unsupervised clustering spotify music

This github is part of the data science bootcamp at WBS. The project goal was to cluster 5K+ songs unsupervised using song features.

1. setting up repository locally

setting up your environment variables please start with making a copy of the .env-dist file and rename it to .env. Inside the created .env file please add your required information.

1.1 install dependencies

It's recommended to create a new virtualenv or conda environment and install the required packages inside that environment.

To install the required packages please run:

pip install -r requirements.txt

1.1 Getting Started with Spotify

First, navigate to Spotify's Developer Website, register, and click the Create app button. Enter an app name, description, and Redirect URI.

Add the redirect URI to your .env file

1.2 Getting the Client ID and Client Secret

From the dashboard of your new app, navigate to Settings and copy the Client ID and Client Secret.

Add the client id and secret to your .env file

1.3 Getting the Username

To get your username go to (https://open.spotify.com/), not the developer site, and click the profile icon in the upper right to get our Username.

Add the user name to your .env file

2. clustering notebooks

The repository contains four notebooks. Three, bisect_kmeans_..., hdbscan_... and kmeans_... are notebooks that were used to analyze the clustering performance using different scalers and with or without data normalization.

In the forth notebook, clustering_spotify_songs I use the bisect_kmeans_ with "best performing" settings, cluster the 5k+ songs into 25 clusters and creating Spotify playlists with 5 sample songs to listen how well the clustering was done.