Smarter Playlist

When I want to listen to music from iTunes library, I basically have two options:

• listen to music from only one album, in order or on shuffle; or
• listen to music from my entire library, on shuffle.

Neither of these options are really ideal for me. Many of my albums are quite long (several hours), so I can get bored with listening to the same style of music for too long. And shuffling through my entire library means there's no continuity between sequential songs. Besides, there's a lot of good music in my library that I forget I have, so I'd like shuffle to be biased toward less recently played songs.

To solve all of these problems, I present Smarter Playlist™. This script will create an iTunes playlist that:

• shuffles through your entire iTunes library, but
• tends to prefer songs that you have listened to less recently, and
• has some inertia, and therefore likes to play a few sequential songs from an album before switching to another album.

Table of contents

• Smarter Playlist
  • Table of contents
  • Setup
  • Basic usage
  • Algorithm description
  • Advanced usage
  • Portability
  • Implementation notes
  • Contributing

Setup

• Clone or download this repository.
• Make sure you have a JDK (or at least a JRE) installed.
• Install Leiningen using Homebrew (brew install leiningen) or from the official website.

On OS X, you can use Homebrew to install all the dependencies. You are using OS X, right? (Smarter Playlist uses AppleScript to interface with iTunes and will only work on OS X.)

% brew cask install java
% brew install leiningen

Basic usage

Create a playlist by executing lein run in the smarter-playlist directory:

% lein run
age-summary = (stats/mean-median) [default]
age-weighting = (stats/log-power) [default]
fraction-distribution = (-> (stats/normal :mean 0.5 :stdev 0.3) (stats/bounded :min 0 :max 1)) [default]
length-distribution = (-> (stats/normal :mean 8 :stdev 5) (stats/bounded :min 0 :max 6) (stats/rounded)) [default]
playlist-length = 100 [default]
playlist-name = "Smarter Playlist" [default]

Making an empty playlist in iTunes.
Reading iTunes library.
Creating playlist.
Exporting playlist to iTunes.

This will add a playlist named "Smarter Playlist" to iTunes, overwriting any previously existing playlist by that name.

Specify the name and/or length of the playlist by passing (optional) keyword arguments:

% lein run --playlist-length 10 --playlist-name '"The Best Playlist"'
age-summary = (stats/mean-median) [default]
age-weighting = (stats/log-power) [default]
fraction-distribution = (-> (stats/normal :mean 0.5 :stdev 0.3) (stats/bounded :min 0 :max 1)) [default]
length-distribution = (-> (stats/normal :mean 8 :stdev 5) (stats/bounded :min 0 :max 6) (stats/rounded)) [default]
playlist-length = 10
playlist-name = "The Best Playlist"

Making an empty playlist in iTunes.
Reading iTunes library.
Creating playlist.
Exporting playlist to iTunes.

Please note the double quotes around the playlist name. The keyword arguments are allowed to be arbitrary Clojure forms which will be evaluated. So, to specify a string, you must pass the double quotes as part of the argument.

After you read the algorithm description below, you may want to configure some of the other parameters. For instance, the following inhibits Smarter Playlist's default behavior of sometimes skipping songs in an album rather than taking them sequentially, and increases the average run length considerably:

% lein run --fraction-distribution '(constantly 1)' --length-distribution '(-> (stats/normal :mean 20 :stdev 5) (stats/bounded :min 0) (stats/rounded))'
age-summary = (stats/mean-median) [default]
age-weighting = (stats/log-power) [default]
fraction-distribution = (constantly 1)
length-distribution = (-> (stats/normal :mean 20 :stdev 5) (stats/bounded :min 0) (stats/rounded))
playlist-length = 100 [default]
playlist-name = "Smarter Playlist" [default]

Making an empty playlist in iTunes.
Reading iTunes library.
Creating playlist.
Exporting playlist to iTunes.

Algorithm description

The age of each album is computed using the age-summary function. Then a random album is selected, with the weighting for each album determined based on its age by the age-weighting function.

Once an album is selected, a "run length" is selected using the length-distribution function, and a random song from the album is selected with the weighting determined again using age-weighting. The "run" of songs under consideration begins with the chosen song, extends either forward or backward in the album (with equal probability, and wrapping around if necessary), and has the chosen length (subject to the constraint that the length must be at most the number of songs in the album).

Now a fraction between 0 and 1 is selected using the fraction-distribution function, and the run length is multiplied by the fraction to determine the number of songs from the run that will actually be added to the playlist. These songs are selected randomly, without replacement, with weighting given again by the age-weighting function.

Before the selected songs are added to the playlist, they are sorted by their order in the original album. All sorting is performed according to iTunes' rules: tracks are sorted first by disc number, then by track number, then by title for tracks lacking a track number (with songs lacking track numbers coming after songs with track numbers, within a given album).

The selected songs are then added to the playlist and removed from the in-memory index of the iTunes library. (This ensures that no single playlist will ever contain duplicates, and implies that the maximum allowable playlist-length is the number of songs in your iTunes library.) The playlist is then extended by continuing to select random albums until it reaches playlist-length. If necessary, the playlist is trimmed to have exactly that length.

Advanced usage

The available options and their default values are printed whenever you run Smarter Playlist. They can all be set in the same way, namely

--some-option '(some Clojure data structure or form)'

The options are evaluated after they are parsed. You have access to anything available in the smarter-playlist.core namespace, including the smarter-playlist.stats functions used in the default values of many of the options.

To be precise:

• age-summary should be a function taking a sequence of ages (doubles) and returning a single age (double) summarizing them.

• age-weighting should be a function taking an age (double) and returning a weighting (double) for the album or song with that age in random selection.

• fraction-distribution and length-distribution should be functions of no arguments returning fractions (doubles) and lengths (longs) respectively.

• playlist-length should be a long.

• playlist-name should be a string.

Portability

You can also create a JAR that can be run from anywhere that has Java installed (without a need for Leiningen or the source code):

% lein uberjar
Compiling smarter-playlist.core
Compiling smarter-playlist.stats
Compiling smarter-playlist.util
Created /Users/raxod502/Desktop/Code/Clojure/smart-playlist/target/smarter-playlist-0.1.0-SNAPSHOT.jar
Created /Users/raxod502/Desktop/Code/Clojure/smart-playlist/target/smarter-playlist-standalone.jar
% java -jar target/smarter-playlist-standalone.jar --playlist-length 10
age-summary = (stats/mean-median) [default]
age-weighting = (stats/log-power) [default]
fraction-distribution = (-> (stats/normal :mean 0.5 :stdev 0.3) (stats/bounded :min 0 :max 1)) [default]
length-distribution = (-> (stats/normal :mean 8 :stdev 5) (stats/bounded :min 0 :max 6) (stats/rounded)) [default]
playlist-length = 10
playlist-name = "Smarter Playlist" [default]

Making an empty playlist in iTunes.
Reading iTunes library.
Creating playlist.
Exporting playlist to iTunes.

Implementation notes

• Uses AppleScript inlined in smarter-playlist.core. Because AppleScript is horrible, songs are added to the iTunes playlist one at a time, which is slow. I would love to find a way to get iTunes to do them all at once.

• Uses a custom defconfig macro to define configuration parameters (dynamic vars that can be configured using command-line options). The macro places special metadata on the created vars to identify them as configuration parameters and to allow the literal form passed to def to be reported in the command-line output rather than the less-pretty evaluated form. The advantage of doing this is that adding a new configuration variable is as easy as adding a new defconfig declaration to smarter-playlist.core (the var will be automatically detected by the option parser).

Contributing

Please do, if you want! Issues and pull requests are welcome.