def test_serialize(self):
random = np.random.RandomState(seed=42)
node_features = generate_random_node_features(random, 1, 3, 1)
input = TensorGraph(
node_features,
np.array([[
[[0, 0, 1], [0, 0, 1], [1, 1, 0]],
[[1, 2, 2], [2, 3, 5], [2, 5, 3]],
]],
dtype=float))
expected_output = TensorGraph(
node_features,
np.array([[
[[0, 0, 0.71], [0, 0, 0.71], [0.71, 0.71, 0]],
[[0.2, 0.28, 0.28], [0.28, 0.3, 0.5], [0.28, 0.5, 0.3]],
]],
dtype=float))
model_input = TensorGraphInput(
TensorGraphShape(num_nodes=3, edge_dims=2))
x = NormaliseGraph(method='spectral')(model_input)
model = tf.keras.Model(model_input, x)
config = model.get_config()
reinitialized_model = tf.keras.Model.from_config(config)
output = reinitialized_model(input)
assert_graph_equal(self,
expected_output,
output,
allow_nodes_close=True,
allow_edges_close=True)
def test_serialize(self):
random = np.random.RandomState(seed=42)
node_features = generate_random_node_features(random, 1, 3, 1)
input = TensorGraph(
node_features,
np.array([[
[[0, 1, 0], [1, 0, 0], [0, 1, 0]],
[[0, 2, 3], [2, 0, 4], [3, 4, 0]],
]],
dtype=float))
expected_output = TensorGraph(
node_features,
np.array([[
[[1, 1, 0], [1, 1, 0], [0, 1, 1]],
[[1, 2, 3], [2, 1, 4], [3, 4, 1]],
]],
dtype=float))
model_input = TensorGraphInput(
TensorGraphShape(num_nodes=3, edge_dims=2))
x = AddSelfLoops()(model_input)
model = tf.keras.Model(model_input, x)
config = model.get_config()
reinitialized_model = tf.keras.Model.from_config(config)
output = reinitialized_model(input)
assert_graph_equal(self,
expected_output,
output,
allow_nodes_close=True)
def _test_call(self, input, expected_output, **kwargs):
graph_input = TensorGraph(None, input)
graph_expected = TensorGraph(None, expected_output)
shiftTensor = AddSelfLoops(**kwargs)
input_shape = graph_tensor_shape(graph_input)
computed_shape = shiftTensor.compute_output_shape(input_shape)
graph_output = shiftTensor(graph_input)
assert_graph_equal(self, graph_expected, graph_output)
assert_graph_shape(self, graph_output, computed_shape)
def _test_call(self, input, expected_output, **kwargs):
graph_input = TensorGraph(None, input)
graph_expected = TensorGraph(None, expected_output)
shiftTensor = NormaliseGraph(**kwargs)
input_shape = graph_tensor_shape(graph_input)
computed_shape = shiftTensor.compute_output_shape(input_shape)
graph_output = shiftTensor(graph_input)
assert_graph_equal(self,
graph_expected,
graph_output,
allow_edges_close=True)
assert_graph_shape(self, graph_output, computed_shape)
def tfrecord_to_tensorgraph(example: dict, name: str = 'graph'):
num_nodes = example[name + ':num_nodes']
node_features = tf.sparse.to_dense(example[name + ':node_features'])
node_features = tf.reshape(node_features, (num_nodes, -1))
edge_features = tf.sparse.to_dense(example[name + ':edge_features'])
edge_features = tf.reshape(edge_features, (-1, num_nodes, num_nodes))
return TensorGraph(node_features, edge_features)
def test_node_and_edge_features(self):
random = np.random.RandomState(seed=42)
node_features = generate_random_node_features(random, 3, 10, 42)
edge_features = generate_random_adjacency(random, 3, 10, 7)
node_tensor = tf.convert_to_tensor(node_features, dtype=tf.float32)
edge_tensor = tf.convert_to_tensor(edge_features, dtype=tf.float32)
graph = TensorGraph(node_tensor, edge_tensor)
self.assertAllEqual(node_tensor, graph[NODE_FEATURES])
self.assertAllEqual(edge_tensor, graph[EDGE_FEATURES])