Extract MusicNet labels

extract_musicnet_labels.py

#!/usr/bin/env python3
# Extracts just a dataset of labels (ingoring audio) from the full
# MusicNet dataset in order to shrink it from 11 GB.
# Note: I was not able to save objects (not numpy arrays) back to npz file,
# so we save to pickle.

import numpy as np
from tqdm import tqdm

# the npz file was picked in Python 2, we're opening in Python 3
musicnet = np.load('musicnet.npz', encoding = 'latin1', fix_imports=True, mmap_mode='r')
musicnet_labels = {id:musicnet[id][1] for id in tqdm(musicnet.keys())}

with open('musicnet_labels.pickle', 'wb') as f:
    pickle.dump(musicnet_labels, f)