def init_with_others(n_genres=163, subset='small', pca_on=True, n_components=3, reuse=False):
global TRACKS
global X_DATA
global Y_DATA
global X_TEST
global Y_TEST
global X_VAL
global Y_VAL
pca = decomposition.PCA(n_components=n_components)
# these are private helper functions
def do_pca(matrix):
nonlocal pca
return pca.fit_transform(preprocessing.normalize(np.transpose(matrix)))
def add_features(x, features):
nonlocal pca_on
if (pca_on):
if (len(x) == 0):
x = np.array([np.hstack(do_pca(to_matrix(features)))])
else:
x = np.vstack([x, np.hstack(do_pca(to_matrix(features)))])
else:
if (len(x) == 0):
x = np.array([np.hstack(to_matrix(features))])
else:
x = np.vstack([x, np.hstack(to_matrix(features))])
return x
genres = load_data.get_n_genres(n_genres-1)
def add_genres(y, y_genres):
nonlocal n_genres
nonlocal genres
labels = [0 for x in range(n_genres)]
if len(y_genres) == 0:
labels[-1] = 1.
for genre in y_genres:
if genre in genres: # the genre is the top n genres
labels[genres.index(genre)] = 1.
if len(y) == 0:
y = labels
else:
y = np.vstack([y, labels])
return y
if not reuse or len(TRACKS) == 0:
print('Loading tracks...')
TRACKS = load_data.get_tracks(subset)
x = []
y = []
x_t = []
y_t = []
x_v = []
y_v = []
# count of how many tracks are added not within specified genres
diff_d = 0
diff_t = 0
diff_v = 0
for track in TRACKS:
if track.split == 'training':
#print(track.genres)
#print(diff_d)
if set(track.genres).intersection(set(genres)) != set():
x = add_features(x, track.features)
y = add_genres(y, track.genres)
diff_d += 1
elif diff_d/(n_genres-1) > 10:
x = add_features(x, track.features)
y = add_genres(y, [])
diff_d -= 1
elif track.split == 'test':
if set(track.genres).intersection(set(genres)) != set():
x_t = add_features(x_t, track.features)
y_t = add_genres(y_t, track.genres)
diff_t += 1
elif diff_t/(n_genres-1) > 10:
x_t = add_features(x_t, track.features)
y_t = add_genres(y_t, [])
diff_t -= 1
else: # validation
if set(track.genres).intersection(set(genres)) != set():
x_v = add_features(x_v, track.features)
y_v = add_genres(y_v, track.genres)
diff_v += 1
elif diff_v/(n_genres-1) > 10:
x_v = add_features(x_v, track.features)
y_v = add_genres(y_v, [])
diff_v -= 1
Y_DATA = y
X_DATA = x
Y_TEST = y_t
X_TEST = x_t
Y_VAL = y_v
X_VAL = x_v
def init(n_genres=163, subset='small', pca_on=True, n_components=3, reuse=False):
global TRACKS
global X_DATA
global Y_DATA
global X_TEST
global Y_TEST
global X_VAL
global Y_VAL
print("Loading tracks... ")
pca = decomposition.PCA(n_components=n_components)
# these are private helper functions
def do_pca(matrix):
nonlocal pca
return pca.fit_transform(preprocessing.normalize(np.transpose(matrix)))
def add_features(x, features):
nonlocal pca_on
if (pca_on):
if (len(x) == 0):
x = np.array([np.hstack(do_pca(to_matrix(features)))])
else:
x = np.vstack([x, np.hstack(do_pca(to_matrix(features)))])
else:
if (len(x) == 0):
x = np.array([np.hstack(to_matrix(features))])
else:
x = np.vstack([x, np.hstack(to_matrix(features))])
return x
genres = load_data.get_n_genres(n_genres)
def add_genres(y, y_genres):
nonlocal n_genres
nonlocal genres
labels = [0 for x in range(n_genres)]
for genre in y_genres:
if genre in genres:
labels[genres.index(genre)] = 1.
if len(y) == 0:
y = labels
else:
y = np.vstack([y, labels])
return y
if not reuse or len(TRACKS) == 0:
TRACKS = load_data.get_tracks(subset)
x = []
y = []
x_t = []
y_t = []
x_v = []
y_v = []
for track in TRACKS:
if track.split == 'training':
#print(track.genres)
#print(diff_d)
if set(track.genres).intersection(set(genres)) != set():
x = add_features(x, track.features)
y = add_genres(y, track.genres)
elif track.split == 'test':
if set(track.genres).intersection(set(genres)) != set():
x_t = add_features(x_t, track.features)
y_t = add_genres(y_t, track.genres)
else: # validation
if set(track.genres).intersection(set(genres)) != set():
x_v = add_features(x_v, track.features)
y_v = add_genres(y_v, track.genres)
Y_DATA = y
X_DATA = x
Y_TEST = y_t
X_TEST = x_t
Y_VAL = y_v
X_VAL = x_v
def init_mel(n_genres=163, subset='medium', reuse=True, pca_on=False):
global TRACKS
global X_DATA
global Y_DATA
global X_TEST
global Y_TEST
global X_VAL
global Y_VAL
genres = load_data.get_n_genres(n_genres)
genres_count = [0 for x in range(n_genres)]
pca = decomposition.PCA(n_components=57)
if not reuse or len(TRACKS) == 0 or len(TRACKS[0].mel) == 0:
print("Loading tracks...")
