Skip to content

Latest commit

 

History

History
22 lines (18 loc) · 2.05 KB

README.md

File metadata and controls

22 lines (18 loc) · 2.05 KB
Raw

Classifying music genre with classical machine learning - Special Topics in Digital Humanities 2019/2020 project

This repository contains files submitted to the Special Topics in Digital Humanities 2019/2020 (ST) course at Leiden University.

Author: Kat Kołodziejczyk.

Data

The data used in this project come from the Million Song Dataset. See documentation on the website for a detailed description of the audio features used. Lyrics (in bag-of-words format) come from the musiXmatch Dataset.

Contents

The repository contains the following files:

  • comb_dataset.csv: csv file with audio features and lyrics per song (50000 most frequent lyrics (rows from "i" to "kad"): frequency of word in given song, 0 if not present)
  • Audio feature names: txt file with the name of the 49 audio features used in the classification
  • Classifier: The main Python code
  • STIDH report: Report I wrote for class

The folder "Results" contains the following output from classifier.py:

  • distribution_genre.png: Bar plot showing the genre distribution of comb_dataset.csv
  • all_pred_results.txt: Accuracy scores (in %) for Gaussian Naive Bayes (GNB), Linear Support Vector Classification (Linear SVC) and k-Nearest Neighbours (KNN) for audio features and lyrics
  • audio_best_predictors.txt: Accuracy scores for audio features for GNB, Linear SVC, and KNN for 26, 25, 20, 15 and 10 best predictors
  • audio_results.png: Bar plot showing the accuracy scores for GNB, Linear SVC, and KNN for 26 best audio predictors
  • lyrics_best_predictors.txt: Accuracy scores for lyrics features for 284, 250, 200 and 100 best predictors
  • lyrics_results.png: Bar plot showing the accuracy scores for GNB, Linear SVC, and KNN for 250 best lyrics predictors