def test_different_inits(self):
np.random.seed(314)
mean = np.ones(3) * 5
cov = np.eye(3) * 0.1
X = stats.multivariate_normal.rvs(mean, cov, 100)
print(X.shape)
Y = stats.multivariate_normal.rvs(mean, cov, 100)
W = stats.ortho_group.rvs(3)
Y = Y @ W
aligner_1 = SeedlessProcrustes(init="2d")
aligner_1.fit(X, Y)
aligner_2 = SeedlessProcrustes(init="sign_flips")
aligner_2.fit(X, Y)
test_sign_flips = SignFlips()
self.assertTrue(
np.all(
test_sign_flips.fit(X, Y).Q_ == aligner_2.selected_initial_Q_))
aligner_3 = SeedlessProcrustes(init="custom")
aligner_3.fit(X, Y)
self.assertTrue(np.all(np.eye(3) == aligner_3.selected_initial_Q_))
aligner_4 = SeedlessProcrustes(init="custom", initial_Q=-np.eye(3))
aligner_4.fit(X, Y)
self.assertTrue(np.all(-np.eye(3) == aligner_4.selected_initial_Q_))
aligner_5 = SeedlessProcrustes(init="custom",
initial_P=np.ones((100, 100)) / 10000)
aligner_5.fit(X, Y)
def test_max_criterion(self):
X = np.arange(6).reshape(3, 2) * (-1)
Y = np.arange(6).reshape(3, 2) @ np.diag([1, -1]) + 0.5
# in this case, Y should be unchanged, and X matched to Y
# so X flips sign in the first dimension
Q_answer = np.array([[-1, 0], [0, 1]])
X_answer = X.copy() @ Q_answer
# set criterion to "max", see if that works
aligner = SignFlips(criterion="max")
aligner.fit(X, Y)
Q_test = aligner.Q_
X_test = aligner.transform(X)
self.assertTrue(np.all(Q_test == Q_answer))
self.assertTrue(np.all(X_test == X_answer))
def test_sim(self):
n = 150
rho = 0.9
n_per_block = int(n / 3)
n_blocks = 3
block_members = np.array(n_blocks * [n_per_block])
block_probs = np.array([[0.2, 0.01, 0.01], [0.01, 0.1, 0.01],
[0.01, 0.01, 0.2]])
directed = False
loops = False
A1, A2 = sbm_corr(block_members,
block_probs,
rho,
directed=directed,
loops=loops)
ase = AdjacencySpectralEmbed(n_components=3, algorithm="truncated")
x1 = ase.fit_transform(A1)
x2 = ase.fit_transform(A2)
xh1 = SignFlips().fit_transform(x1, x2)
S = xh1 @ x2.T
res = self.barygm.fit(A1, A2, S=S)
self.assertTrue(0.7 <= (sum(res.perm_inds_ == np.arange(n)) / n))
A1 = A1[:-1, :-1]
xh1 = xh1[:-1, :]
S = xh1 @ x2.T
res = self.barygm.fit(A1, A2, S=S)
self.assertTrue(0.6 <= (sum(res.perm_inds_ == np.arange(n)) / n))
def test_bad_kwargs(self):
with self.assertRaises(TypeError):
SignFlips(criterion={"this is a": "dict"})
with self.assertRaises(ValueError):
SignFlips(criterion="cep")
# check delayed ValueError
with self.assertRaises(ValueError):
aligner = SignFlips(criterion="median")
X = np.arange(6).reshape(6, 1)
Y = np.arange(6).reshape(6, 1)
aligner.criterion = "something"
aligner.fit(X, Y)
def test_two_datasets(self):
X = np.arange(6).reshape(3, 2) * (-1)
Y = np.arange(6).reshape(3, 2) @ np.diag([1, -1]) + 0.5
# X flips sign in the first dimension
Q_answer = np.array([[-1, 0], [0, 1]])
X_answer = X.copy() @ Q_answer
# first, do fit and transform separately
aligner_1 = SignFlips()
aligner_1.fit(X, Y)
Q_test = aligner_1.Q_
X_test = aligner_1.transform(X)
self.assertTrue(np.all(Q_test == Q_answer))
self.assertTrue(np.all(X_test == X_answer))
# now, do fit_transform
aligner_2 = SignFlips()
X_test = aligner_2.fit_transform(X, Y)
Q_test = aligner_2.Q_
self.assertTrue(np.all(Q_test == Q_answer))
self.assertTrue(np.all(X_test == X_answer))
# try giving a different matrix as the sole input (I)
I_test = aligner_2.transform(np.eye(2))
I_answer = np.diag([-1, 1])
self.assertTrue(np.all(I_test == I_answer))
def test_bad_datasets(self):
X = np.arange(6).reshape(6, 1)
Y = np.arange(6).reshape(6, 1)
Y_wrong_d = np.arange(12).reshape(6, 2)
# check passing weird stuff as input (caught by us)
with self.assertRaises(TypeError):
aligner = SignFlips()
aligner.fit("hello there", Y)
with self.assertRaises(TypeError):
aligner = SignFlips()
aligner.fit(X, "hello there")
with self.assertRaises(TypeError):
aligner = SignFlips()
aligner.fit({"hello": "there"}, Y)
with self.assertRaises(TypeError):
aligner = SignFlips()
aligner.fit(X, {"hello": "there"})
# check passing arrays of weird ndims (caught by check_array)
with self.assertRaises(ValueError):
aligner = SignFlips()
aligner.fit(X, Y.reshape(3, 2, 1))
with self.assertRaises(ValueError):
aligner = SignFlips()
aligner.fit(X.reshape(3, 2, 1), Y)
# check passing arrays with different dimensions (caught by us)
with self.assertRaises(ValueError):
aligner = SignFlips()
aligner.fit(X, Y_wrong_d)
# check passing array with wrong dimensions to transform (caught by us)
with self.assertRaises(ValueError):
aligner = SignFlips()
aligner.fit(X, Y)
aligner.transform(Y_wrong_d)