def optimal_attack_full_network():
# when there is no limit on the type of attacks,
# a very good type of attack is obtained
# when they are all routes to the same station
# we proceed so by choosing the best station to route the attacks to
# which is the initialization, then there is not much room for progress
# with the block-coordinate descent algorithm
nw = load_network('data/queueing_params.mat')
nw.balance()
nw.combine()
nw.budget = 20.
nw.optimal_attack(max_iters=3, alpha=10., beta=1., max_iters_attack_rate=5)
def optimal_attack_full_network():
# when there is no limit on the type of attacks,
# a very good type of attack is obtained
# when they are all routes to the same station
# we proceed so by choosing the best station to route the attacks to
# which is the initialization, then there is not much room for progress
# with the block-coordinate descent algorithm
nw = load_network('data/queueing_params.mat')
nw.balance()
nw.combine()
nw.budget = 20.
nw.optimal_attack(max_iters=3, alpha=10., beta=1., max_iters_attack_rate=5)
def network_simulation():
nw = load_network('data/queueing_params.mat')
target = get_availabilities(nw.station_names)
bal_rates, bal_routing = nw.balance()
nw.combine()
n = Network(nw.size, nw.rates, nw.travel_times, nw.routing, [20]* nw.size)
for i in range(100):
if i % 10 == 0:
print i
n.jump()
T()
def network_simulation():
nw = load_network('data/queueing_params.mat')
target = get_availabilities(nw.station_names)
bal_rates, bal_routing = nw.balance()
nw.combine()
n = Network(nw.size, nw.rates, nw.travel_times, nw.routing, [20] * nw.size)
for i in range(100):
if i % 10 == 0:
print i
n.jump()
T()
def cal_logo_experiment(adj):
nw = load_network('data/queueing_params.mat')
target = get_availabilities(nw.station_names)
bal_rates, bal_routing = nw.balance()
nw.combine()
res = []
for i in adj:
nw.update_adjacency(i)
att_rates, att_routing = nw.min_attack(target, full_adj=False)
T()
res.append(int(np.sum(att_rates)))
print 'Passenger Arrival Rate:', np.sum(nw.rates)
print 'Balance Cost: ', np.sum(bal_rates)
print 'Attack After Balance Cost (adjacency {}): {}'.format(adj, res)
return res
def cal_logo_experiment(adj):
nw = load_network('data/queueing_params.mat')
target = get_availabilities(nw.station_names)
bal_rates, bal_routing = nw.balance()
nw.combine()
res = []
for i in adj:
nw.update_adjacency(i)
att_rates, att_routing = nw.min_attack(target, full_adj=False)
T()
res.append(int(np.sum(att_rates)))
print 'Passenger Arrival Rate:', np.sum(nw.rates)
print 'Balance Cost: ', np.sum(bal_rates)
print 'Attack After Balance Cost (adjacency {}): {}'.format(adj, res)
return res
def optimal_attack_with_radius(r, save_to=None):
# try to compute the optimal attacks with different radii of adjacencies
nw = load_network('data/queueing_params.mat')
nw.set_weights_to_min_time_usage()
#nw.rates += np.ones(nw.size) * 100
nw.balance()
nw.combine()
nw.budget = 1000
nw.update_adjacency(r)
# k has been pre-processed and is given by best_single_destination_attack()
k = 302
nw.optimal_attack(max_iters=1, full_adj=False, alpha=10., beta=1., \
max_iters_attack_rate=3, k=k)
rates = nw.attack_rates / (nw.attack_rates + nw.rates)
T()
if save_to:
obj = {'rates': rates, 'routing': nw.attack_routing}
pickle.dump(obj, open(save_to, 'wb'))
def optimal_attack_with_radius(r, save_to=None):
# try to compute the optimal attacks with different radii of adjacencies
nw = load_network('data/queueing_params.mat')
nw.set_weights_to_min_time_usage()
#nw.rates += np.ones(nw.size) * 100
nw.balance()
nw.combine()
nw.budget = 1000
nw.update_adjacency(r)
# k has been pre-processed and is given by best_single_destination_attack()
k = 302
nw.optimal_attack(max_iters=1, full_adj=False, alpha=10., beta=1., \
max_iters_attack_rate=3, k=k)
rates = nw.attack_rates / (nw.attack_rates + nw.rates)
T()
if save_to:
obj = {'rates': rates, 'routing': nw.attack_routing}
pickle.dump(obj, open(save_to, 'wb'))