def test_bruteforce_gen(scenario, seed, fully_obs, flat_actions, flat_obs):
"""Tests all generated benchmark scenarios using every possible environment
setting, using bruteforce agent, checking for any errors
"""
env = nasim.make_benchmark(scenario,
seed=seed,
fully_obs=fully_obs,
flat_actions=flat_actions,
flat_obs=flat_obs)
run_bruteforce_agent(env, verbose=False)
def run_scenario(args):
scenario_name, seed = args
print_msg(f"Running '{scenario_name}' scenario with seed={seed}")
env = nasim.make_benchmark(scenario_name, seed, False, True, True)
steps, total_reward, done = run_random_agent(env, verbose=False)
return {
"Name": scenario_name,
"Seed": seed,
"Steps": steps,
"Total reward": total_reward
}
parser.add_argument("--replay_size", type=int, default=500000,
help="(default=100000)")
parser.add_argument("--final_epsilon", type=float, default=0.05,
help="(default=0.05)")
parser.add_argument("--init_epsilon", type=float, default=1.0,
help="(default=1.0)")
parser.add_argument("--exploration_steps", type=int, default=800000,
help="(default=10000)")
parser.add_argument("--gamma", type=float, default=0.9,
help="(default=0.99)")
parser.add_argument("--quite", action="store_false",
help="Run in Quite mode")
args = parser.parse_args()
env = nasim.make_benchmark(args.env_name,
args.seed,
fully_obs=not args.partially_obs,
flat_actions=True,
flat_obs=True)
#env.render_network_graph(show=True)
dqn_agent = NoisyDoubleDuelingDQN_PERAgent(env, verbose=args.quite, **vars(args))
num_actions = env.action_space.n
#for a in range(num_actions):
# print(env.action_space.get_action(a))
dqn_agent.train()
dqn_agent.save("D:\\Experiments\\Experiment_Record\\E0812\\Saved_model\\NDSPI-medium-multi-site-0820.pkl")
#dqn_agent.save("D:\\Experiments\\NetworkAttackSimulator\\medium-multi-site-honeypot\\NoisyDoubleDueling_PER_Mar15.pkl")
#save_data(dqn_agent.rewards_episode,'D:\\Experiments\\NetworkAttackSimulator\\medium-multi-site\\ NoisyDoubleDueling_PER_rewards_episode_Mar1.csv')
#save_data(dqn_agent.rewards_step,'D:\\Experiments\\NetworkAttackSimulator\\medium-multi-site\\ NoisyDoubleDueling_PER_rewards_step_Mar1.csv')
#dqn_agent.run_eval_episode(render=args.render_eval)
parser.add_argument("-p",
"--param_actions",
action="store_true",
help="Use Parameterised action space")
parser.add_argument("-f",
"--box_obs",
action="store_true",
help="Use 2D observation space")
args = parser.parse_args()
seed = args.seed
run_steps = []
run_rewards = []
run_goals = 0
for i in range(args.runs):
env = nasim.make_benchmark(args.env_name, seed, not args.partially_obs,
not args.param_actions, not args.box_obs)
steps, reward, done = run_random_agent(env, verbose=False)
run_steps.append(steps)
run_rewards.append(reward)
run_goals += int(done)
seed += 1
if args.runs > 1:
print(f"Run {i}:")
print(f"\tSteps = {steps}")
print(f"\tReward = {reward}")
print(f"\tGoal reached = {done}")
run_steps = np.array(run_steps)
run_rewards = np.array(run_rewards)
"""Environment network graph visualizer
This script allows the user to visualize the network graph for a chosen
benchmark scenario.
"""
import nasim
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser()
parser.add_argument("scenario_name",
type=str,
help="benchmark scenario name")
parser.add_argument("-s",
"--seed",
type=int,
default=0,
help="random seed (default=0)")
args = parser.parse_args()
env = nasim.make_benchmark(args.scenario_name, args.seed)
env.render_network_graph(show=True)
def test_generator(scenario, seed):
"""Tests generating all generated benchmark scenarios using a range of
seeds, checking for any errors
"""
nasim.make_benchmark(scenario, seed=seed)
" AI hacker."))
parser.add_argument("env_name", type=str, help="benchmark scenario name")
parser.add_argument("-ai",
"--run_ai",
action="store_true",
help=("Run AI policy (currently ony supported for"
" 'tiny' and 'small' environments"))
args = parser.parse_args()
if args.run_ai:
assert args.env_name in DQN_POLICIES, \
("AI demo only supported for the following environments:"
f" {list(DQN_POLICIES)}")
env = nasim.make_benchmark(args.env_name,
fully_obs=True,
flat_actions=True,
flat_obs=True)
line_break = f"\n{'-'*60}"
print(line_break)
print(f"Running Demo on {args.env_name} environment")
if args.run_ai:
print("Using AI policy")
print(line_break)
dqn_agent = DQNAgent(env, verbose=False, **vars(args))
dqn_agent.load(DQN_POLICIES[args.env_name])
ret, steps, goal = dqn_agent.run_eval_episode(env, True, 0.01,
"readable")
else:
print("Player controlled")
print(line_break)
