def test__initialize_covariances_case4(self):
""" diagonal covariance, multi-init.
"""
random_state = np.random.RandomState(12)
n_init = 3
means_init_true = random_state.rand(n_init, self.td.K, self.td.P)
covariances_init_true = random_state.rand(n_init, self.td.K, self.td.P)
gmm = GaussianMixture(n_components=self.td.K,
num_feat_full=self.td.P,
num_feat_comp=self.td.Q,
num_feat_shared=self.td.Qs,
num_samp=self.td.N,
transform=self.td.transform,
D_indices=self.td.D_indices,
mask=self.td.mask,
means_init=means_init_true,
covariances_init=covariances_init_true,
covariance_type='diag',
n_init=n_init,
random_state=random_state)
gmm.fit_sparsifier(X=self.td.X)
# init means twice to cycle covariances
_ = gmm._initialize_means()
means_init = gmm._initialize_means()
covariances_init_test = gmm._initialize_covariances(means_init)
self.assertArrayEqual(covariances_init_test, covariances_init_true[1])
def test__initialize_weights_case3(self):
""" multi-init
"""
random_state = np.random.RandomState(12)
n_init = 3
means_init_true = random_state.rand(n_init, self.td.K, self.td.P)
weights_init_true = random_state.rand(n_init, self.td.K)
weights_init_true /= weights_init_true.sum(axis=1)[:, np.newaxis]
gmm = GaussianMixture(n_components=self.td.K,
num_feat_full=self.td.P,
num_feat_comp=self.td.Q,
num_feat_shared=self.td.Qs,
num_samp=self.td.N,
transform=self.td.transform,
D_indices=self.td.D_indices,
mask=self.td.mask,
means_init=means_init_true,
weights_init=weights_init_true,
n_init=n_init,
covariance_type='diag',
random_state=random_state)
gmm.fit_sparsifier(X=self.td.X)
# init means twice to cycle covariances
_ = gmm._initialize_means()
means_init = gmm._initialize_means()
weights_init_test = gmm._initialize_weights(means_init)
self.assertArrayEqual(weights_init_test, weights_init_true[1])
def test__initialize_means_case2(self):
""" means_init is a 3D array.
"""
random_state = np.random.RandomState(12)
n_init = 3
means_init_true = random_state.rand(n_init, self.td.K, self.td.P)
gmm = GaussianMixture(n_components=self.td.K,
num_feat_full=self.td.P,
num_feat_comp=self.td.Q,
num_feat_shared=self.td.Qs,
num_samp=self.td.N,
transform=self.td.transform,
D_indices=self.td.D_indices,
mask=self.td.mask,
means_init=means_init_true,
n_init=n_init,
random_state=random_state)
gmm.fit_sparsifier(X=self.td.X)
# first one is discarded for this test
_ = gmm._initialize_means()
# this should recover the second one
means_init_test = gmm._initialize_means()
self.assertArrayEqual(means_init_test, means_init_true[1])
def test__initialize_means_case1(self):
""" means_init is a 2D array.
"""
random_state = np.random.RandomState(12)
means_init_true = random_state.rand(self.td.K, self.td.P)
gmm = GaussianMixture(n_components=self.td.K,
num_feat_full=self.td.P,
num_feat_comp=self.td.Q,
num_feat_shared=self.td.Qs,
num_samp=self.td.N,
transform=self.td.transform,
D_indices=self.td.D_indices,
mask=self.td.mask,
means_init=means_init_true,
random_state=random_state)
gmm.fit_sparsifier(X=self.td.X)
means_init_test = gmm._initialize_means()
self.assertArrayEqual(means_init_test, means_init_true)
def test__initialize_means_case4(self):
""" means_init is None, init_params is 'random'.
Only checks that the initialized means are of the correct shape.
"""
random_state = np.random.RandomState(12)
gmm = GaussianMixture(n_components=self.td.K,
num_feat_full=self.td.P,
num_feat_comp=self.td.Q,
num_feat_shared=self.td.Qs,
num_samp=self.td.N,
transform=self.td.transform,
D_indices=self.td.D_indices,
mask=self.td.mask,
means_init=None,
init_params='random',
random_state=random_state)
gmm.fit_sparsifier(X=self.td.X)
means_init_shape_test = gmm._initialize_means().shape
means_init_shape_true = np.array([self.td.K, self.td.P])
self.assertArrayEqual(means_init_shape_test, means_init_shape_true)
def test__initialize_covariances_case3(self):
""" No covariances given, just check shape.
"""
random_state = np.random.RandomState(12)
means_init_true = random_state.rand(self.td.K, self.td.P)
gmm = GaussianMixture(n_components=self.td.K,
num_feat_full=self.td.P,
num_feat_comp=self.td.Q,
num_feat_shared=self.td.Qs,
num_samp=self.td.N,
transform=self.td.transform,
D_indices=self.td.D_indices,
mask=self.td.mask,
means_init=means_init_true,
covariances_init=None,
covariance_type='diag',
random_state=random_state)
gmm.fit_sparsifier(X=self.td.X)
means_init = gmm._initialize_means()
covariances_init_test = gmm._initialize_covariances(means_init)
true_shape = np.array((self.td.K, self.td.P))
self.assertArrayEqual(covariances_init_test.shape, true_shape)
def test__initialize_covariances_case1(self):
""" spherical covariance, 1 init.
"""
random_state = np.random.RandomState(12)
means_init_true = random_state.rand(self.td.K, self.td.P)
covariances_init_true = random_state.rand(self.td.K)
gmm = GaussianMixture(n_components=self.td.K,
num_feat_full=self.td.P,
num_feat_comp=self.td.Q,
num_feat_shared=self.td.Qs,
num_samp=self.td.N,
transform=self.td.transform,
D_indices=self.td.D_indices,
mask=self.td.mask,
means_init=means_init_true,
covariances_init=covariances_init_true,
covariance_type='spherical',
random_state=random_state)
gmm.fit_sparsifier(X=self.td.X)
means_init = gmm._initialize_means()
covariances_init_test = gmm._initialize_covariances(means_init)
self.assertArrayEqual(covariances_init_test, covariances_init_true)