Jukebox - A tiny music recommendation app

One of Spotify's best features is the "create similar playlist" function. You can use this on your existing playlists to have Spotify generate playlists with songs from the same artists and artists that are similar to those artists.

I've gotten into collecting records over the past couple of years - but that music isn't in my Spotify library unless I go through and manually save each album to a playlist. I'm lazy and I don't want to go through and save all my albums to Spotify, so I can't use the "create similar playlist" function on my record collection.

I use an app called Discogs to track the records in my collection. Since all my records are catalogued on their app, I had an idea. Could I use python to get a list of all the artists in my Discogs collection and generate a Spotify playlist with songs from similar artists?

Python has pretty great packages for interfacing with both the Discogs and Spotify APIs. For Discogs, I used discogs_client and for Spotify, I used spotipy.

1. Register for API Access

In order to get API access, I had to register on the Discogs and Spotify developer sites. On Discogs, I was able to get an access token for my own account without registering a full application. On Spotify, I had to register my application, and then get an access token via their API, which I did using an application called Postman.

2. The Script

Let's import the libraries we'll need:

import random
import discogs_client as dc
import spotipy as sp

First, we'll need to retrieve a list of all the artists in my record collection. To do that, let's log in to Discogs and retrieve the list of records in my collection:

discogs_agent = dc.Client('JukeboxAppUWPCE/0.1', user_token=DISCOGS_ACCESS_TOKEN)
discogs_user = discogs_agent.identity()
record_collection = discogs_user.collection_folders[1].releases

record_collection will be a list of records, each of which is a dict containing information about that specific album. Let's process that list to get a set of all the artists in my collection. (I'm using a set here to avoid duplicate artist entries.)

artists = {record.data['basic_information']['artists'][0]['name'].split(" (")[0]
           for record in record_collection}

Now, let's log in to Spotify:

spotify_agent = sp.Spotify(auth=SPOTIFY_ACCESS_TOKEN)
spotify_id = spotify_agent.current_user()['id']

Thinking ahead - in order to add tracks to a playlist, I'm going to need each track's Spotify ID. Let's make a set to hold those IDs (again, using a set to avoid duplicates):

track_ids = set()

Now, we're going to have to loop through each artist in my collection and query the Spotify API to find artists in Spotify's library that are similar to that artist. Since there could be an arbitrary number of artists in my record collection, I'm going to take a random sample of 10 artists to use for this script. For each artist, our first step will be to get that artist's Spotify ID to use in future requests.

for artist in random.sample(artists, 10):
    print("Getting artists similar to: ", artist)
    artist_data = spotify_agent.search(q='artist:' + artist, type='artist')
    try:
        artist_id = artist_data['artists']['items'][0]['id']
    except IndexError:
        continue # we couldn't find the artist on Spotify

Let's query Spotify for similar artists:

    similar_artists = spotify_agent.artist_related_artists(artist_id)['artists']
    try:
        random_similar_artists = random.sample(similar_artists, 5)
    except ValueError:
        continue # we didn't get at least 5 similar artists from Spotify

For each similar artist, I want to get some tracks to put into my new playlist. I'm going to query Spotify for a list of the artist's top tracks and add a random track from that list to my track_ids set.

    for similar_artist in random_similar_artists:
        print("Found similar artist: ", similar_artist['name'])
        similar_artist_top_tracks = spotify_agent.artist_top_tracks(similar_artist['id'])['tracks']
        track_ids.add(random.sample(similar_artist_top_tracks, 1)[0]['id'])

So now we've got our list of track IDs - let's add them to a new playlist:

new_playlist = spotify_agent.user_playlist_create(
                user=spotify_id,
                name='Jukebox Recommendations',
                public=False,
                collaborative=False,
                description="Playlist of automatically generated song recommendations."
                )
spotify_agent.user_playlist_add_tracks(user=spotify_id, playlist_id=new_playlist['id'],
                                       tracks=track_ids, position=None)