def dtw_kl_divergence(x, y):
""" Kullback-Leibler divergence """
if x.shape[0] > 0 and y.shape[0] > 0:
d = dtw.dtw(x, y, kullback_leibler.kl_divergence, True)
elif x.shape[0] == y.shape[0]:
d = 0
else:
d = np.inf
return d
def kl_divergence(x, y, normalized):
"""Kullback-Leibler divergence"""
if x.shape[0] > 0 and y.shape[0] > 0:
d = dtw.dtw(x, y, kullback_leibler.kl_divergence, normalized)
elif x.shape[0] == y.shape[0]:
d = 0
else:
d = np.inf
return d
def dtw_kl_divergence(x, y):
""" Kullback-Leibler divergence
"""
if x.shape[0] > 0 and y.shape[0] > 0:
d = dtw.dtw(x, y, kullback_leibler.kl_divergence)
elif x.shape[0] == y.shape[0]:
d = 0
else:
d = np.inf
return d
def dtw_kl_divergence(x, y):
""" Kullback-Leibler divergence """
# times on rows, features on columns
if x.shape[0] > 0 and y.shape[0] > 0:
d = dtw.dtw(x, y, kullback_leibler.kl_divergence, True)
elif x.shape[0] == y.shape[0]:
d = 0
else:
d = np.inf
return d
def distance(x, y, normalized):
"""Dynamic time warping cosine distance
The "feature" dimension is along the columns and the "time" dimension
along the lines of arrays x and y
"""
if x.shape[0] > 0 and y.shape[0] > 0:
# x and y are not empty
d = dtw.dtw(x, y, cosine.cosine_distance, normalized)
elif x.shape[0] == y.shape[0]:
# both x and y are empty
d = 0
else:
# x or y is empty
d = np.inf
return d
def default_distance(x, y):
""" Dynamic time warping cosine distance
The "feature" dimension is along the columns and the "time" dimension
along the lines of arrays x and y
"""
if x.shape[0] > 0 and y.shape[0] > 0:
# x and y are not empty
d = dtw.dtw(x, y, cosine.cosine_distance)
elif x.shape[0] == y.shape[0]:
# both x and y are empty
d = 0
else:
# x or y is empty
d = np.inf
return d
def dtw_cosine_distance(x, y):
return dtw.dtw(x, y, cosine.cosine_distance)
def dtw_cosine_distance(x, y, normalized):
return dtw.dtw(x, y, cosine.cosine_distance, normalized)
def dtw_cosine_distance(x, y, normalized):
return dtw.dtw(x, y, cosine.cosine_distance, normalized)
def dtw_cosine_distance(x, y):
return dtw.dtw(x, y, cosine.cosine_distance)
help=('if true, take mean distance along'
'dtw path length instead of sum'))
args = parser.parse_args()
assert path.exists(args.feat_file), ("No such file "
"{}".format(args.feat_file))
assert path.exists(args.task_file), ("No such file "
"{}".format(args.task_file))
dis_file = path.join(args.res_folder, 'distances',
args.res_id + '.distances')
score_file = path.join(args.res_folder, 'scores', args.res_id + '.scores')
result_file = path.join(args.res_folder, 'results', args.res_id + '.txt')
assert not(path.exists(dis_file)), \
"{} already exists".format(dis_file)
assert not(path.exists(score_file)), \
"{} already exists".format(score_file)
assert not(path.exists(result_file)), \
"{} already exists".format(result_file)
assert args.distance in ['kl', 'cos', 'dur'], \
"Distance function {} not supported".format(args.distance)
if args.distance == 'kl':
frame_dis = kl.kl_divergence
elif args.distance == 'cos':
frame_dis = cos.cosine_distance
if args.distance in ['kl', 'cos']:
distance = lambda x, y, normalized: dtw.dtw(
x, y, frame_dis, normalized=normalized)
else:
distance = lambda x, y, normalized: np.abs(x.shape[0] - y.shape[0])
run_ABX(args.feat_file, args.task_file, dis_file, score_file, result_file,
distance, args.normalized)
