def get_node_ns(graph, k=3):
"""
Get k nodes to attack based on the Netshield algorithm :cite`tong2010vulnerability`.
:param graph: an undirected NetworkX graph
:param k: number of nodes to attack
:return: a list of nodes to attack
"""
if not scipy.sparse.issparse(graph):
sparse_graph = get_sparse_graph(graph)
else:
sparse_graph = graph
lam, u = eigsh(sparse_graph, k=1, which='LA')
lam = lam[0]
u = np.abs(np.real(u).flatten())
v = (2 * lam * np.ones(len(u))) * np.power(u, 2)
nodes = []
for i in range(k):
B = sparse_graph[:, nodes]
b = B * u[nodes]
score = v - 2 * b * u
score[nodes] = -1
nodes.append(np.argmax(score))
return nodes
def __init__(self, graph, runs, steps, **kwargs):
# TODO: efficiency improvement--store edge and node difference rather than whole graph twice
self.graph_og = graph.copy()
self.graph = graph
self.prm = {
'runs': runs,
'steps': steps,
'seed': 1,
'max_val': 1,
'gif_animation': False,
'gif_snaps': False,
'plot_transition': False,
'edge_style': None,
'node_style': None,
'fa_iter': 200
}
self.sim_info = defaultdict()
self.sparse_graph = get_sparse_graph(self.graph)
if self.prm['seed'] is not None:
random.seed(self.prm['seed'])
np.random.seed(self.prm['seed'])