def test_pca(n_components, n_individual_components, multiview_output):
gpca = GroupPCA(
n_components=n_components,
n_individual_components=n_individual_components,
multiview_output=multiview_output,
)
n_samples = 100
n_features = [6, 4, 5]
rng = np.random.RandomState(0)
Xs = [
rng.multivariate_normal(np.zeros(p), np.eye(p), size=n_samples)
for p in n_features
]
# check the shape of fit.transform
X_r = gpca.fit(Xs).transform(Xs)
if multiview_output:
assert len(X_r) == 3
for X in X_r:
assert X.shape[0] == n_samples
if n_components is not None:
assert X.shape[1] == n_components
else:
assert X_r.shape[0] == n_samples
if n_components is not None:
assert X_r.shape[1] == n_components
# check the equivalence of fit.transform and fit_transform
X_r2 = gpca.fit_transform(Xs)
X_r = gpca.transform(Xs)
assert_allclose(X_r, X_r2)
def test_whitening(n_individual_components, prewhiten, multiview_output):
# Check that PCA output has unit-variance
rng = np.random.RandomState(0)
n_samples = 100
n_features = 80
n_components = 30
rank = 50
# some low rank data with correlated features
X = np.dot(
rng.randn(n_samples, rank),
np.dot(np.diag(np.linspace(10.0, 1.0, rank)),
rng.randn(rank, n_features)),
)
# the component-wise variance of the first 50 features is 3 times the
# mean component-wise variance of the remaining 30 features
X[:, :50] *= 3
assert X.shape == (n_samples, n_features)
# the component-wise variance is thus highly varying:
assert X.std(axis=0).std() > 43.8
Xs = np.array_split(X, 3, axis=1)
print([x.shape for x in Xs])
Xs_ = Xs.copy() # make sure we keep an original across iterations.
gpca = GroupPCA(
n_components=n_components,
whiten=True,
prewhiten=prewhiten,
random_state=0,
n_individual_components=n_individual_components,
multiview_output=multiview_output,
)
# test fit_transform
X_whitened = gpca.fit_transform(Xs_)
X_whitened2 = gpca.transform(Xs_)
assert_allclose(X_whitened, X_whitened2, rtol=5e-4)
if multiview_output:
assert len(X_whitened) == 3
for X in X_whitened:
assert X.shape == (n_samples, n_components)
else:
assert X_whitened.shape == (n_samples, n_components)
assert_allclose(X_whitened.std(ddof=1, axis=0), np.ones(n_components))
Xs_ = Xs.copy()
gpca = GroupPCA(
n_components=n_components,
whiten=False,
prewhiten=prewhiten,
n_individual_components=n_individual_components,
multiview_output=multiview_output,
random_state=rng,
).fit(Xs)
X_unwhitened = gpca.transform(Xs_)
if multiview_output:
assert len(X_unwhitened) == 3
for X in X_unwhitened:
assert X.shape == (n_samples, n_components)
else:
assert X_unwhitened.shape == (n_samples, n_components)
def test_grouppca_deterministic_output():
n_samples = 100
n_features = [6, 4, 5]
rng = np.random.RandomState(0)
Xs = [
rng.multivariate_normal(np.zeros(p), np.eye(p), size=n_samples)
for p in n_features
]
transformed_X = np.zeros((20, 2))
for i in range(20):
pca = GroupPCA(
n_components=2,
n_individual_components=3,
multiview_output=False,
random_state=rng,
)
transformed_X[i, :] = pca.fit_transform(Xs)[0]
assert_allclose(transformed_X,
np.tile(transformed_X[0, :], 20).reshape(20, 2))