def node_label_and_degree_joint_stats(graph_ref_list, graph_pred_list):
sample_ref = []
sample_pred = []
# in case an empty graph is generated
graph_pred_list = [
G for G in graph_pred_list if not G.number_of_nodes() == 0]
prev = datetime.now()
node_map = {}
for graph in graph_ref_list + graph_pred_list:
for u in graph.nodes():
if (graph.degree[u], graph.nodes[u]['label']) not in node_map:
node_map[(graph.degree[u], graph.nodes[u]
['label'])] = len(node_map)
with concurrent.futures.ProcessPoolExecutor(max_workers=MAX_WORKERS) as executor:
for node_label_hist in executor.map(partial(
node_label_and_degree_worker, node_map=node_map), graph_ref_list):
sample_ref.append(node_label_hist)
with concurrent.futures.ProcessPoolExecutor(max_workers=MAX_WORKERS) as executor:
for node_label_hist in executor.map(partial(
node_label_and_degree_worker, node_map=node_map), graph_pred_list):
sample_pred.append(node_label_hist)
mmd_dist = mmd.compute_mmd(
sample_ref, sample_pred, mmd.gaussian_emd, n_jobs=MAX_WORKERS)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing joint node label and degree mmd: ', elapsed)
return mmd_dist
def orbit_stats_all(graph_ref_list, graph_pred_list):
total_counts_ref = []
total_counts_pred = []
graph_pred_list = [
G for G in graph_pred_list if not G.number_of_nodes() == 0]
prev = datetime.now()
with concurrent.futures.ProcessPoolExecutor(max_workers=MAX_WORKERS) as executor:
for orbit_counts_graph in executor.map(orbits_counts_worker, graph_ref_list):
if orbit_counts_graph is not None:
total_counts_ref.append(orbit_counts_graph)
with concurrent.futures.ProcessPoolExecutor(max_workers=MAX_WORKERS) as executor:
for orbit_counts_graph in executor.map(orbits_counts_worker, graph_pred_list):
if orbit_counts_graph is not None:
total_counts_pred.append(orbit_counts_graph)
total_counts_ref = np.array(total_counts_ref)
total_counts_pred = np.array(total_counts_pred)
mmd_dist = mmd.compute_mmd(total_counts_ref, total_counts_pred, metric=partial(
mmd.gaussian, sigma=30.0), is_hist=False, n_jobs=MAX_WORKERS)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing orbit mmd: ', elapsed)
return mmd_dist
def degree_stats(graph_ref_list, graph_pred_list):
"""
Compute the distance between the degree distributions of two unordered sets of graphs.
Args:
graph_ref_list, graph_target_list: two lists of networkx graphs to be evaluated
"""
sample_ref = []
sample_pred = []
# in case an empty graph is generated
graph_pred_list = [
G for G in graph_pred_list if not G.number_of_nodes() == 0]
prev = datetime.now()
with concurrent.futures.ProcessPoolExecutor(max_workers=MAX_WORKERS) as executor:
for deg_hist in executor.map(degree_worker, graph_ref_list):
sample_ref.append(deg_hist)
with concurrent.futures.ProcessPoolExecutor(max_workers=MAX_WORKERS) as executor:
for deg_hist in executor.map(degree_worker, graph_pred_list):
sample_pred.append(deg_hist)
mmd_dist = mmd.compute_mmd(
sample_ref, sample_pred, mmd.gaussian_emd, n_jobs=MAX_WORKERS)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing degree mmd: ', elapsed)
return mmd_dist
def clustering_stats(graph_ref_list, graph_pred_list, bins=100):
sample_ref = []
sample_pred = []
graph_pred_list = [
G for G in graph_pred_list if not G.number_of_nodes() == 0]
prev = datetime.now()
with concurrent.futures.ProcessPoolExecutor(max_workers=MAX_WORKERS) as executor:
for clustering_hist in executor.map(clustering_worker,
[(G, bins) for G in graph_ref_list]):
sample_ref.append(clustering_hist)
with concurrent.futures.ProcessPoolExecutor(max_workers=MAX_WORKERS) as executor:
for clustering_hist in executor.map(clustering_worker,
[(G, bins) for G in graph_pred_list]):
sample_pred.append(clustering_hist)
mmd_dist = mmd.compute_mmd(sample_ref, sample_pred, metric=partial(
mmd.gaussian_emd, sigma=0.1, distance_scaling=bins), n_jobs=MAX_WORKERS)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing clustering mmd: ', elapsed)
return mmd_dist
def node_label_stats(graph_ref_list, graph_pred_list):
"""
Compute the distance between the node label distributions of two unordered sets of graphs.
Args:
graph_ref_list, graph_target_list: two lists of networkx graphs to be evaluated
"""
sample_ref = []
sample_pred = []
# in case an empty graph is generated
graph_pred_list = [G for G in graph_pred_list if not G.number_of_nodes() == 0]
prev = datetime.now()
node_map = {}
for graph in graph_ref_list + graph_pred_list:
for u in graph.nodes():
if graph.nodes[u]['label'] not in node_map:
node_map[graph.nodes[u]['label']] = len(node_map)
with concurrent.futures.ProcessPoolExecutor(max_workers=MAX_WORKERS) as executor:
for node_label_hist in executor.map(partial(node_label_worker, node_map=node_map), graph_ref_list):
sample_ref.append(node_label_hist)
with concurrent.futures.ProcessPoolExecutor(max_workers=MAX_WORKERS) as executor:
for node_label_hist in executor.map(partial(node_label_worker, node_map=node_map), graph_pred_list):
sample_pred.append(node_label_hist)
mmd_dist = mmd.compute_mmd(sample_ref, sample_pred, mmd.gaussian_emd, n_jobs=MAX_WORKERS)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing node label mmd: ', elapsed)
return mmd_dist
def nspdk_stats(graph_ref_list, graph_pred_list):
graph_pred_list = [G for G in graph_pred_list if not G.number_of_nodes() == 0]
prev = datetime.now()
mmd_dist = mmd.compute_mmd(graph_ref_list, graph_pred_list, metric='nspdk',
is_hist=False, n_jobs=MAX_WORKERS)
elapsed = datetime.now() - prev
if PRINT_TIME:
print('Time computing NSPDK mmd: ', elapsed)
return mmd_dist