def find_top_k_similar_graphs(self,
graph_dot_file,
graph_name,
k,
num_iter,
cluster_json=None):
gk = GraphKernel(graph_name)
gk.read_dot_graph(graph_dot_file)
if cluster_json:
label_map = gk.read_cluster_info(cluster_json)
gk.relabel_graph(label_map)
gk.init_wl_kernel()
wl = gk.compute_wl_kernel(num_iter)
#graph_size = gk.g.number_of_nodes()
similarity_vector = self.compute_similarity_using_stored_vectors(
wl, num_iter)
wl_pairs = list(zip(self.graphs, similarity_vector))
wl_pairs.sort(key=lambda x: x[1], reverse=True)
if k > len(wl_pairs):
logging.warning(
"Trying to select {0} programs out of only {1} programs.".
format(k, len(wl_pairs)))
#return zip(*wl_pairs)[0]
return wl_pairs[0]
else:
#return zip(*wl_pairs)[0][:k]
return wl_pairs[:k]
kernel_file = sys.argv[2] # write to
num_iter = 3 # WL-Kernel iteration number
total_node_count = 0
total_relabel_count = 0
fo = open(kernel_file, 'w')
for r, ds, fs in os.walk(repo_dir):
for f in fnmatch.filter(fs, '*.dot'):
# build graph kerenel
# print f
gk = GraphKernel(f)
gk.read_dot_graph(os.path.join(r, f))
if len(sys.argv) == 4:
label_map = gk.read_cluster_info(sys.argv[3])
relabel_count = gk.relabel_graph(label_map)
total_node_count += gk.g.number_of_nodes()
total_relabel_count += relabel_count
print("Relabeled {0} out of {1} nodes in {2}.".format(
relabel_count, gk.g.number_of_nodes(), gk.dot_file))
gk.init_wl_kernel()
wls = gk.compute_wl_kernel(num_iter)
wl_str = "###".join([
";;;".join([",,,".join([str(x), str(y)]) for (x, y) in wl])
for wl in wls
])
fo.write(
os.path.join(
os.path.abspath(os.path.join(r, f)) + '\t' + wl_str + '\t' +
str(gk.g.number_of_nodes()) + '\n'))