def test_correct_labels_resampled():
dataset = datasets.omitted_with_actions(['LT1001_TA05.net'], resample=True)
labels = [get_label_type(dataset, graph) for graph in dataset]
unique, counts = np.unique(labels, return_counts=True)
label_dist = dict(zip(unique, counts))
for (label, count) in label_dist.items():
max_count = expected_label_count[label]
assert count <= max_count, f'Expected <= {max_count} #{label} labels but found {count}'
def test_targets_should_be_different_types():
dataset = datasets.omitted_with_actions(['LT1001_TA05.net'],
resample=False,
normalize=False)
for graph in dataset:
prototype_idx = (graph.x[:, -1] == 1).nonzero()[0]
valid_target_idx = (graph.y > -1).nonzero()[0]
for (prototype, target) in zip(prototype_idx, valid_target_idx):
assert prototype == target
def test_only_one_prototype_per_type():
dataset = datasets.omitted_with_actions(['LT1001_TA05.net'],
resample=False,
normalize=False)
for graph in dataset:
prototype_idx = (graph.x[:, -1] == 1).nonzero()[0]
prototypes = graph.x[prototype_idx]
type_counts = np.sum(prototypes, axis=0)[0:-1]
assert (type_counts > 1).nonzero()[0].size == 0
def test_components_should_be_connected():
dataset = datasets.omitted_with_actions(['LT1001_TA05.net'],
resample=False,
normalize=False,
shuffle=False)
for (i, graph) in enumerate(dataset):
adj = graph.a.toarray()
edge_counts = np.sum(adj, axis=0)
for (edge_index, edge_count) in enumerate(edge_counts):
assert edge_count > 0, f'Found disconnected node: {edge_index}'
def test_no_dupe_types_in_proto():
dataset = datasets.omitted_with_actions(['LT1001_TA05.net'],
resample=False,
normalize=False,
shuffle=False)
for (i, graph) in enumerate(dataset):
targets = graph.y
prototypes = graph.x[(targets > -1).nonzero()[0]]
node_types = dataset.get_node_types(prototypes)
types, counts = np.unique(node_types, return_counts=True)
type_dict = dict(zip(types, counts))
for (node_type, count) in type_dict.items():
assert count == 1, f'Found {count} nodes of type {node_type}'
def test_get_node_types():
dataset = datasets.omitted_with_actions(['LT1001_TA05.net'],
resample=False)
graph = dataset[0]
types = dataset.get_node_types(graph.x)
unique, counts = np.unique(types, return_counts=True)
single_graph_count = dict(zip(unique, counts))
found_omitted_type = False
for (label, component_count) in expected_label_count.items():
count = single_graph_count[label]
if count != (component_count + 1):
if count == component_count and not found_omitted_type:
found_omitted_type = True
continue
raise Exception(
f'Expected {expected+1} #{label} labels but found {count}')
def test_to_networkx():
dataset = datasets.omitted_with_actions([filename], shuffle=False)
nx_data = datasets.helpers.to_networkx(dataset)
assert len(dataset) == len(
nx_data), f'Expected to find 20 graphs. Found {len(nx_data)}'
for (i, nx_graph) in enumerate(nx_data):
# Check node counts
assert len(
nx_graph.nodes
) == 19, f'Expected 19 nodes. Found {nx_graph.node_feature.shape[0]}'
# Check edge counts
graph = dataset[i]
expected_edge_count = graph.n_edges / 2 # Converting to bidirectional graph
edge_count = len(nx_graph.edges)
print('----', len(dataset.helpers.component_types))
assert edge_count == expected_edge_count, f'Expected {expected_edge_count} edges. Found {edge_count}'
def test_load_graphs():
dataset = datasets.omitted_with_actions(['LT1001_TA05.net'],
resample=False)
assert len(
dataset) == 20, f'Expected to find 20 graphs. Found {len(dataset)}'
def test_has_edges():
dataset = datasets.omitted_with_actions(['LT1001_TA05.net'],
resample=False,
normalize=False)
for graph in dataset:
assert graph.n_edges > 0