def launch_tests():
for name in networks:
network = network_from_file("graphs//" + name)[0]
for b in examples.botnets(network, 0.9, nb_trials):
tests.hyper_parameter_influence(b, network, nb_trials,
hyper_parameter, values,
redundancy)
def launch_tests():
for name in networks:
print(name)
network = network_from_file("graphs//" + name)[0]
results = []
for b in examples.botnets(network, 0.9, nb_trials):
print(b.type)
real = []
exp = []
policy = []
action = []
perf = []
time = []
for i in range(redundancy):
print("Trial ", i)
real_rewards, expected_rewards, policy_rewards = tests.train(
b, network, nb_trials)
real.append(real_rewards)
exp.append(expected_rewards)
policy.append(policy_rewards)
action.append(b.compute_policy())
p = Policy(network, b.compute_policy())
perf.append(p.expected_reward(b.gamma))
time.append(p.expected_time())
b.clear(all=True)
results.append((b.type, real, exp, policy, action, perf, time))
tests.dump_actions("botnet_Compare", name, "all", results, nb_trials)
def show_results(nb):
window_size = int(nb / 20)
for name in networks:
res = tests.load_file("results/botnet_Compare_%s_all_%d.out" %
(name, nb))
network = network_from_file("graphs//" + name)[0]
real_res = []
exp_res = []
pol_res = []
final_perf = []
for x in res:
(b, real_rewards, expected_rewards, policy_rewards, actions, perf,
time) = x
real_rewards = np.array(real_rewards)
expected_rewards = np.array(expected_rewards)
policy_rewards = np.array(policy_rewards)
perf = np.array(perf)
time = np.array(time)
real_rewards = np.mean(real_rewards, axis=0)
expected_rewards = np.mean(expected_rewards, axis=0)
policy_rewards = np.mean(policy_rewards, axis=0)
real_res.append((b, real_rewards))
exp_res.append((b, expected_rewards))
pol_res.append((b, policy_rewards))
final_perf.append((b, np.mean(perf), np.max(perf), np.mean(time),
np.min(time), actions))
subplot(2, 2, 1)
for (name, perf) in real_res:
tests.plot_perf(perf, window_size, name)
ylabel("Real rewards")
#f = figure(2)
axis(sharex=True)
subplot(2, 2, 3)
for (name, perf) in exp_res:
tests.plot_perf(perf, window_size, name)
ylabel("Expected rewards")
#f = figure(3)
axis(sharex=True)
subplot(1, 2, 2)
for (name, perf) in pol_res:
tests.plot_perf(perf, window_size, name)
ylabel("Policy rewards")
tests.show_with_legend(f)
for (b_name, mean_p, max_p, mean_t, min_t, actions) in final_perf:
print(b_name,
"Mean perf:",
mean_p,
"Max perf:",
max_p,
"Mean time:",
mean_t,
"Min time:",
min_t,
"Actions:",
actions,
sep='\t')
x = input("Enter something to exit :")
def show_results():
for name in networks:
network = network_from_file("graphs//" + name)[0]
b = "ThompsonSampling"
[actions, times, rewards, values] = tests.load_file(
"results/beta_%d_%s_%d.out" % (network.size, b, nb_trials))
tests.plot_with_legend(values, times, "Time")
tests.plot_with_legend(values, rewards, "Reward")
tests.show_with_legend()
def show_results():
botnets_names = [
"RewardTentative", "Sarsa", "QLearning", "QLearning - Potential",
"ThompsonSampling", "ModelBasedThompson", "FullModelBasedThompson"
]
for name in networks:
network = network_from_file("graphs//" + name)[0]
print(name)
for b in botnets_names:
print(b)
[actions, times, rewards, values] = tests.load_file(
"results/gamma_%d_%s_%d.out" % (network.size, b, nb_trials))
tests.plot_with_legend(values, times, b)
tests.show_with_legend()
def show_results():
for name in networks:
network = network_from_file("graphs//" + name)[0]
b = FullModelBasedThompson(thompson_standard,
network.graph,
gamma=0.9,
nb_trials=nb_trials,
initial_nodes=network.initial_nodes)
[actions, times, rewards, values] = tests.load_file(
"results/alpha_p_%d_%s_%d.out" % (network.size, b.type, nb_trials))
values = [math.log(v, 10) for v in values]
tests.plot_with_legend(values, times, "Time")
tests.plot_with_legend(values, rewards, "Reward")
tests.show_with_legend()
def launch_tests():
for name in networks:
network = network_from_file("graphs//" + name)[0]
botnet = Thompson(thompson_standard,
network.graph,
gamma=0.9,
nb_trials=nb_trials,
initial_nodes=network.initial_nodes)
tests.hyper_parameter_influence(botnet,
network,
nb_trials,
hyper_parameter,
values,
redundancy,
is_log=False)
def show_results():
botnets_names = [
"Sarsa", "QLearning", "QLearning - Potential", "ThompsonSampling"
]
for name in networks:
network = network_from_file("graphs//" + name)[0]
print(name)
for b in botnets_names:
print(b)
[actions, times, rewards, values] = tests.load_file(
"results/alpha_%d_%s_%d.out" % (network.size, b, nb_trials))
values = [math.log(v, 10) for v in values]
tests.plot_with_legend(values, times, "Time")
tests.plot_with_legend(values, rewards, "Reward")
tests.show_with_legend()
def test(file_name):
net = network.network_from_file(file_name)
new_game = game.Game()
score = new_game.play(net)
print("Evaluated score: {0}".format(score))
import strategy
import thompson_sampling as thom
import time
from qlearning import QLearning
#n = 16
#difficulty = 2
##
##net = network.Network(1)
##
##for i in range(n):
## net.add_node(i**1.5, i, i)
##
##net.generate_random_connected()
#net = network.random_network(n, difficulty, big_nodes=log(n)/float(n), complete=False)
net, _ = network.network_from_file("./graphs/W08atk.gr")
q = thom.Thompson(strategy.thompson_standard, net.graph, 0.9, 0.01)
#q = QLearning(strategy.full_exploration, net.graph, nb_trials=200)
tests.train(q, net, 100)
q.clear()
l = tests.sample_optimal(q, net)
net.viz.layout()
#fig = plt.figure()
#ax = fig.add_subplot(111)
#plt.ion()
#plt.show()
#