def LoadMetadata(self):
"""Loads metadata from csv files"""
plt.rcParams['figure.figsize'] = (17, 5)
# Load metadata and features.
self.tracks = utils.load("../fma/data/tracks.csv")
self.genres = utils.load("../fma/data/genres.csv")
self.features = utils.load("../fma/data/features.csv")
# self.echonest = utils.load("fma_metadata/echonest.csv")
np.testing.assert_array_equal(self.features.index, self.tracks.index)
def main_orig():
tracks = utils.load('tracks.csv')
features = pd.DataFrame(index=tracks.index,
columns=columns(), dtype=np.float32)
# More than usable CPUs to be CPU bound, not I/O bound. Beware memory.
nb_workers = int(1.5 * len(os.sched_getaffinity(0)))
# Longest is ~11,000 seconds. Limit processes to avoid memory errors.
table = ((5000, 1), (3000, 3), (2000, 5), (1000, 10), (0, nb_workers))
for duration, nb_workers in table:
print('Working with {} processes.'.format(nb_workers))
tids = tracks[tracks['track', 'duration'] >= duration].index
tracks.drop(tids, axis=0, inplace=True)
pool = multiprocessing.Pool(nb_workers)
it = pool.imap_unordered(compute_features, tids)
for i, row in enumerate(tqdm(it, total=len(tids))):
features.loc[row.name] = row
if i % 1000 == 0:
save(features, 10)
save(features, 10)
test(features, 10)
def test(features, ndigits):
indices = features[features.isnull().any(axis=1)].index
if len(indices) > 0:
print('Failed tracks: {}'.format(', '.join(str(i) for i in indices)))
tmp = utils.load('features.csv')
np.testing.assert_allclose(tmp.values, features.values, rtol=10**-ndigits)
def LoadQuery(self, csv, feature="mfcc"):
self.query = utils.load("featureFiles/" + csv)
f = self.query[0] == feature
self.query = self.query.loc[f]
return self.query[3].values
from fma import utils
from utils.load_transform_data import VALID_GENRES, get_tracks_locations,\
get_filename_and_ids
music_files_locations = get_tracks_locations()
tracks = utils.load('fma/tracks.csv')
small = tracks['set', 'subset'] <= 'small'
train = tracks['set', 'split'] == 'training'
val = tracks['set', 'split'] == 'validation'
test = tracks['set', 'split'] == 'test'
selected_genres = tracks['track', 'genre_top'].isin(VALID_GENRES)
x_train_ids = tracks.loc[small & train & selected_genres].index.values
x_val_ids = tracks.loc[small & val & selected_genres].index.values
x_test_ids = tracks.loc[small & test & selected_genres].index.values
train_valid_tracks = get_filename_and_ids(x_train_ids, music_files_locations)
validation_valid_tracks = get_filename_and_ids(x_val_ids,
music_files_locations)
test_valid_tracks = get_filename_and_ids(x_test_ids, music_files_locations)
if args.wavelet:
requested_wavelet = args.wavelet
if requested_split == "training":
requested_split_path = "train"
elif requested_split == "validation":
requested_split_path = "validation"
elif requested_split == "test":
requested_split_path = "test"
# Load the metadata files
tracks = fma_utils.load(input_dir + 'tracks.csv')
features = fma_utils.load(input_dir + 'features.csv')
# Make sure everything in features is in tracks and vice versa
np.testing.assert_array_equal(features.index, tracks.index)
# Use the specified data subset:
subset = tracks['set', 'subset'] <= requested_subset
split = tracks['set', 'split'] == requested_split
rel_track_ids = tracks.loc[subset & split].index
y_values = tracks.loc[subset & split, ('track', 'genre_top')]
unique_genres = y_values.unique().categories
def __load_data(self, path_prefix):
self.tracks = utils.load(path_prefix + "/tracks.csv")
self.genres = utils.load(path_prefix + "/genres.csv")
self.features = utils.load(path_prefix + "/features.csv")