def test_predict_all_with_novelties(self):
"""Test scoring all triples with labeling as novel w.r.t. training and testing."""
all_df = get_all_prediction_df(
model=self.model,
triples_factory=self.dataset.training,
testing=self.testing_mapped_triples,
)
self.assertIsInstance(all_df, pd.DataFrame)
self.assertEqual(
[
'head_id',
'head_label',
'relation_id',
'relation_label',
'tail_id',
'tail_label',
'score',
'in_training',
'in_testing',
],
list(all_df.columns),
)
possible = self.dataset.training.num_relations * self.model.num_entities**2
self.assertEqual(possible, len(all_df.index))
self.assertEqual(self.dataset.training.num_triples,
all_df['in_training'].sum())
self.assertEqual(self.testing_mapped_triples.shape[0],
all_df['in_testing'].sum())
def test_predict_all_no_novelties(self):
"""Test scoring all triples without labeling as novel w.r.t. training and testing."""
all_df = get_all_prediction_df(model=self.model, testing=self.testing_mapped_triples, add_novelties=False)
self.assertIsInstance(all_df, pd.DataFrame)
self.assertEqual(
['head_id', 'head_label', 'relation_id', 'relation_label', 'tail_id', 'tail_label', 'score'],
list(all_df.columns),
)
possible = self.model.triples_factory.num_relations * self.model.num_entities ** 2
self.assertEqual(possible, len(all_df.index))
def test_predict_all_remove_known(self):
"""Test scoring all triples while removing non-novel triples w.r.t. training and testing."""
all_df = get_all_prediction_df(model=self.model, testing=self.testing_mapped_triples, remove_known=True)
self.assertIsInstance(all_df, pd.DataFrame)
self.assertEqual(
['head_id', 'head_label', 'relation_id', 'relation_label', 'tail_id', 'tail_label', 'score'],
list(all_df.columns),
)
possible = self.model.triples_factory.num_relations * self.model.num_entities ** 2
known = self.model.triples_factory.num_triples + self.testing_mapped_triples.shape[0]
self.assertNotEqual(possible, known, msg='testing and training triples cover all possible triples')
self.assertEqual(possible - known, len(all_df.index))
def test_get_all_prediction_df(self):
"""Test consistency of top-k scoring."""
ks = [5, 10]
dfs = [
get_all_prediction_df(
model=self.instance,
triples_factory=self.factory,
batch_size=1,
k=k,
).nlargest(n=min(ks), columns="score").reset_index(drop=True)
for k in ks
]
assert set(dfs[0].columns) == set(dfs[0].columns)
for column in dfs[0].columns:
numpy.testing.assert_equal(dfs[0][column].values,
dfs[1][column].values)