specifically geocoding. I want to feed information from Eventful into this to get a longitude and latitude. Once I have that I can perform simple geometry to get the proximity to Sasquatch. Even though some Eventful events have lon/lat, I don't particularly trust their accuracy and I do not believe all do.

What is currently missing:
News. I need to find a good way to incorporate music news into this thing. If it comes down to it I could modify a twitter parser I built before, and twitter has some pretty decent API's I can use but it's not really what I want to focus my time on. We'll see, more to come on this one. Artist_hotttness might end up being enough to cover this.

[wonk]

You will be wrong. Math helped me predict it.

[Pea]

HO-LY SHIT

[totinoboy]

[StormyPinkness]

Wonk knows about math.

[Horned Gramma]

...and yet he is a stranger to the ways of the heart.