def frank_wolfe_on_I210():
'''
Frank-Wolfe on I210
'''
graph = np.loadtxt('data/I210_attack_net.csv', delimiter=',', skiprows=1)
demand = np.loadtxt('data/I210_od.csv', delimiter=',', skiprows=1)
demand[:,2] = 1. * demand[:,2] / 4000
# run solver
f = solver_3(graph, demand, max_iter=1000, q=50, display=1, stop=1e-2)
# display cost
for a,b in zip(cost(f, graph), f*4000): print a,b
# visualization
node = np.loadtxt('data/I210_node.csv', delimiter=',', skiprows=1)
# extract features: 'capacity', 'length', 'fftt'
feat = extract_features('data/I210_attack_Sketch_net.csv')
ratio = cost_ratio(f, graph)
# merge features with the cost ratios
features = np.zeros((feat.shape[0],4))
features[:,:3] = feat
features[:,3] = ratio
# join features with (lat1,lon1,lat2,lon2)
links = process_links(graph, node, features)
color = features[:,3] # we choose the costs
names = ['capacity', 'length', 'fftt', 'tt_over_fftt']
geojson_link(links, names, color)
def reduce_demand():
net, demand, node = load_LA()
# features = extract_features('data/LA_net.txt')
f = np.loadtxt('data/LA/LA_output_3.csv', delimiter=',', skiprows=0)
cr = cost_ratio(f, net)
for row in range(net.shape[0]):
if cr[row] >= 10.:
out = []
for i in range(demand.shape[0]):
if int(demand[i, 0]) == int(net[row, 1]):
out.append(demand[i, 2])
demand[i, 2] = demand[i, 2] / 10.
if len(out) > 0:
out = np.array(out)
print '\nratio:', cr[row]
print 'origin: {}\nflow: {}'.format(int(demand[i, 0]),
np.sum(out))
print np.sort(out).tolist()
for row in range(net.shape[0]):
if cr[row] >= 10.:
out = []
for i in range(demand.shape[0]):
if int(demand[i, 1]) == int(net[row, 2]):
out.append(demand[i, 2])
demand[i, 2] = demand[i, 2] / 10.
if len(out) > 0:
out = np.array(out)
print '\nratio:', cr[row]
print 'destination: {}\nflow: {}'.format(
int(demand[i, 0]), np.sum(out))
print np.sort(out).tolist()
def frank_wolfe_on_I210():
'''
Frank-Wolfe on I210
'''
graph = np.loadtxt('data/I210_attack_net.csv', delimiter=',', skiprows=1)
demand = np.loadtxt('data/I210_od.csv', delimiter=',', skiprows=1)
demand[:, 2] = 1. * demand[:, 2] / 4000
# run solver
f = solver_3(graph, demand, max_iter=1000, q=50, display=1, stop=1e-2)
# display cost
for a, b in zip(cost(f, graph), f * 4000):
print a, b
# visualization
node = np.loadtxt('data/I210_node.csv', delimiter=',', skiprows=1)
# extract features: 'capacity', 'length', 'fftt'
feat = extract_features('data/I210_attack_Sketch_net.csv')
ratio = cost_ratio(f, graph)
# merge features with the cost ratios
features = np.zeros((feat.shape[0], 4))
features[:, :3] = feat
features[:, 3] = ratio
# join features with (lat1,lon1,lat2,lon2)
links = process_links(graph, node, features)
color = features[:, 3] # we choose the costs
names = ['capacity', 'length', 'fftt', 'tt_over_fftt']
geojson_link(links, names, color)
def reduce_demand():
net, demand, node = load_LA()
features = extract_features('data/LA_net.txt')
f = np.loadtxt('data/LA/LA_output_3.csv', delimiter=',', skiprows=0)
cr = cost_ratio(f, net)
for row in range(net.shape[0]):
if cr[row] >= 10.:
out = []
for i in range(demand.shape[0]):
if int(demand[i,0]) == int(net[row,1]):
out.append(demand[i,2])
demand[i,2] = demand[i,2] / 10.
if len(out) > 0:
out = np.array(out)
print '\nratio:', cr[row]
print 'origin: {}\nflow: {}'.format(int(demand[i,0]), np.sum(out))
print np.sort(out).tolist()
for row in range(net.shape[0]):
if cr[row] >= 10.:
out = []
for i in range(demand.shape[0]):
if int(demand[i,1]) == int(net[row,2]):
out.append(demand[i,2])
demand[i,2] = demand[i,2] / 10.
if len(out) > 0:
out = np.array(out)
print '\nratio:', cr[row]
print 'destination: {}\nflow: {}'.format(int(demand[i,0]), np.sum(out))
print np.sort(out).tolist()
def check_LA_result():
net, demand, node, features = load_LA_2()
demand[:, 2] = demand[:, 2] / 4000.
f = np.loadtxt('data/LA/LA_output_4.csv', delimiter=',', skiprows=0)
costs = cost(f, net)
cr = cost_ratio(f, net)
print np.sort(cr)[-20:]
for row in range(net.shape[0]):
if cr[row] >= 10.:
print cr[row]
print net[row, :3], features[row, :]
L = all_or_nothing_assignment(costs, net, demand)
print costs.dot(L) / np.sum(demand[:, 2])
def check_LA_result():
net, demand, node, features = load_LA_2()
demand[:,2] = demand[:,2] / 4000.
