(Awkwardly) mono-repo huge project for learning purposes.

Contains code for:

• collecting a number of pseudo-random songs from the Spotify API through spotipy. The dataset is then comprised of both musicality features fetched from the API and audio features extracted through librosa from a 30 sec long audio sample.
• the (pretty much toy) model to classify a song as popular or unpopular based on the above features. The model has been currently published to the TestPyPi index.
• the heroku-deployed app to make predictions through a web-served REST API.

Todos (?):

• Rework the app or add an endpoint to make predictions based on the author/song name (provided it is on Spotify)
• Make it work with audio features only by adding a post method to upload an audio sample (and training a "light" model along with the full one)