Awesome

NTDS project: Transitional playlists for new musical discoveries

This project is done in the frame of the EPFL course Network tour of data science. The main goal is to take some of the most listened tracks found in the Free Music Archive and help someone wishing to discover a new style or a new artist to transition smoothly from a song he knows and likes to an other unknown music. Basically the idea is to let the user choose two points in a graph representing the music tracks and return him a 10(max)-tracks playlist gradually changing from musics similar to the first track to others more like the the second track.

Code and notebooks

• Load_save_dataset.ipynb loads the metadata from the dataset used in this project. The data with scripts to process it can be found on the fma github repository, it is contained in this archive: fma_metadata.zip. The two files used for this project are tracks.csv and echonest.csv, the relative path to these files can be modified at the beginning of the notebook. This first notebook loads the data with the Pandas library, keeps a subset to create and save an adjacency matrix as well as key features of the selected music tracks.

• Project_ntds_2018_team36.ipynb loads and uses the data processed by the first notebook (.csv and .npy files). Additional generated files can be found in the Project generated files folder. It uses the NetworkX library to create a graph out of the adjacency matrix and compute paths between selected nodes (music tracks) to create a new playlist. It is important to NOT run the whole notebook in one go! The user should click on the two desired nodes in the graph created in section Node selection before running subsequent cells.

Dependencies:

• numpy
• pandas
• matplotlib
• scipy
• math
• collections
• pyunlocbox
• networkx