def two_points_test_aux(self,data): t=2 epsilon=0.5 diff = diffusion_map(data, epsilon, t) # no real reason for this epsilon diff_truth = 0.4101401*ones((2,1), dtype=np.float32) # computed in scilab diff_truth[0] *= -1 np.testing.assert_almost_equal(diff_truth, diff, 3)
def test_single_3D_point(self): """ one point is always reduced to zero dimensions since we always throw away the first eigenvector """ data = zeros((1,3), dtype=np.float32) diff = diffusion_map(data,d=1) self.assertEqual(diff.size, 0)
def test_single_3D_point(self):
""" one point is always reduced to zero dimensions since we always throw
away the first eigenvector
"""
data = zeros((1, 3), dtype=np.float32)
diff = diffusion_map(data, d=1)
self.assertEqual(diff.size, 0)
def two_points_test_aux(self, data):
t = 2
epsilon = 0.5
diff = diffusion_map(data, epsilon,
t) # no real reason for this epsilon
diff_truth = 0.4101401 * ones(
(2, 1), dtype=np.float32) # computed in scilab
diff_truth[0] *= -1
np.testing.assert_almost_equal(diff_truth, diff, 3)
def test_dimension_augmentation(self): """"diff can actually increase dimension!""" data = zeros((2,1), dtype=np.float32) diff = diffusion_map(data, d=1) self.assertEqual(diff.size, 2) # output will be zeros since P will be constant 0.5 # which has a non-empty null space # moreover, even though we specified d=2, it forces 1D since its all we # have after discarding the higest eigenvalue dimension self.assertEqual(diff[0,0], [0]) self.assertEqual(diff[1,0], [0]) # picking the right #dims does same diff = diffusion_map(data, d=1) self.assertEqual(diff.size, 2) self.assertEqual(diff[0,0], [0]) self.assertEqual(diff[1,0], [0]) # picking #dims zero gives empty diff = diffusion_map(data, d=0) self.assertEqual(diff.size, 0)
def test_dimension_augmentation(self):
""""diff can actually increase dimension!"""
data = zeros((2, 1), dtype=np.float32)
diff = diffusion_map(data, d=1)
self.assertEqual(diff.size, 2)
# output will be zeros since P will be constant 0.5
# which has a non-empty null space
# moreover, even though we specified d=2, it forces 1D since its all we
# have after discarding the higest eigenvalue dimension
self.assertEqual(diff[0, 0], [0])
self.assertEqual(diff[1, 0], [0])
# picking the right #dims does same
diff = diffusion_map(data, d=1)
self.assertEqual(diff.size, 2)
self.assertEqual(diff[0, 0], [0])
self.assertEqual(diff[1, 0], [0])
# picking #dims zero gives empty
diff = diffusion_map(data, d=0)
self.assertEqual(diff.size, 0)
def test_repeated_points_still_map_together(self):
# for i in range(10000):
data = 10*rand(3,5)
# diff = diffusion_map(data,d=2,t=1)
data = vstack((data, data[1]))
data = vstack((data, data[1]))
diff_repeat = diffusion_map(data,d=2,t=1)
# result gets altered by repeating point
# np.testing.assert_not_almost_equal(diff[0],diff_repeat[0],10)
# but at least repeated point still maps to same so. In theory only one
# dimension could have been better used though
#
# Also note how the precision is very bad in absolute numbers!
# If you use equality up to 2 decimal places, this test sometimes fails.
np.testing.assert_almost_equal(diff_repeat[1],diff_repeat[3],1)
np.testing.assert_almost_equal(diff_repeat[1],diff_repeat[4],1)
def test_repeated_points_still_map_together(self):
# for i in range(10000):
data = 10 * rand(3, 5)
# diff = diffusion_map(data,d=2,t=1)
data = vstack((data, data[1]))
data = vstack((data, data[1]))
diff_repeat = diffusion_map(data, d=2, t=1)
# result gets altered by repeating point
# np.testing.assert_not_almost_equal(diff[0],diff_repeat[0],10)
# but at least repeated point still maps to same so. In theory only one
# dimension could have been better used though
#
# Also note how the precision is very bad in absolute numbers!
# If you use equality up to 2 decimal places, this test sometimes fails.
np.testing.assert_almost_equal(diff_repeat[1], diff_repeat[3], 1)
np.testing.assert_almost_equal(diff_repeat[1], diff_repeat[4], 1)
def test_sanity1(self): data = zeros((1,1), dtype=np.float32) diff = diffusion_map(data) self.assertEqual(diff.size, 0)
def test_sanity1(self):
data = zeros((1, 1), dtype=np.float32)
diff = diffusion_map(data)
self.assertEqual(diff.size, 0)
