-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathmain.py
156 lines (136 loc) · 6.91 KB
/
main.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
154
155
156
import supersuit
import wandb
import torch as th
from wandb.integration.sb3 import WandbCallback
import os
from stable_baselines3 import PPO, A2C, TD3, SAC, DQN, HerReplayBuffer, HER, DDPG
from stable_baselines3.common.evaluation import evaluate_policy
from stable_baselines3.common.env_util import make_vec_env
from sb3_contrib import TQC, QRDQN, MaskablePPO, TRPO, ARS, RecurrentPPO
# pip install git+https://github.com/Stable-Baselines-Team/stable-baselines3-contrib
from uav_env import parallel_env
from result_buffer import ResultBuffer
from supersuit import frame_stack_v1, normalize_obs_v0
from input_config import InputConfig
os.environ["WANDB_SILENT"] = "True"
project_name = "multi-agent-uav-offloading"
print("starting {} project".format(project_name))
change_processing = True
eval_episodes = 250
p = [1.75 + k * 0.05 for k in range(11)]
o = [0.5 + k * 1 for k in range(11)]
min_mu = p[0] if change_processing else o[0]
max_mu = p[-1] if change_processing else o[-1]
step_mu = 0.05 if change_processing else 1
print("processing rates:", p)
algo = PPO
n_uavs = 8
algorithms = ["fcto", "woto", "MULTIAGENT", "ldo", "us"]
for alg in algorithms:
res_buffer = ResultBuffer(min_n_drone=n_uavs, max_n_drone=n_uavs, min_mu=min_mu, max_mu=max_mu, step_mu=step_mu,
net_slice=1, change_processing=change_processing, alg=alg)
for mu_p in p:
config = {"algo": algo if alg == "MULTIAGENT" else alg,
"n_cpus": 10,
"uavs": n_uavs,
"frame_stack": 4,
"processing_rate": mu_p,
"offloading_rate": 1, # 2.5
"transition_probability_low": 1 / 180,
"transition_probability_high": 1 / 60,
"shifting_probs": True, # True gives better learning curve, while False gives nice results quicker
"lambda_low": 1.3,
"lambda_high": 2.6,
"policy_type": "MlpPolicy",
"total_timesteps": 500000 # 1000000
}
input_config = InputConfig(uavs=config["uavs"],
frame_stack=config["frame_stack"],
processing_rate=config["processing_rate"],
offloading_rate=config["offloading_rate"],
lmbda=[config["lambda_low"], config["lambda_high"]],
prob_trans=[config["transition_probability_low"], config["transition_probability_high"]],
shifting_probs=config["shifting_probs"],
algorithm=alg,
)
input_config.print_settings()
uav_env = parallel_env(input_c=input_config)
uav_env = supersuit.pettingzoo_env_to_vec_env_v1(uav_env)
uav_env = supersuit.concat_vec_envs_v1(uav_env, 1, num_cpus=config["n_cpus"],
base_class='stable_baselines3')
if change_processing:
res_buffer.set_save_runs(n_drones=config["uavs"], mu=config["processing_rate"])
else:
res_buffer.set_save_runs(n_drones=config["uavs"], mu=config["offloading_rate"])
eval_env = parallel_env(input_c=input_config, result_buffer=res_buffer)
eval_env = supersuit.pettingzoo_env_to_vec_env_v1(eval_env)
eval_env = supersuit.concat_vec_envs_v1(eval_env, 1, num_cpus=config["n_cpus"],
base_class='stable_baselines3')
# uav_env = supersuit.normalize_obs_v0(uav_env, env_min=0, env_max=1)
# uav_env = supersuit.frame_stack_v1(uav_env, 3)
"""
model = PPO('MlpPolicy', uav_env, verbose=0, gamma=0.95, n_steps=256, ent_coef=0.0905168, learning_rate=0.00062211,
vf_coef=0.042202, max_grad_norm=0.9, gae_lambda=0.99, n_epochs=5, clip_range=0.3, batch_size=256)
"""
# need to initialize it even for static policies to use evaluate_policy from sb3
model = algo(config["policy_type"], uav_env, verbose=0, gamma=0.95, tensorboard_log=f"runs")
if alg == "MULTIAGENT":
# for i in range(3):
run = wandb.init(
project=project_name,
tags=["n {}".format(config["uavs"]),
"pr {}".format(config["processing_rate"]),
"or {}".format(config["offloading_rate"]),
"lmbda_l {:.2f}, lmbda_h {:.2f}".format(config["lambda_low"], config["lambda_high"]),
"prob_l {:.2f}, prob_h {:.2f}".format(config["transition_probability_low"],
config["transition_probability_high"]),
"alg {}".format(alg),
],
entity="xraulz",
reinit=True,
sync_tensorboard=True,
config=config,
)
if os.name == 'nt':
model.learn(
total_timesteps=config["total_timesteps"],
)
else:
model.learn(
total_timesteps=config["total_timesteps"],
callback=WandbCallback(
gradient_save_freq=100,
model_save_path=f"models/{project_name}/{run.name}",
verbose=2,
),
)
# model is already saved via model_save_path inside model.learn
# model.save("policy {}-{}".format(run.name, i))
run.finish()
print("initializing evaluation...")
run = wandb.init(
project=project_name,
tags=["n {}".format(config["uavs"]),
"pr {}".format(config["processing_rate"]),
"or {}".format(config["offloading_rate"]),
"lmbda_l {:.2f}, lmbda_h {:.2f}".format(config["lambda_low"], config["lambda_high"]),
"prob_l {:.2f}, prob_h {:.2f}".format(config["transition_probability_low"],
config["transition_probability_high"]),
"evaluation",
"alg {}".format(alg),
],
entity="xraulz",
reinit=True,
sync_tensorboard=True,
config=config,
)
evaluate_policy(model, eval_env, n_eval_episodes=eval_episodes, deterministic=True)
run.finish()
"""
# not used anymore
print("saving results...")
if change_processing:
res_buffer.save_and_reset(n_drones=config["uavs"], mu=config["processing_rate"])
else:
res_buffer.save_and_reset(n_drones=config["uavs"], mu=config["offloading_rate"])
"""