Spot-On-Playlist-Organizer

About

Everybody who uses Spotify has come across playlists. Playlists are the way to store various tracks in a single list. Lists usually tend to get cluttered over time. So do also Spotify playlists. Ever been on a morning commute, rocking your favorite songs and then suddenly the next song queuing up just blew your tune because it just does not quite fit? Spot-On-Playlist-Organizer solves this problem by decluttering playlists and reorganizing a given playlist into multiple smaller ones. It takes various features of your tracks into account to group them.

How does it work?

The application is quite simple. It follows the following steps which are common for an ETL (Extract-Transform-Load) process:

1. Upon running this application, we will extract the track data (for a playlist you previously specified) directly from Spotify.
2. The track data will be transformed and passed onto a clustering algorithm.
3. The tracks will then be stored into new playlists defined by the previously retrieved clusters.

Note: No real validation for correctness of the clusters can be made. This has various reasons:

• No ground truth can be provided to the system. (We don't think you would actually have ground truth on how the playlists should be organized :-)).
• Good playlists and music is a subjective topic, so it may be a good playlist for one but a bad playlist for another.

How do you make good playlists then? We simply use the silhouette score of the clusters to judge the quality of the clusters.

How to get started?

1. Clone this repository.
2. Run python setup.py install
3. Copy config.yml.tmpl to config.yml and adjust the placeholders.
4. Run app.py using the following command: python app.py
5. Enjoy your freshly created playlists.

Use python app.py --help for help.

Want to use example data? Or any specific dump of data? Specify the timestamp of a past run using (in this case example data): python app.py --past-run 201907201413

Packages required

requests>=2.3.0,
six>=1.10.0,
spotipy>=2.4.4,
spotify-token>=0.1.0,
pandas>=0.23.4,
numpy>=1.15.4,
scikit-learn>=0.20.1

Configuration explained

Below you can see an explanation of the config file. You need to specify everything in the auth and spotify section.

auth:                                           # --- Authentication section
    username: YOUR_USERNAME_GOES_HERE           # Username of Spotify goes here
    password: YOUR_PASSWORD_GOES_HERE           # Password of Spotify goes here
spotify:                                        # --- Spotify section
    playlist: PLAYLIST_TO_ANALYZE_GOES_HERE     # Define the playlist to work with here
    username: YOUR_USERNAME_GOES_HERE           # Username to which playlist belongs, should be your own username
    output:                                     # --- Spotify output section
        prefix: PREFIX_GOES_HERE                # Define a prefix which will be used to tag the generated playlists
app:                                            # --- App section
    logging:                                    # --- App logging section
        level: DEBUG                            # Specify logging level. You see everything happening in the application using DEBUG.
        file: true                              # Specify whether logging should be stored to file. If set to false logging is effectively disabled.
        logpath: log                            # Specify where to store the logs. Default is the 'log' folder.
    data:                                       # --- App data section
        path: data                              # Path where to read and write data (in our case we use pickles)
    transformation:                             # --- App transformation section
        use_standard_scaler: false              # Use a standard scaler or use the manually defined transformation rules. Recommended: false
    clustering:                                 # --- App clustering section
        sophisticated: true                     # Use a sophisticated clustering algorithm. Recommended: true (true = AffinityPropagation, false = kMeans)
        unsophisticated_cluster_count: 2        # Define how many clusters should be used when using kMeans. Recommended: 2

Structure

- data                                          # Default data folder
- log                                           # Default log folder
- spoton                                        # Application code folder
| - extract.py                                  # Script that holds the extraction logic
| - transform.py                                # Script that holds the transformation rules and clustering
| - load.py                                     # Script that loads the results back to Spotify
| - utils.py                                    # Utils class that holds function to authenticate with Spotify and provides an interface to the config.
- app.py                                        # Application script - run this to run the code.
- config.yml.tmpl                               # The template for a new config file. Create your own config file called "config.yml".
- data_exploration.[ipynb,html]                 # Iphython notebook which holds analysis of the problem.
- README.md                                     # This readme
- setup.py                                      # Setup script

Discussion

My extended insight can be found on Medium: https://medium.com/@nsiicm0/spot-on-a-way-to-organize-spotify-playlists-c792ee8ade74