def setUp(self):
self.test_matrix = np.loadtxt(get_data_path('full_sym_matrix.txt'))
self.nystrom = Nystrom()
self.big_seed_matrix = self.test_matrix[:49]
self.small_seed_matrix = self.test_matrix[:25]
def setUp(self):
self.test_matrix = np.loadtxt(get_data_path('full_sym_matrix.txt'))
self.nystrom = Nystrom()
self.big_seed_matrix = self.test_matrix[:49]
self.small_seed_matrix = self.test_matrix[:25]
class TestNystrom(unittest.TestCase):
"""
Tests functions to do nystrom fast metric scaling
"""
def setUp(self):
self.test_matrix = np.loadtxt(get_data_path('full_sym_matrix.txt'))
self.nystrom = Nystrom()
self.big_seed_matrix = self.test_matrix[:49]
self.small_seed_matrix = self.test_matrix[:25]
def test_nystrom(self):
self.nystrom.run(self.test_matrix)
def test_nystrom_output_dimensions(self):
""" Ensure that an incorrect num_dimensions_out parameter triggers
an exception """
with self.assertRaises(ValueError):
self.nystrom.run(self.test_matrix,
num_dimensions_out=self.test_matrix.shape[0])
def test_nystrom_build_seed_matrix(self):
"""build_seed_matrix() should return a seedmatrix and an order
"""
seedmat_dim = 10
(seedmat, order) = self.nystrom._build_seed_matrix(
self.test_matrix.shape[0], seedmat_dim,
lambda x, y: (self.test_matrix[x, y]))
self.assertTrue(len(order) == self.test_matrix.shape[0])
self.assertTrue(sorted(order) == range(self.test_matrix.shape[0]))
self.assertTrue(seedmat.shape == (
seedmat_dim, self.test_matrix.shape[0]))
# build_seed_matrix randomises order
ind = argsort(order)
i = random.randint(0, seedmat.shape[0])
j = random.randint(0, seedmat.shape[1])
self.assertAlmostEqual(seedmat[i, j], self.test_matrix[ind[i], ind[j]],
places=5)
def test_calc_matrix_a(self):
"""calc_matrix_a should calculate a k x k matrix of
(predefined) association matrix K of certain (predefined)
value"""
nseeds = self.small_seed_matrix.shape[0]
matrix_e = self.small_seed_matrix[:, 0:nseeds]
matrix_a = self.nystrom._calc_matrix_a(matrix_e)
self.assertAlmostEqual(matrix_a[0, 0], 250.032270, places=5)
self.assertAlmostEqual(matrix_a[-1][-1], 316.875461, places=5)
def test_calc_matrix_b(self):
"""calc_matrix_b should calculate a k x n-k matrix of
association matrix K
"""
nseeds = self.small_seed_matrix.shape[0]
matrix_e = self.small_seed_matrix[:, 0:nseeds]
matrix_f = self.small_seed_matrix[:, nseeds:]
matrix_b = self.nystrom._calc_matrix_b(matrix_e, matrix_f)
self.assertTrue(matrix_b.shape == matrix_f.shape)
self.assertAlmostEqual(matrix_b[0, 0], -272.711227,
places=5)
self.assertAlmostEqual(matrix_b[-1, -1], -64.898372,
places=5)
def test_nystrom_internal(self):
"""nystrom() should return an MDS approximation"""
dim = 3
mds_coords = self.nystrom._nystrom(self.big_seed_matrix, dim)
self.assertTrue(len(mds_coords.shape) == 2)
self.assertTrue(mds_coords.shape[0] == self.big_seed_matrix.shape[1])
self.assertTrue(mds_coords.shape[1] == dim)
self.assertAlmostEqual(mds_coords[0, 0], -10.709626, places=5)
self.assertAlmostEqual(mds_coords[-1, -1], -1.778160, places=5)
def test_nystrom_seed_number(self):
"""nystrom() should give better MDS approximations the more
seeds were used"""
dim = 3
mds_coords = self.nystrom._nystrom(self.big_seed_matrix, dim)
stress = kruskal_stress.KruskalStress(self.test_matrix, mds_coords)
kruskal_stress_big_mat = stress.calc_kruskal_stress()
if PRINT_STRESS:
print("INFO: Kruskal stress for Nystrom MDS "
"(big_seed_matrix, dim=%d) = %f" %
(dim, kruskal_stress_big_mat))
self.assertTrue(kruskal_stress_big_mat < 0.04)
mds_coords = self.nystrom._nystrom(self.small_seed_matrix, dim)
stress = kruskal_stress.KruskalStress(self.test_matrix, mds_coords)
kruskal_stress_small_mat = stress.calc_kruskal_stress()
if PRINT_STRESS:
print("INFO: Kruskal stress for Nystrom MDS "
"(small_seed_matrix, dim=%d) = %f" %
(dim, kruskal_stress_small_mat))
self.assertTrue(kruskal_stress_small_mat < 0.06)
self.assertTrue(kruskal_stress_small_mat > kruskal_stress_big_mat)
class TestNystrom(unittest.TestCase):
"""
Tests functions to do nystrom fast metric scaling
"""
def setUp(self):
self.test_matrix = np.loadtxt(get_data_path('full_sym_matrix.txt'))
self.nystrom = Nystrom()
self.big_seed_matrix = self.test_matrix[:49]
self.small_seed_matrix = self.test_matrix[:25]
def test_nystrom(self):
self.nystrom.run(self.test_matrix)
def test_nystrom_output_dimensions(self):
""" Ensure that an incorrect num_dimensions_out parameter triggers
an exception """
with self.assertRaises(ValueError):
self.nystrom.run(self.test_matrix,
num_dimensions_out=self.test_matrix.shape[0])
def test_nystrom_build_seed_matrix(self):
"""build_seed_matrix() should return a seedmatrix and an order
"""
seedmat_dim = 10
(seedmat, order) = self.nystrom._build_seed_matrix(
self.test_matrix.shape[0], seedmat_dim,
lambda x, y: (self.test_matrix[x, y]))
self.assertTrue(len(order) == self.test_matrix.shape[0])
self.assertTrue(
sorted(order) == list(range(self.test_matrix.shape[0])))
self.assertTrue(seedmat.shape == (
seedmat_dim, self.test_matrix.shape[0]))
# build_seed_matrix randomises order
ind = argsort(order)
i = random.randint(0, seedmat.shape[0])
j = random.randint(0, seedmat.shape[1])
self.assertAlmostEqual(seedmat[i, j], self.test_matrix[ind[i], ind[j]],
places=5)
def test_calc_matrix_a(self):
"""calc_matrix_a should calculate a k x k matrix of
(predefined) association matrix K of certain (predefined)
value"""
nseeds = self.small_seed_matrix.shape[0]
matrix_e = self.small_seed_matrix[:, 0:nseeds]
matrix_a = self.nystrom._calc_matrix_a(matrix_e)
self.assertAlmostEqual(matrix_a[0, 0], 250.032270, places=5)
self.assertAlmostEqual(matrix_a[-1][-1], 316.875461, places=5)
def test_calc_matrix_b(self):
"""calc_matrix_b should calculate a k x n-k matrix of
association matrix K
"""
nseeds = self.small_seed_matrix.shape[0]
matrix_e = self.small_seed_matrix[:, 0:nseeds]
matrix_f = self.small_seed_matrix[:, nseeds:]
matrix_b = self.nystrom._calc_matrix_b(matrix_e, matrix_f)
self.assertTrue(matrix_b.shape == matrix_f.shape)
self.assertAlmostEqual(matrix_b[0, 0], -272.711227,
places=5)
self.assertAlmostEqual(matrix_b[-1, -1], -64.898372,
places=5)
def test_nystrom_internal(self):
"""nystrom() should return an MDS approximation"""
dim = 3
mds_coords = self.nystrom._nystrom(self.big_seed_matrix, dim)
self.assertTrue(len(mds_coords.shape) == 2)
self.assertTrue(mds_coords.shape[0] == self.big_seed_matrix.shape[1])
self.assertTrue(mds_coords.shape[1] == dim)
self.assertAlmostEqual(mds_coords[0, 0], -10.709626, places=5)
self.assertAlmostEqual(mds_coords[-1, -1], -1.778160, places=5)
def test_nystrom_seed_number(self):
"""nystrom() should give better MDS approximations the more
seeds were used"""
dim = 3
mds_coords = self.nystrom._nystrom(self.big_seed_matrix, dim)
stress = kruskal_stress.KruskalStress(self.test_matrix, mds_coords)
kruskal_stress_big_mat = stress.calc_kruskal_stress()
if PRINT_STRESS:
print(("INFO: Kruskal stress for Nystrom MDS "
"(big_seed_matrix, dim=%d) = %f" %
(dim, kruskal_stress_big_mat)))
self.assertTrue(kruskal_stress_big_mat < 0.04)
mds_coords = self.nystrom._nystrom(self.small_seed_matrix, dim)
stress = kruskal_stress.KruskalStress(self.test_matrix, mds_coords)
kruskal_stress_small_mat = stress.calc_kruskal_stress()
if PRINT_STRESS:
print(("INFO: Kruskal stress for Nystrom MDS "
"(small_seed_matrix, dim=%d) = %f" %
(dim, kruskal_stress_small_mat)))
self.assertTrue(kruskal_stress_small_mat < 0.06)
self.assertTrue(kruskal_stress_small_mat > kruskal_stress_big_mat)
