def generate_results(network, overwrite=False):
dt = real_networks[network]
roc_df_path = os.path.join("results", network) + "_roc_res.p"
graph, comms, mmatrix, nmap = load_network(dt["path"], network,
dt["clusters"], dt["index"],
dt["node_type"])
if overwrite or not os.path.exists(roc_df_path):
tmm, cmap = construct_true_memberships_matrix(nmap, comms)
print(network, "gen_roc_curves")
roc_results = get_rocs(mmatrix, nmap, comms, tmm)
with open(roc_df_path, "wb+") as roc_df:
pickle.dump(roc_results, roc_df)
roc_df_path = os.path.join("results", network) + "_roc_res_rwr.p"
if overwrite or not os.path.exists(roc_df_path):
nmap = dict([(j, i) for i, j in enumerate(graph.nodes())])
ngraph = nx.Graph()
for edge in graph.edges():
ngraph.add_edge(nmap[edge[0]], nmap[edge[1]])
tmm, cmap = construct_true_memberships_matrix(nmap, comms)
print(network, "gen_roc_curves_rwr")
roc_results = get_rocs_rwr(ngraph, nmap, comms, tmm)
with open(roc_df_path, "wb+") as roc_df:
pickle.dump(roc_results, roc_df)
def auc_compute_rwr(seed, n, ol, results_folder):
pset = _base_params.copy()
pset['n'] = int(n)
pset['overlapping_nodes'] = ol
pset['seed'] = int(seed)
rp = "{}_{}_{}_rwr.json".format(n, ol, seed)
results_file = os.path.abspath(os.path.join(results_folder, rp))
graph, communities = lfr_benchmark_graph(**pset)
nmap = dict([(j, i) for i, j in enumerate(sorted(graph.nodes()))])
tmm, cmap = construct_true_memberships_matrix(nmap, communities)
results = []
for c, comm in communities.items():
for seed_size in _seed_sizes:
if len(comm) > seed_size:
# Seed of AUC scores for node this size
scom = [nmap[i] for i in comm]
auc_s = get_auc_scores_community_rwr(seed_size, scom, graph, nmap, tmm)
results.append([int(n), int(ol), int(seed), c, seed_size, len(comm), np.mean(auc_s), np.std(auc_s)])
with open(results_file, "w+") as rf:
json.dump(results, rf)
def auc_compute(seed, n, ol, results_folder):
pset = _base_params.copy()
pset['n'] = int(n)
pset['overlapping_nodes'] = ol
pset['seed'] = int(seed)
rp = "{}_{}_{}.json".format(n, ol, seed)
results_file = os.path.abspath(os.path.join(results_folder, rp))
graph, communities, index = get_benchmark(pset)
membership_ma, nmap = membership_matrix(graph.nodes(), index)
tmm, cmap = construct_true_memberships_matrix(nmap, communities)
nodes = np.array(list(nmap.values()))
results = []
for c, comm in communities.items():
for seed_size in _seed_sizes:
if len(comm) > seed_size:
# Seed of AUC scores for node this size
scom = np.array([nmap[i] for i in comm])
auc_s = get_auc_scores_community(seed_size, scom, nodes, membership_ma, tmm)
results.append([int(n), int(ol), int(seed), c, seed_size, len(comm), np.mean(auc_s), np.std(auc_s)])
with open(results_file, "w+") as rf:
json.dump(results, rf)