deeplearningunb/pop-music

Visualising Popular Songs using Self Organising Maps

Visualising the one million popular songs extracted from the Million Song Dataset (the MillionSongSubset) [1]. The Self Organising Map is built using the minisom library [2].

How to use/build

1. Get the SQLite database with the songs' metadata

2. Export it to CSV filtering the invalid songs

sqlite3 -header -csv track_metadata.db < filter_data.sql > track_metadata.csv

The attribute artist_mbid is dropped from the dataset because it is only an external identifier for the artist in the musicbrainz.org database.

The attribute track_7digitalid is dropped from the dataset because it is only an external identifier for the artist in the external 7digital database.

Songs without a year, shs_work or shs_perf information are discarded.

10000 songs should be exported to the CSV due to memory constraints

3. Run the script and log its result

python -u analyse.py 2>&1 | tee "$(date --iso-8601='minutes').log"

[1] BERTIN-MAHIEUX, Thierry. Million Song Dataset, official website. Available at: http://millionsongdataset.com/. Accessed on 05 May 2022.

[2] VETTIGLI, Giuseppe. MiniSom: minimalistic and NumPy-based implementation of the Self Organizing Map. Available at: https://github.com/JustGlowing/minisom. Accessed on 05 May 2022.