Project Ventoux Overview
Project Parts:
- Part 1 technical overview: single feature linear regression with regularization, normalization and gradient descent on Beatport and Spotify data
- Part 2 technical overview: multiple polynomial regression on Beatport and Spotify electronic music data
- Summit: multilayer neural network with Tensor Flow
Project Ventoux is named after the famous mountain in France that has featured heavily in the Tour de France over the years. It's so famous that, in 2021, they sent the peloton up it twice in a stage run by Belgian cycling powerhouse Wout van Aert. It's a vicious climb that a lot of everyday cyclists dream of climbing.
The core of the data used is from this Kaggle dataset of Spotify and Beatport track data. In addition, the early part of the project accepts at face value how Spotify defines their Audio Features metrics and how Beatports calculates their rankings. Later iterations of the
When I started this project, I had (and still have) a general thesis: what does increased popularity for an artist, label, or genre in the eletronic music space do to their music, and how do they respond to increased popularity? I have a borderline conspiracy theory that melodic techno has been ruined over the last few years by its explosion in popularity. If you want to see what I mean, sample some songs from 'Afterlife's Realm of Consciousness in 2016 and then take a listen to part VI from 2023. Recondite, one of the earliest artists on the label, has posted about this trend in Instagram.
I finished Andrew Ng's famous, popular, and well done Machine Learning specialization on Coursera and my approach is heavily influenced by his classes.
General Links: