def test_wasserstein_vectorizer_lists():
vectorizer = WassersteinVectorizer(random_state=42)
result = vectorizer.fit_transform(distributions_data_list,
vectors=vectors_data_list)
transform_result = vectorizer.transform(distributions_data_list,
vectors=vectors_data_list)
assert np.allclose(result, transform_result, rtol=1e-3, atol=1e-6)
def test_wasserstein_vectorizer_list_based_blockwise():
lil_data = normalize(distributions_data, norm="l1").tolil()
distributions = [np.array(x) for x in lil_data.data]
vectors = [vectors_data[x] for x in lil_data.rows]
vectorizer = WassersteinVectorizer(random_state=42, memory_size="50k")
result = vectorizer.fit_transform(distributions, vectors=vectors)
transform_result = vectorizer.transform(distributions, vectors=vectors)
assert np.allclose(result, transform_result, rtol=1e-3, atol=1e-6)
def test_wasserstein_vectorizer_generators_blockwise():
distributions_data_generator = (x for x in distributions_data_list)
vectors_data_generator = (x for x in vectors_data_list)
vectorizer = WassersteinVectorizer(random_state=42, memory_size="50k")
result = vectorizer.fit_transform(
distributions_data_generator,
vectors=vectors_data_generator,
reference_distribution=generator_reference_dist,
reference_vectors=generator_reference_vectors,
n_distributions=distributions_data.shape[0],
vector_dim=vectors_data.shape[1],
)
distributions_data_generator = (x for x in distributions_data_list)
vectors_data_generator = (x for x in vectors_data_list)
transform_result = vectorizer.transform(
distributions_data_generator,
vectors=vectors_data_generator,
n_distributions=distributions_data.shape[0],
vector_dim=vectors_data.shape[1],
)
assert np.allclose(result, transform_result, rtol=1e-3, atol=1e-6)
def test_wasserstein_vectorizer_list_compared_to_sparse():
lil_data = normalize(distributions_data.astype(np.float64), norm="l1").tolil()
distributions = [np.array(x) for x in lil_data.data]
vectors = [vectors_data[x] for x in lil_data.rows]
vectorizer_sparse = WassersteinVectorizer(random_state=42)
result_sparse = vectorizer_sparse.fit_transform(
distributions_data, vectors=vectors_data
)
vectorizer_list = WassersteinVectorizer(random_state=42)
result_list = vectorizer_list.fit_transform(
distributions,
vectors=vectors,
reference_distribution=vectorizer_sparse.reference_distribution_,
reference_vectors=vectorizer_sparse.reference_vectors_,
)
assert np.allclose(result_sparse, result_list, rtol=1e-3, atol=1e-6)
def test_wasserstein_vectorizer_generator_compared_to_sparse():
distributions_data_generator = (x for x in distributions_data_list)
vectors_data_generator = (x for x in vectors_data_list)
vectorizer_sparse = WassersteinVectorizer(random_state=42)
result_sparse = vectorizer_sparse.fit_transform(distributions_data,
vectors=vectors_data)
vectorizer_gen = WassersteinVectorizer(random_state=42)
result_list = vectorizer_gen.fit_transform(
distributions_data_generator,
vectors=vectors_data_generator,
reference_distribution=vectorizer_sparse.reference_distribution_,
reference_vectors=vectorizer_sparse.reference_vectors_,
n_distributions=distributions_data.shape[0],
vector_dim=vectors_data.shape[1],
)
assert np.allclose(result_sparse, result_list, rtol=1e-3, atol=1e-6)
def test_wasserstein_based_vectorizer_bad_params(wasserstein_class):
with pytest.raises(ValueError):
vectorizer = wasserstein_class()
vectorizer.fit(distributions_data)
with pytest.raises(ValueError):
vectorizer = wasserstein_class()
vectorizer.fit(mixed_token_data, vectors=vectors_data)
with pytest.raises(ValueError):
vectorizer = wasserstein_class()
vectorizer.fit(point_data, vectors=vectors_data)
distributions_data_generator = (x for x in distributions_data_list)
vectors_data_generator = (x for x in vectors_data_list)
with pytest.raises(ValueError):
vectorizer = WassersteinVectorizer()
vectorizer.fit(distributions_data_generator, vectors=vectors_data_generator)
distributions_data_generator = (x for x in distributions_data_list)
vectors_data_generator = (x for x in vectors_data_list)
with pytest.raises(ValueError):
vectorizer = WassersteinVectorizer()
vectorizer.fit(
distributions_data_generator,
vectors=vectors_data_generator,
reference_vectors=np.random.random((10, vectors_data.shape[1])),
)
distributions_data_generator = (x for x in distributions_data_list)
vectors_data_generator = (x for x in vectors_data_list)
with pytest.raises(ValueError):
vectorizer = WassersteinVectorizer(reference_size=20)
vectorizer.fit(
distributions_data_generator,
vectors=vectors_data_generator,
reference_vectors=np.random.random((10, vectors_data.shape[1])),
)
def test_wasserstein_vectorizer_blockwise():
vectorizer = WassersteinVectorizer(random_state=42, memory_size="50k")
result = vectorizer.fit_transform(distributions_data, vectors=vectors_data)
transform_result = vectorizer.transform(distributions_data, vectors=vectors_data)
assert np.allclose(result, transform_result, rtol=1e-3, atol=1e-6)