def test_all_embedding_models_as_feature(self):
df = get_available_models_for_node_embedding()
bar = tqdm(df.iterrows(),
total=df.shape[0],
leave=False,
desc="Testing embedding methods")
for _, row in bar:
if row.requires_edge_weights:
graph_name = "Usair97"
repository = "networkrepository"
else:
graph_name = "CIO"
repository = "kgobo"
bar.set_description(
f"Testing {row.model_name} from library {row.library_name}")
edge_prediction_evaluation(
holdouts_kwargs=dict(train_size=0.8),
models="Perceptron",
node_features=row.model_name,
evaluation_schema="Connected Monte Carlo",
graphs=graph_name,
repositories=repository,
number_of_holdouts=self._number_of_holdouts,
verbose=False,
smoke_test=True,
)
def test_all_edge_embedding_methods(self):
for edge_embedding_method in GraphSAGEEdgePrediction.get_available_edge_embedding_methods(
):
edge_prediction_evaluation(
holdouts_kwargs=dict(train_size=0.8),
models=GraphSAGEEdgePrediction(
edge_embedding_method=edge_embedding_method, epochs=1),
evaluation_schema="Connected Monte Carlo",
graphs=self._graph,
node_features="SPINE",
number_of_holdouts=self._number_of_holdouts,
verbose=False)
def test_evaluate_edge_prediction_in_subgraph(self):
df = get_available_models_for_edge_prediction()
holdouts = edge_prediction_evaluation(
holdouts_kwargs=dict(train_size=0.8),
models=df.model_name,
library_names=df.library_name,
node_features=SPINE(embedding_size=5),
graphs=self._graph,
number_of_holdouts=self._number_of_holdouts,
verbose=True,
smoke_test=True,
subgraph_of_interest=self._subgraph_of_interest)
self.assertEqual(holdouts.shape[0],
self._number_of_holdouts * 2 * df.shape[0])
def test_evaluate_edge_prediction(self):
df = get_available_models_for_edge_prediction()
holdouts = edge_prediction_evaluation(
holdouts_kwargs=dict(train_size=0.8),
models=df.model_name,
library_names=df.library_name,
node_features=SPINE(embedding_size=5),
evaluation_schema="Connected Monte Carlo",
graphs=[self._graph, self._graph_without_node_types],
number_of_holdouts=self._number_of_holdouts,
verbose=True,
smoke_test=True)
self.assertEqual(holdouts.shape[0],
self._number_of_holdouts * 2 * 2 * df.shape[0])
def test_tree_with_cosine(self):
for evaluation_schema in AbstractEdgePredictionModel.get_available_evaluation_schemas(
):
holdouts = edge_prediction_evaluation(
holdouts_kwargs=dict(train_size=0.8),
models=DecisionTreeEdgePrediction(
edge_embedding_method="CosineSimilarity"),
node_features=SPINE(embedding_size=10),
evaluation_schema=evaluation_schema,
graphs="CIO",
number_of_holdouts=self._number_of_holdouts,
verbose=True,
smoke_test=True,
validation_unbalance_rates=(
1.0,
2.0,
),
)
self.assertEqual(holdouts.shape[0],
self._number_of_holdouts * 2 * 2)
self.assertTrue(
set(holdouts.validation_unbalance_rate) == set((1.0, 2.0)))