TRACKS = load_data.get_tracks(subset, pca=False, mel=True)
print("Tracks loaded")
#genres = load_data.get_n_genres(n_genres-1)
def do_pca(matrix):
nonlocal pca
return pca.fit_transform(preprocessing.normalize(np.transpose(matrix)))
def add_features(x, features):
nonlocal pca_on
if (len(x) == 0):
if pca_on:
features = np.reshape(features, [128,128])
features = np.reshape(do_pca(features), [128,57,1])
x = np.array([features])
else:
if pca_on:
features = np.reshape(features, [128,128])
features = np.reshape(do_pca(features), [128,57,1])
x = np.vstack([x, [features]])
return x
def add_genres(y, y_genres):
nonlocal n_genres
nonlocal genres
nonlocal genres_count
labels = [0 for x in range(n_genres)]
for genre in y_genres:
if genre in genres: # the genre is the top n genres
labels[genres.index(genre)] = 1.
genres_count[genres.index(genre)] += 1
if len(y) == 0:
y = labels
else:
y = np.vstack([y, labels])
return y
x = []
y = []
x_t = []
y_t = []
x_v = []
y_v = []
progress = 1
for track in TRACKS:
print("Processing track {} of {}".format(progress, len(TRACKS)), sep=' ', end='\r', flush=True)
progress += 1
if len(track.mel) != 0:
if track.split == 'training':
if set(track.genres).intersection(set(genres)) != set():
x = add_features(x, track.mel)
y = add_genres(y, track.genres)
elif track.split == 'test':
if set(track.genres).intersection(set(genres)) != set():
x_t = add_features(x_t, track.mel)
y_t = add_genres(y_t, track.genres)
elif track.split == 'validation':
if set(track.genres).intersection(set(genres)) != set():
x_v = add_features(x_v, track.mel)
y_v = add_genres(y_v, track.genres)
print('')
print('X shape: {}'.format(x.shape))
print('Y shape: {}'.format(y.shape))
print('Count of each genre:\n\t{}'.format(genres_count))
Y_DATA = y
X_DATA = x
Y_TEST = y_t
X_TEST = x_t
Y_VAL = y_v
X_VAL = x_v
def init_mel_with_others(n_genres=163, subset='small', reuse=True):
global TRACKS
global X_DATA
global Y_DATA
global X_TEST
global Y_TEST
global X_VAL
global Y_VAL
genres = load_data.get_n_genres(n_genres-1)
if not reuse or len(TRACKS) == 0 or len(TRACKS[0].mel) == 0:
print("Loading tracks...")
TRACKS = load_data.get_tracks(subset, pca=False, mel=True)
print("Tracks loaded")
genres = load_data.get_n_genres(n_genres-1)
def add_features(x, features):
if (len(x) == 0):
x = np.array([features])
else:
x = np.vstack([x, [features]])
return x
def add_genres(y, y_genres):
nonlocal n_genres
nonlocal genres
labels = [0 for x in range(n_genres)]
if len(y_genres) == 0:
labels[-1] = 1.
for genre in y_genres:
if genre in genres: # the genre is the top n genres
labels[genres.index(genre)] = 1.
if len(y) == 0:
y = labels
else:
y = np.vstack([y, labels])
return y
x = []
y = []
x_t = []
y_t = []
x_v = []
y_v = []
# count of how many tracks are added not within specified genres
diff_d = 0
diff_t = 0
diff_v = 0
progress = 1
for track in TRACKS:
print("Processing track {} of {}".format(progress, len(TRACKS)), sep=' ', end='\r', flush=True)
progress += 1
if len(track.mel) != 0:
if track.split == 'training':
if set(track.genres).intersection(set(genres)) != set():
x = add_features(x, track.mel)
y = add_genres(y, track.genres)
diff_d += 1
elif diff_d/(n_genres-1) > 10:
x = add_features(x, track.mel)
y = add_genres(y, [])
diff_d -= 1
elif track.split == 'test':
if set(track.genres).intersection(set(genres)) != set():
x_t = add_features(x_t, track.mel)
y_t = add_genres(y_t, track.genres)
diff_t += 1
elif diff_t/(n_genres-1) > 10:
x_t = add_features(x_t, track.mel)
y_t = add_genres(y_t, [])
diff_t -= 1
elif track.split == 'validation':
if set(track.genres).intersection(set(genres)) != set():
x_v = add_features(x_v, track.mel)
y_v = add_genres(y_v, track.genres)
diff_v += 1
elif diff_v/(n_genres-1) > 10:
x_v = add_features(x_v, track.mel)
y_v = add_genres(y_v, [])
diff_v -= 1
print('')
print('X shape: {}'.format(x.shape))
print('Y shape: {}'.format(y.shape))
Y_DATA = y
X_DATA = x
Y_TEST = y_t
X_TEST = x_t
Y_VAL = y_v
X_VAL = x_v
return np.mean(avg, axis=0)
if __name__ == '__main__':
if len(sys.argv) > 1 and sys.argv[1] == 'debug':
pass
else:
if len(TRACKS) == 0:
TRACKS = load_data.get_tracks('small')
import ply_export
n_genres = 2
genres = load_data.get_n_genres(n_genres)
pca = decomposition.PCA(3)
# these are private helper functions
def do_pca(matrix, pca):
return pca.fit_transform(preprocessing.normalize(np.transpose(matrix)))
# Calculate average .ply for each of n_genres
genre_map = load_data.load_genres()[1]
max_examples = 3
mixed_examples = []