f = np.loadtxt('data/LA/LA_output_4.csv', delimiter=',', skiprows=0)
costs = cost(f, net)
cr = cost_ratio(f, net)
print np.sort(cr)[-20:]
for row in range(net.shape[0]):
if cr[row] >= 10.:
print cr[row]
print net[row,:3], features[row,:]
L = all_or_nothing_assignment(costs, net, demand)
print costs.dot(L) / np.sum(demand[:,2])
def visualize_equilibrium_in_chicago():
"""
visualize costs in the network of Chicago
"""
net, demand, node, f = load_chicago()
flow = np.loadtxt("data/Chicago_results_2.csv", delimiter=",")
# flow = np.loadtxt('data/Chicago_results.csv', delimiter=',', skiprows=1)[:,2]
print average_cost(flow / 4000.0, net)
costs = cost_ratio(flow / 4000.0, net)
features = np.zeros((f.shape[0], 4))
features[:, :3] = f
features[:, 3] = costs
links = process_links(net, node, features)
color = features[:, 3] - 1.0 # we choose the costs
geojson_link(links, ["capacity", "length", "fftt", "tt_over_fftt"], color)
def visualize_equilibrium_in_chicago():
'''
visualize costs in the network of Chicago
'''
net, demand, node, f = load_chicago()
flow = np.loadtxt('data/Chicago_results_2.csv', delimiter=',')
# flow = np.loadtxt('data/Chicago_results.csv', delimiter=',', skiprows=1)[:,2]
print average_cost(flow/4000., net, demand)
costs = cost_ratio(flow/4000., net)
features = np.zeros((f.shape[0],4))
features[:,:3] = f
features[:,3] = costs
links = process_links(net, node, features)
color = features[:,3] - 1.0 # we choose the costs
geojson_link(links, ['capacity', 'length', 'fftt', 'tt_over_fftt'], color)
def chicago_tt_over_fftt():
"""
study the test_*.csv files generated by chicago_parametric_study()
visualize tt/fftt for each edge
"""
net, demand, node, geometry = load_chicago()
g_nr, small_capacity = add_cognitive_cost(net, geometry, 2000.0, 1000.0)
# f = np.loadtxt('data/test_0.csv', delimiter=',', skiprows=0)
alpha = 0.01
fs = np.loadtxt("data/test_{}.csv".format(int(100.0 * alpha)), delimiter=",", skiprows=0)
# f = np.loadtxt('data/test_100.csv', delimiter=',', skiprows=0)
f = np.sum(fs, axis=1)
costs = cost_ratio(f, net)
features = np.zeros((f.shape[0], 4))
features[:, :3] = geometry
features[:, 3] = costs
links = process_links(net, node, features)
color = (costs - 1.0) * 2.0 + 1.0
geojson_link(links, ["capacity", "length", "fftt", "tt_over_fftt"], color)
def chicago_tt_over_fftt():
'''
study the test_*.csv files generated by chicago_parametric_study()
visualize tt/fftt for each edge
'''
net, demand, node, geometry = load_chicago()
g_nr, small_capacity = add_cognitive_cost(net, geometry, 2000., 1000.)
# f = np.loadtxt('data/test_0.csv', delimiter=',', skiprows=0)
alpha = 0.01
fs = np.loadtxt('data/test_{}.csv'.format(int(100.*alpha)), delimiter=',', skiprows=0)
# f = np.loadtxt('data/test_100.csv', delimiter=',', skiprows=0)
f = np.sum(fs, axis=1)
costs = cost_ratio(f, net)
features = np.zeros((f.shape[0], 4))
features[:,:3] = geometry
features[:,3] = costs
links = process_links(net, node, features)
color = (costs - 1.0) * 2.0 + 1.0
geojson_link(links, ['capacity', 'length', 'fftt', 'tt_over_fftt'], color)
def test_cost_ratio(self):
net = np.loadtxt('data/braess_net.csv', delimiter=',', skiprows=1)
net[:,5] = np.array([2.]*5)
flow = np.array([0., 1., 2., 3., 4.])
result = np.array([1.0, 3.0, 1e8, 19, 1e8])
self.assertTrue(np.linalg.norm(result - cost_ratio(flow, net)) < 1e-8)
def check_results():
# check highest ration of tt / fftt (should be less than 5)
net, d, node, features = load_I210()
f = np.loadtxt('data/I210/test_0.csv', delimiter=',')
print np.max(cost_ratio(f, net))
def increase_capacity():
net, demand, node = load_LA()
f = np.loadtxt('data/LA/LA_output_3.csv', delimiter=',', skiprows=0)
cr = cost_ratio(f, net)
def check_results():
# check highest ration of tt / fftt (should be less than 5)
net, d, node, features = load_I210()
f = np.loadtxt('data/I210/test_0.csv', delimiter=',')
print np.max(cost_ratio(f, net))
def increase_capacity():
net, demand, node = load_LA()
f = np.loadtxt('data/LA/LA_output_3.csv', delimiter=',', skiprows=0)
cr = cost_ratio(f, net)
def test_cost_ratio(self):
net = np.loadtxt('data/braess_net.csv', delimiter=',', skiprows=1)
net[:, 5] = np.array([2.] * 5)
flow = np.array([0., 1., 2., 3., 4.])
result = np.array([1.0, 3.0, 1e8, 19, 1e8])
self.assertTrue(np.linalg.norm(result - cost_ratio(flow, net)) < 1e-8)
