def test_bad_inputs():
np.random.seed(1)
single_graph = er_np(100, 0.2)
different_size_graphs = [er_np(100, 0.2)] + [er_np(200, 0.2)]
with pytest.raises(TypeError):
"Invalid unscaled"
mase = MultipleASE(scaled="1")
with pytest.raises(ValueError):
"Test single graph input"
MultipleASE().fit(single_graph)
with pytest.raises(ValueError):
"Test 3-d tensor with 1 graph"
single_graph_tensor = single_graph.reshape(1, 100, -1)
MultipleASE().fit(single_graph_tensor)
with pytest.raises(ValueError):
"Empty list"
MultipleASE().fit([])
with pytest.raises(ValueError):
"Test graphs with different sizes"
MultipleASE().fit(different_size_graphs)
def test_diag_aug():
np.random.seed(5)
n = 100
p = 0.25
graphs_list = [er_np(n, p) for _ in range(2)]
graphs_arr = np.array(graphs_list)
# Test that array and list inputs results in same embeddings
mase_arr = MultipleASE(diag_aug=True).fit_transform(graphs_arr)
mase_list = MultipleASE(diag_aug=True).fit_transform(graphs_list)
assert array_equal(mase_list, mase_arr)
def run(diag_aug, scaled):
# undirected case
np.random.seed(4 + diag_aug + scaled)
X = make_train_undirected(n, m)
res = MultipleASE(n_components=2).fit(X).latent_left_
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 2
# directed case
np.random.seed(5 + diag_aug + scaled)
X = make_train_directed(n, m)
mase = MultipleASE(n_components=2).fit(X)
res = np.hstack([mase.latent_left_, mase.latent_right_])
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 2
def run(diag_aug, scaled):
# undirected case
np.random.seed(2 + diag_aug + scaled)
X = make_train_undirected(n, m)
res = (MultipleASE(2, diag_aug=diag_aug,
scaled=scaled).fit(X).scores_.reshape((m * 4, -1)))
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 4
# directed case
np.random.seed(3 + diag_aug + scaled)
X = make_train_directed(n, m)
res = MultipleASE(2, diag_aug=diag_aug).fit(X).scores_.reshape(
(m * 4, -1))
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 4
def test_vertex():
"""
There should be 2 clusters since each graph is a 2 block model
"""
# undirected case
np.random.seed(4)
n = [128, 128]
m = 10
X = make_train_undirected(n, m)
# undirected case
res = MultipleASE(n_components=2).fit(X).latent_left_
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 2
# directed case
np.random.seed(5)
X = make_train_directed(n, m)
mase = MultipleASE(n_components=2).fit(X)
res = np.hstack([mase.latent_left_, mase.latent_right_])
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 2
# Scaled and undirected case
np.random.seed(4)
X = make_train_undirected(n, m)
res = MultipleASE(n_components=2, scaled=True).fit_transform(X)
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 2
# Scaled and directed case
np.random.seed(5)
X = make_train_directed(n, m)
left, right = MultipleASE(n_components=2, scaled=True).fit_transform(X)
res = np.hstack([left, right])
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 2
def test_graph_clustering():
"""
There should be 4 total clusters since 4 class problem.
n_components = 2
"""
# undirected case
np.random.seed(2)
n = [128, 128]
m = 10
X = make_train_undirected(n, m)
res = MultipleASE(2).fit(X).scores_.reshape((m * 4, -1))
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 4
# directed case
np.random.seed(3)
X = make_train_directed(n, m)
res = MultipleASE(2).fit(X).scores_.reshape((m * 4, -1))
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 4
# Scaled cases
# undirected case
np.random.seed(12)
X = make_train_undirected(n, m)
res = MultipleASE(2, scaled=True).fit(X).scores_.reshape((m * 4, -1))
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 4
# directed case
np.random.seed(13)
X = make_train_directed(n, m)
res = MultipleASE(2, scaled=True).fit(X).scores_.reshape((m * 4, -1))
gmm = GaussianCluster(10, covariance_type="all").fit(res)
assert gmm.n_components_ == 4