ezorrio / genre-classification

Genre Classification using Locality Sensitive Hashing (LSH)

MMD-Assignment #1: Local Sensitive Hashing (LSH) for Item Search and Genre Classification for FMA dataset.

Team: Emin Guliev, Justus Rass & Christian Wiskott.

Requirements

• Python 3
• Numpy
• Matplotlib
• Jupyter

Cloning

Github repo can be cloned from:

https://github.com/ezorrio/genre-classification.git

If you have data files fetched from original source, clone using following command:

GIT_LFS_SKIP_SMUDGE=1 git clone https://github.com/ezorrio/genre-classification

Description

The 4 Python files should be placed in a directory with the .csv files in a folder called 'metadata'.

• FMA.py
• LSH.py
• main.py
• MusicSearch.py
• metadata/
  • features.csv
  • tracks.csv
• experiments/
• paper/ - contains LaTeX source files for paper

To run it with the hyperparameters used for the final results, simply execute the 'experiments_manual.ipynb' file located in the experiments folder.

The experiments folder contains the ipython notebooks used to carry out the tests for validation. The experiments/results folder contains the raw results for the test suite used for k=3, 5, 7 nearest neighbours, as well as a condensed list of the best achieved results and the manual tests carried out with those parameters.