It started with a post on Practical Developer about finding your new favorite song with Spotify. “I should try that”, I thought.
The comments mentioned Last.FM, a service which tracks your music for you. Eric mentioned a playlist of 45 tracks that are his favorites, but I don’t have that. I have a number of playlists on Spotify, including some designed to get me going on Monday Mornings and some intended to help keep me in flow state, but I didn’t have a list of songs I go back to again and again.
But, with some poking, I soon had it. In CSV form, because I didn’t immediately write my own. And, after a few queries, I decided re-parsing the CSV each time was too slow, and threw it into MySQL.
My thought, then, was to go all DataVis on it and run it through R. I mostly know R as a plotting library, handling data I pre-munge with DBs and other languages, and I have long wanted to “get good” on actually using it as a language.
I can get to top artists fairly easily, and break it down by year with a little bit more work.
artist y2008 y2009 y2010 y2011 y2012 y2013 y2014 y2015 y2016 y2017 Happy_Mondays 0 1 0 154 581 270 135 96 111 82 Mogwai 0 55 8 124 49 61 380 292 173 63 Orbital 10 143 41 269 391 28 28 148 24 84 Wilco 1 137 71 63 37 16 225 152 58 193 Daft_Punk 2 38 39 204 137 294 73 65 30 0
I would hope that, in essence,
"Orbital" , 10, 143, 41, ... would be enough to draw a line plot, but nope.
With struggle and help remembered from the last time I tried to struggle with R, I got to
years Happy_Mondays Mogwai Orbital Wilco Daft_Punk <chr> <dbl> <dbl> <dbl> <dbl> <dbl> 1 y2008 0 0 10.0 1.00 2.00 2 y2009 1.00 55.0 143 137 38.0 3 y2010 0 8.00 41.0 71.0 39.0 4 y2011 154 124 269 63.0 204 5 y2012 581 49.0 391 37.0 137 6 y2013 270 61.0 28.0 16.0 294 7 y2014 135 380 28.0 225 73.0 8 y2015 96.0 292 148 152 65.0 9 y2016 111 173 24.0 58.0 30.0 10 y2017 82.0 63.0 84.0 193 0
Which is nice enough. MySQL won’t let you have columns that are numeric, so I was thinking that eventually I would be able to use
sapply to pull the initial
y from the
years column, but that should be sufficient to allow me to see my top artists.
But, each time I got to the point of
ggplot(), it didn’t like it. I looped through each half-remembered plot I wrote in the past, trying to remember how I got it to work, and searching through StackOverflow and trying to get a sense of what R wanted of me.
Until I saw an example that used a data frame they threw together right before plotting. I tried it, looking at the generated frame before plotting, and I figured out the problem. This is what it desired:
year,plays,artist 2008,0,"Happy Mondays" 2008,0,Mogwai 2008,10,Orbital 2008,1,Wilco 2008,2,"Daft Punk" 2008,2,"The Jayhawks" 2008,45,R.E.M. 2008,7,"Explosions in the Sky" 2008,0,Rush 2008,0,"Ozric Tentacles" 2009,1,"Happy Mondays" 2009,55,Mogwai ...
It would group the artists without me. I just needed to get to this point, which, I confess, I broke down and implemented it with Perl, because I could, in the fullness of time, find the right query to get my all-time top-ten artists and their results by year in
year,plays,artist format, but it was easier to take my existing query, munge the year, and spit out CSV.
And, from there, it was downright easy to actually plot the thing.
#!/usr/bin/Rscript suppressPackageStartupMessages( require( "Cairo" , quietly=TRUE ) ) suppressPackageStartupMessages( require( "ggplot2" , quietly=TRUE ) ) suppressPackageStartupMessages( require( "readr" , quietly=TRUE ) ) data <- read_csv( 'top_artists.csv' ) CairoPNG( filename = "/home/jacoby/top_artists.png" , height = 600 , width = 800 , pointsize = 12 ) ggplot(data=data, aes(x=year, y=plays, colour=artist)) + geom_line() + ggtitle( "Top Artists, Plotted by Plays / Year" ) + theme( plot.title = element_text( size=24) )
I am pretty sure that I left something repeating Explosions in the Sky overnight in 2012, which is why there’s that spike. Also, I should give Daft Punk some love this year.
While digging through this, it struck me that Programming is like Parenting an Infant. The keys are learning what input it wants, What he output should look like, and what do if either don’t look the way they should.
I suppose the next step is to start with the tracks and plays themselves, building them up a step at a time until I get to that CSV-like level. In fact, I should, if only as a learning experience.
And I think a count of the number of artists per year to whom I listened to one track or less would be interesting.