Music to My Ears

Sameer Bibikar, Aimun Khan, Nimay Kumar, Jason Nguyen, Shrey Sachdeva

We worked in separate Jupyter Notebooks and serialized the data between runs to offer some persistence. Because scikit-learn models support pickling/unpickling, it was easy for us to generate models, save them to a file, and then test them separately.

Files in this repository

• feature_selection.ipynb - Directly reads the data from /mnt/snap, which is the mountpoint for the MSD, selects only certain features, and dumps to different HDF5 files. Then, it combines those into one for convenience.
• Scratch File Clustering*.ipynb - Initial supervised methods, feature exploration, adding the genre dataset
• filter_users_20.ipynb - Filters out user histories which have less than 20 songs in them and outputs HDF5.
• kdtree.ipynb - Trying out nearest neighbors algorithms (KDTree, Ball tree)
• kdtree_users.ipynb - Implementing nearest neighbors algorithm with user histories, pickling these models
• evaluation.ipynb - Evaluating pickled nearest neighbors models.