forked from Limmen/idsgame
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathreinforce.py
153 lines (128 loc) · 9.33 KB
/
reinforce.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
import os
import gym
import sys
from gym_idsgame.agents.training_agents.policy_gradient.pg_agent_config import PolicyGradientAgentConfig
from gym_idsgame.agents.training_agents.policy_gradient.reinforce.reinforce import ReinforceAgent
from util import util
import argparse
import json
parser = argparse.ArgumentParser(description='')
parser.add_argument('--experiment_id', type=str, default="REINF.0.MinDef19.1000.00.123456789")
parser.add_argument('--seed', type=int, default=123456789)
parser.add_argument('--batchsize', type=int, default=8)
parser.add_argument('--lr_decay_rate', type=float, default=0.999)
parser.add_argument('--env_name', type=str, default="idsgame-minimal_defense-v19")
parser.add_argument('--input_dim_attacker', type=int, default=40)
parser.add_argument('--output_dim_attacker', type=int, default=44)
#parser.add_argument('--input_dim_defender', type=int, default=44)
#parser.add_argument('--output_dim_defender', type=int, default=44)
parser.add_argument('--alpha_attacker', type=float, default=0.0001)
parser.add_argument('--alpha_defender', type=float, default=0.0001)
parser.add_argument('--eval_episodes', type=int, default=100)
parser.add_argument('--num_episodes', type=int, default=100001)
parser.add_argument('--eval_frequency', type=int, default=1000)
parser.add_argument('--train_log_frequency', type=int, default=100)
parser.add_argument('--eval_log_frequency', type=int, default=1000)
parser.add_argument('--hidden_dim', type=int, default=128)
parser.add_argument('--num_hidden_layers', type=int, default=2)
parser.add_argument('--discount_factor', type=int, default=0.999)
parser.add_argument('--defender', type=bool, default=False)
parser.add_argument('--attacker', type=bool, default=True)
parser.add_argument('--lr_exp_decay', type=bool, default=False)
parser.add_argument('--gpu', type=bool, default=False)
args = parser.parse_args()
def get_script_path():
"""
:return: the script path
"""
return os.path.dirname(os.path.realpath(sys.argv[0]))
def default_output_dir() -> str:
"""
:return: the default output dir
"""
script_dir = get_script_path()
return script_dir
# Program entrypoint
if __name__ == '__main__':
random_seed = args.seed
with open(default_output_dir() + "/results/" + args.experiment_id + "_args.txt", 'w') as f:
json.dump(args.__dict__, f, indent=2)
util.create_artefact_dirs(default_output_dir(), args.experiment_id)
# these parameter are changed
'''
pg_agent_config = PolicyGradientAgentConfig(gamma=args.discount_factor,
alpha_attacker=args.alpha_attacker,
alpha_defender=args.alpha_defender,
epsilon=1, render=False, eval_epsilon=0.,
eval_sleep=0.9,
min_epsilon=0.01, eval_episodes=args.eval_episodes, train_log_frequency=args.train_log_frequency,
epsilon_decay=0., video=False, eval_log_frequency=args.eval_log_frequency,
video_fps=5, video_dir=default_output_dir() + "/results/videos/" + str(random_seed),
num_episodes=args.num_episodes,
eval_render=False, gifs=True,
gif_dir=default_output_dir() + "/results/gifs/" + str(random_seed),
eval_frequency=args.eval_frequency,
attacker=args.attacker,
defender=args.defender,
video_frequency=101,
save_dir=default_output_dir() + "/results/data/" + str(random_seed),
checkpoint_freq=5000,
input_dim_attacker=args.input_dim_attacker,
output_dim_attacker=args.output_dim_attacker,
#input_dim_defender=args.input_dim_defender,
#output_dim_defender=args.output_dim_defender,
hidden_dim=args.hidden_dim,
num_hidden_layers=args.num_hidden_layers, batch_size=args.batchsize,
gpu=args.gpu, tensorboard=True,
tensorboard_dir=default_output_dir() + "/results/tensorboard/" + str(random_seed),
optimizer="Adam", lr_exp_decay=args.lr_exp_decay, lr_decay_rate=args.lr_decay_rate)
'''
pg_agent_config = PolicyGradientAgentConfig(gamma=args.discount_factor, alpha_attacker=args.alpha_attacker, epsilon=1, render=False,
alpha_defender=args.alpha_defender,
eval_sleep=0.9,
min_epsilon=0.01, eval_episodes=args.eval_episodes, train_log_frequency=args.train_log_frequency,
epsilon_decay=0.9999, video=False, eval_log_frequency=args.eval_log_frequency,
video_fps=5, video_dir=default_output_dir() + "/results/videos",
num_episodes=args.num_episodes,
eval_render=False, gifs=False,
gif_dir=default_output_dir() + "/results/gifs/" + args.experiment_id,
eval_frequency=args.eval_frequency, attacker=args.attacker, defender=args.defender,
video_frequency=1001,
save_dir=default_output_dir() + "/results/data/" + args.experiment_id,
checkpoint_freq=500,
input_dim_attacker=((4 + 2) * 4),
output_dim_attacker=(4 + 1) * 4,
input_dim_defender=((4 + 1) * 4),
output_dim_defender=5 * 4,
hidden_dim=args.hidden_dim, num_hidden_layers=args.num_hidden_layers,
pi_hidden_layers=1, pi_hidden_dim=128, vf_hidden_layers=1,
vf_hidden_dim=128,
batch_size=args.batchsize,
gpu=args.gpu, tensorboard=True,
tensorboard_dir=default_output_dir() + "/results/tensorboard/" + args.experiment_id,
optimizer="Adam", lr_exp_decay=args.lr_exp_decay, lr_decay_rate=args.lr_decay_rate,
state_length=1, normalize_features=False, merged_ad_features=True,
zero_mean_features=False, gpu_id=0,
lstm_network=False,
lstm_seq_length=4, num_lstm_layers=2, optimization_iterations=10,
eps_clip=0.2, max_gradient_norm=0.5, gae_lambda=0.95,
cnn_feature_extractor=False, features_dim=512,
flatten_feature_planes=False, cnn_type=5, vf_coef=0.5, ent_coef=0.001,
render_attacker_view=False, lr_progress_power_decay=4,
lr_progress_decay=False, use_sde=False, sde_sample_freq=4,
one_hot_obs=False, lstm_core=False, lstm_hidden_dim=32,
multi_channel_obs=False,
channel_1_dim=32, channel_1_layers=2, channel_1_input_dim=16,
channel_2_dim=32, channel_2_layers=2, channel_2_input_dim=16,
channel_3_dim=32, channel_3_layers=2, channel_3_input_dim=4,
channel_4_dim=32, channel_4_layers=2, channel_4_input_dim=4,
mini_batch_size=64, ar_policy=False,
attacker_node_input_dim=((4 + 2) * 4),
attacker_at_net_input_dim=(4 + 2), attacker_at_net_output_dim=(4 + 1),
attacker_node_net_output_dim=4)
env_name = args.env_name
env = gym.make(env_name, save_dir=default_output_dir() + "/results/data/" + args.experiment_id)
attacker_agent = ReinforceAgent(env=env, config=pg_agent_config)
attacker_agent.train()
train_result = attacker_agent.train_result
eval_result = attacker_agent.eval_result