def test_tensor_input():
X = np.random.normal(size=(100, 5, 5))
mds = ClassicalMDS(n_components=3, dissimilarity="euclidean")
mds.fit(X)
assert_equal(mds.dissimilarity_matrix_.shape, (100, 100))
X_transformed = mds.fit_transform(X)
assert_equal(X_transformed.shape, (100, 3))
def use_fit():
A = np.ones((4, 4)) - np.identity(4)
mds = ClassicalMDS(n_components=3, dissimilarity="precomputed")
mds.fit(A)
B = np.dot(mds.components_, np.diag(mds.singular_values_))
Ahat = _compute_dissimilarity(B)
# Checks up to 7 decimal points
assert_almost_equal(A, Ahat)
def test_input():
X = np.random.normal(0, 1, size=(10, 3))
# X cannot be tensor when precomputed dissimilarity
with pytest.raises(ValueError):
tensor = np.random.normal(0, 1, size=(10, 3, 3))
mds = ClassicalMDS(n_components=3, dissimilarity="precomputed")
mds.fit(tensor)
with pytest.raises(ValueError):
one_dimensional = np.random.normal(size=10)
mds = ClassicalMDS(n_components=2, dissimilarity="euclidean")
mds.fit(one_dimensional)
# n_components > n_samples
with pytest.raises(ValueError):
mds = ClassicalMDS(n_components=100)
mds.fit(X)
# Invalid n_components
with pytest.raises(ValueError):
mds = ClassicalMDS(n_components=-2)
with pytest.raises(TypeError):
mds = ClassicalMDS(n_components="1")
# Invalid dissimilarity
with pytest.raises(ValueError):
mds = ClassicalMDS(dissimilarity="abc")
# Invalid input for fit function
with pytest.raises(ValueError):
mds = ClassicalMDS(n_components=3, dissimilarity="precomputed")
mds.fit(X="bad_input")
# Must be square and symmetric matrix if precomputed dissimilarity
with pytest.raises(ValueError):
mds = ClassicalMDS(n_components=3, dissimilarity="precomputed")
mds.fit(X)