🎵 Spotify playlist sorter

Sorting playlists into a mathematically ideal order, using graph theory.

What?

I like building playlists, but they're a lot of work.

Avoiding sudden, jarring changes in the vibe/mood takes careful attention which I can't be bothered to give them. I want to throw a bunch of tracks together and let the tracks sort themselves out.

Thankfully, by combining a bit of graph theory with the machine learning inferred features for each track in the Spotify API, this problem can be solved automatically!

How?

Reframing the task as a version of the travelling salesman problem where each 'city' is a point in track-feature space, we end up minimising the overall path length between tracks and creating the smoothest possible transition from one mood to another!

Getting started

You can run the notebooks for this project in a container by running docker compose up --build from the notebooks/ directory.

The container will install all of the necessary dependencies, and environment variables will be pulled in from your local .env file.

The logs will output a localhost URL which you should use to access jupyter, just as you normally would.

Environment variables

You'll need to fill out the .env.blank file and rename it .env to get the project running properly.

• SPOTIPY_CLIENT_ID and SPOTIPY_CLIENT_SECRET are credentials for working with the spotify API. You'll need to go to developer.spotify.com to create an app and copy over its client ID and secret.
• PLAYLIST_ID is the ID of the playlist you'll sort. You can find the ID of your playlist by clicking the 'share' button and getting a shareable link. The playlist ID is the code after /playlist in the URL, eg https://open.spotify.com/playlist/THIS_BIT_HERE?si=NOT_THIS_BIT