def _calc_secondary_distance(self):
"""Calculate secondary distances (e.g. Mutual Proximity)"""
sec_dist_fun = SEC_DIST[self.secondary_distance_type]
try:
self.secondary_distance = sec_dist_fun(
D=self.original_distance, metric=self.metric)
except TypeError: # centering has no keyword 'D='
if self.secondary_distance_type in ['cent', 'wcent']:
self.secondary_distance = \
cosine_distance(sec_dist_fun(X=self.vectors))
elif self.secondary_distance_type in ['lcent']:
self.secondary_distance = 1. - sec_dist_fun(X=self.vectors)
elif self.secondary_distance_type in ['dsg', 'dsl']:
self.secondary_distance = sec_dist_fun(X=self.vectors)
else:
raise ValueError("Erroneous secondary distance type: {}".
format(self.secondary_distance_type))
return self
def load_dexter():
"""Load the example data set (dexter).
Returns
-------
D : ndarray
Distance matrix
classes : ndarray
Class label vector
vectors : ndarray
Vector data matrix
"""
n = 300
dim = 20000
# Read class labels
classes_file = os.path.dirname(os.path.realpath(__file__)) +\
'/example_datasets/dexter_train.labels'
classes = np.loadtxt(classes_file)
# Read data
vectors = np.zeros((n, dim))
data_file = os.path.dirname(os.path.realpath(__file__)) + \
'/example_datasets/dexter_train.data'
with open(data_file, mode='r') as fid:
data = fid.readlines()
row = 0
for line in data:
line = line.strip().split() # line now contains pairs of dim:val
for word in line:
col, val = word.split(':')
vectors[row][int(col)-1] = int(val)
row += 1
# Calc distance
D = cosine_distance(vectors)
return D, classes, vectors
def test_cosine_dist_equal_to_scipy_pdist_cos(self):
cos_dist = cosine_distance(self.vectors)
cos_dist_scipy = squareform(pdist(self.vectors, 'cosine'))
result = np.allclose(cos_dist, cos_dist_scipy)
return self.assertTrue(result)