def test_transform(Xs):
ica = MultiviewICA(n_components=2)
with pytest.raises(ValueError):
ica.transform(Xs)
assert(ica.fit_transform(Xs).shape == (Xs.shape[0], Xs.shape[1], 2))
ica = MultiviewICA()
assert(ica.fit_transform(Xs).shape == Xs.shape)
def test_fit_errors(Xs):
with pytest.raises(ValueError):
ica = MultiviewICA()
ica.fit(Xs[:,:5, :])
with pytest.raises(ValueError):
ica = MultiviewICA(init='WRONG')
ica.fit(Xs)
with pytest.raises(TypeError):
ica = MultiviewICA(init=list())
ica.fit(Xs)
def test_inverse_transform(Xs):
ica = MultiviewICA(n_components=2)
with pytest.raises(ValueError):
ica.transform(Xs)
ica = ica.fit(Xs)
Xs_mixed = ica.inverse_transform()
avg_mixed = np.mean([X @ C for X,C in zip(Xs, ica.components_)], axis=0)
for X_mixed in Xs_mixed:
assert np.linalg.norm(X_mixed - avg_mixed) < 0.5
def test_inverse_transform(Xs, multiview_output):
ica = MultiviewICA(n_components=2, multiview_output=multiview_output)
with pytest.raises(ValueError):
ica.inverse_transform(Xs)
S = ica.fit_transform(Xs)
Xs_mixed = ica.inverse_transform(S)
avg_mixed = np.mean(
[X @ np.linalg.pinv(C) for X, C in zip(Xs, ica.pca_components_)],
axis=0,
)
avg_mixed2 = np.mean(
[X @ np.linalg.pinv(C) for X, C in zip(Xs_mixed, ica.pca_components_)],
axis=0,
)
assert np.linalg.norm(avg_mixed2 - avg_mixed) < 0.2
def test_index(index, inverse_index, multiview_output):
# Test that the projection of data can be inverted
rng = np.random.RandomState(0)
n, p = 50, 3
X = rng.randn(n, p) # spherical data
X[:, 1] *= 0.00001 # make middle component relatively small
X += [5, 4, 3] # make a large mean
X2 = np.copy(X)
X2[:, 1] += rng.randn(n) * 0.00001
X2 = X2.dot(rng.rand(p, p))
X3 = np.copy(X)
X3[:, 1] += rng.randn(n) * 0.00001
X3 = X3.dot(rng.rand(p, p))
Xs = [X, X2, X3]
ica = MultiviewICA(
n_components=2,
init="groupica",
multiview_output=multiview_output,
).fit(Xs)
if index is not None:
index_ = np.atleast_1d(index)
Xs_transform = [Xs[i] for i in index_]
len_index = len(index_)
else:
len_index = 3
Xs_transform = np.copy(Xs)
if inverse_index is not None:
inverse_index_ = np.atleast_1d(inverse_index)
Xs_inverse = [Xs[i] for i in inverse_index_]
len_inverse_index = len(inverse_index_)
else:
len_inverse_index = 3
Xs_inverse = np.copy(Xs)
Y = ica.transform(Xs_transform, index=index)
Y_inverse = ica.inverse_transform(Y, index=inverse_index)
for X, X_estimated in zip(Xs_inverse, Y_inverse):
np.testing.assert_allclose(X, X_estimated, atol=1e-3)
def test_fit(Xs, capfd):
ica = MultiviewICA(verbose=True)
ica.fit(Xs)
out, err = capfd.readouterr()
assert out[:2] == "it"
def test_inverse_transform_no_preproc(Xs):
ica = MultiviewICA()
S = ica.fit_transform(Xs)
Xs_mixed = ica.inverse_transform(S)
assert np.mean((Xs_mixed - Xs)**2) / np.mean(Xs**2) < 0.05