from_timestamp = timestamp(edge_list[start])
to_timestamp = timestamp(edge_list[end])
g = subgraph_by_timestamp(
mg,
from_timestamp,
to_timestamp,
)
average_degree = 2 * g.number_of_edges() / g.number_of_nodes()
embedding = model.deepwalk(g)
t0 = time.time()
comm_list = model.ddcrp(g, embedding, ddcrp_scale=scale, receptive_hop=hop)
ddcrp_time = time.time() - t0
comm_list = comm_list[ddcrp_cutoff:]
init_comm, mapping = model.mcla(comm_list, comm)
predicted_cluster_size = len(init_comm)
new_comm = model.kmeans(embedding, init_comm)
def response(new_comm: List[Set[int]]) -> Any:
out = []
for i, c in enumerate(mapping):
if len(c) == 0:
new = new_comm[i]
out.append({
"type": "new",
"join": list(new),
})
elif len(c) == 1:
old = comm[i]
new = new_comm[i]
out.append({
"type": "old",
embedding = Model(seed, g.number_of_nodes(),
dim).deepwalk(g, deepwalk_epochs)
for scale in range(1000, 30000, 1000):
t0 = time.time()
comm_list = Model(seed, g.number_of_nodes(),
dim).ddcrp(g,
embedding,
ddcrp_scale=scale,
ddcrp_iterations=ddcrp_iterations)
ddcrp_time = time.time() - t0
comm_list = comm_list[ddcrp_cutoff:]
comm, _ = Model.mcla(comm_list)
predicted_cluster_size = len(comm)
modularity = nx.algorithms.community.quality.modularity(g, comm)
performance = nx.algorithms.community.quality.performance(g, comm)
improved_comm = Model.kmeans(embedding, comm)
improved_modularity = nx.algorithms.community.quality.modularity(
g, improved_comm)
improved_performance = nx.algorithms.community.quality.performance(
g, improved_comm)
naive_comm = Model.kmeans(embedding, len(comm))
naive_modularity = nx.algorithms.community.quality.modularity(
g, naive_comm)
naive_performance = nx.algorithms.community.quality.performance(
g, naive_comm)
write_line(graph_size, average_degree, cluster_size,
max_modularity, max_performance, scale,
predicted_cluster_size, modularity, performance,
improved_modularity, improved_performance,
naive_modularity, naive_performance, ddcrp_time)