def test_suggested_sigma(sess):
data = np.arange(25, dtype=float).reshape((5, 5))
data = tf.convert_to_tensor(data, dtype=tf.float32)
s_sigma = sess.run(mvsc.suggested_sigma(data))
real_s_sigma = 7.0
array_eq(s_sigma, real_s_sigma, decimal=4)
def test_mvmds_multiple(sess, data, is_distance, k, real_result):
data_tf = tf.convert_to_tensor(data, dtype=tf.float32)
result = sess.run(mvmds.mvmds(data_tf, is_distance, k=2))
from multiview.mvmds import mvmds as mds_cpu
print("Multiview Result")
np.set_printoptions(precision=10, suppress=True)
print(mds_cpu(data, is_distance, k=k))
array_eq(np.abs(result[:, 0]), np.abs(real_result[:, 0]), decimal=4)
def test_gaussian_similarity(sess):
data = np.arange(25, dtype=float).reshape((5, 5))
data = tf.convert_to_tensor(data, dtype=tf.float32)
similarity = sess.run(mvsc.distance_gaussian_similarity(data, 2))
real_similarity = np.array([[1., 0.8824, 0.6065, 0.3246, 0.1353],
[0.0439, 0.0110, 0.0021, 0.0003, 0.],
[0., 0., 0., 0., 0.], [0., 0., 0., 0., 0.],
[0., 0., 0., 0., 0.]])
array_eq(similarity, real_similarity, decimal=4)
def test_laplacian_ng(sess):
data = np.arange(25, dtype=float).reshape((5, 5))
data = tf.convert_to_tensor(data, dtype=tf.float32)
laplacian = sess.run(mvsc.laplacian_ng(data))
real_laplacian = np.array([[0., 0.0534, 0.0816, 0.1028, 0.1206],
[0.2672, 0.1714, 0.1527, 0.1466, 0.1450],
[0.4082, 0.2400, 0.2, 0.1820, 0.1723],
[0.5144, 0.2933, 0.2380, 0.2117, 0.1964],
[0.6030, 0.3384, 0.2708, 0.2378, 0.2181]])
array_eq(laplacian, real_laplacian, decimal=4)
def test_preprocess_mds(sess):
data = np.arange(25, dtype=float).reshape((5, 5))
data = tf.convert_to_tensor(data, dtype=tf.float32)
preprocessed_data = sess.run(mvmds.preprocess_mvmds(data))
sim = np.array([[40., 20., 0., -20., -40.],
[20., 10., 0., -10., -20.],
[0., 0., 0., 0., 0.],
[-20., -10., 0., 10., 20.],
[-40., -20., 0., 20., 40.]])
array_eq(preprocessed_data, sim, decimal=4)
def test_mvsc_multiple(sess, data, is_distance, k, real_clust, real_evalues,
real_evectors, real_sigmas):
data_tf = tf.convert_to_tensor(data, dtype=tf.float64)
result = sess.run(mvsc.mvsc(data_tf, is_distance, k=k))
(clust, evalues, evectors, sigmas) = result
# from multiview.mvsc import mvsc as sc_cpu
# print("Multiview Result")
# np.set_printoptions(precision=10, suppress=True)
# print(sc_cpu(data, is_distance, k=k))
array_eq(normalize_labels(clust), normalize_labels(real_clust), decimal=4)
array_eq(evalues, real_evalues, decimal=4)
array_eq(np.abs(evectors), np.abs(real_evectors), decimal=4)
array_eq(sigmas, real_sigmas, decimal=4)
def validateEval(self, a, b, Z, L, J, Y, dx, dy, dz):
array_eq(a + Z.dot(dx) + L.dot(abs(dz)), dz)
array_eq(b + J.dot(dx) + Y.dot(abs(dz)), dy)