def test_model_wasserstein():
random_seed = check_random_state(42)
model = PersistenceVectorizer(n_components=4,
random_state=random_seed).fit(base_data)
model_dmat = model.pairwise_p_wasserstein_distance(base_data[:10])
random_seed = check_random_state(42)
gmm = GaussianMixture(n_components=4,
random_state=random_seed).fit(base_data)
vec_data = [vectorize_diagram(base_data[i], gmm) for i in range(10)]
raw_ground_distance = pairwise_gaussian_ground_distance(
gmm.means_,
gmm.covariances_,
)
ground_distance = add_birth_death_line(
raw_ground_distance,
gmm.means_,
gmm.covariances_,
y_axis="lifetime",
)
util_dmat = pairwise_distances(
vec_data,
metric=persistence_wasserstein_distance,
ground_distance=ground_distance,
)
assert np.allclose(model_dmat, util_dmat)
def test_vectorize_diagram():
gmm = GaussianMixture(n_components=4, random_state=42).fit(base_data)
result = vectorize_diagram(np.array([[0.5, 0.2], [0.75, 0.1]]), gmm)
assert np.allclose(
result, np.array([6.24722853e-02, 5.50441490e-33, 0.00000000e00, 1.93752771e00])
)
def test_persistence_p_wasserstein_distance():
gmm = GaussianMixture(n_components=4, random_state=42).fit(base_data[10:90])
v1 = vectorize_diagram(base_data[:10], gmm)
v2 = vectorize_diagram(base_data[-10:], gmm)
raw_ground_distance = pairwise_gaussian_ground_distance(
gmm.means_, gmm.covariances_,
)
ground_distance = add_birth_death_line(
raw_ground_distance, gmm.means_, gmm.covariances_, y_axis="lifetime",
)
d = persistence_p_wasserstein_distance(v1, v2, ground_distance)
assert np.isclose(d, 0.94303)
def test_pervect_transform_umap():
random_seed = check_random_state(42)
gmm = GaussianMixture(n_components=4,
random_state=random_seed).fit(base_data)
util_result = np.array(
[vectorize_diagram(diagram, gmm) for diagram in base_data])
model = PersistenceVectorizer(
n_components=4,
random_state=42,
apply_umap=True,
).fit(base_data)
model_result = model.transform(base_data)
assert np.allclose(model.mixture_model_.means_, gmm.means_)
assert np.allclose(model.mixture_model_.covariances_, gmm.covariances_)
umap_util_result = umap.UMAP(metric=umap_metric,
random_state=42).fit_transform(util_result)
assert np.allclose(model_result, umap_util_result)
def test_pervect_transform_base():
random_seed = check_random_state(42)
model = PersistenceVectorizer(n_components=4,
random_state=random_seed).fit(base_data)
model_result = model.transform(base_data)
random_seed = check_random_state(42)
gmm = GaussianMixture(n_components=4,
random_state=random_seed).fit(base_data)
util_result = np.array(
[vectorize_diagram(diagram, gmm) for diagram in base_data])
assert np.allclose(model.mixture_model_.means_, gmm.means_)
assert np.allclose(model.mixture_model_.covariances_, gmm.covariances_)
assert np.allclose(model_result, util_result)
random_seed = check_random_state(42)
model_result = PersistenceVectorizer(
n_components=4, random_state=random_seed).fit_transform(base_data)
assert np.allclose(model_result, util_result)