def test_truncated_fit_transform(self):
trans = FastWLGraphKernelTransformer(h=2, use_early_stopping=False, truncate_to_highest_label=True)
X = helper.get_graphs()
num_train = int(len(X) * 0.9)
X_train, X_test = X[:num_train], X[num_train:]
trans.fit(X)
phi_total = trans.transform(X)
trans = sklearn.base.copy.copy(trans)
trans.fit(X_train)
phi_train = trans.transform(X_train)
phi_test = trans.transform(X_test)
phi_total_ = phi_total[-1]
phi_train_ = phi_train[-1]
phi_test_ = phi_test[-1]
phi_total_splitted_ = scipy.sparse.vstack([phi_train_, phi_test_])
self.assertEqual(phi_total_.shape[1], np.max(phi_total_.nonzero()[1]) + 1)
def get_gram(phi):
return phi.dot(phi.T)
gram_total = get_gram(phi_total_)
gram_total_splitted = get_gram(phi_total_splitted_)
self.assertNotEqual((gram_total - gram_total_splitted).nnz, 0)
def test_early_stop(self):
t = FastWLGraphKernelTransformer(h=10,
use_early_stopping=True,
truncate_to_highest_label=True)
X = helper.get_graphs()
X = X[:100]
phi_list = t.fit_transform(X)
feature_nums = [phi.shape[1] for phi in phi_list]
self.assertEqual(len(set(feature_nums)), len(feature_nums))
def test_truncated_with_no_phi_dim_with_transform(self):
t = FastWLGraphKernelTransformer(h=2, use_early_stopping=False, truncate_to_highest_label=True)
X = helper.get_graphs()
num_train = 1000
X_train, X_test = X[:num_train], X[num_train:]
phi_list = t.fit_transform(X_train)
phi_list_test = t.transform(X_test)
for phi_train, phi_test in zip(phi_list, phi_list_test):
self.assertEqual(phi_train.shape[1], phi_test.shape[1])
def test_normalize_wl_without_node_weights(self):
t = FastWLGraphKernelTransformer(
h=2,
use_early_stopping=False,
truncate_to_highest_label=True,
# This is important
norm='l1'
)
X = helper.get_graphs()
X = X[:100]
phi_list = t.fit_transform(X)
for phi in phi_list:
# Assert that phi is normalized
self.assertTrue(np.allclose(phi.sum(axis=1), np.ones(phi.shape[0])))
def get_phi_list(self, **fast_wl_params):
t = FastWLGraphKernelTransformer(**fast_wl_params)
X = helper.get_graphs()
phi_list = t.fit_transform(X)
return phi_list