-
Notifications
You must be signed in to change notification settings - Fork 1
/
eval_td3.py
176 lines (170 loc) · 6.39 KB
/
eval_td3.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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
# This file can generate two results:
# (a) Trajectories of AUVs and USV(fig 4)
# (b) Positioning error of the AUV
# you can use fig_draw_example/* to generate analysis
import math
import os
from env import Env
import numpy as np
import argparse
import copy
import pickle
# pytorch
from td3 import TD3
# args, same as train_ddpg.py, but some are dummy
parser = argparse.ArgumentParser()
# ------ training paras ------
parser.add_argument("--is_train", type=int, default=1)
parser.add_argument("--lr", type=float, default=0.001)
parser.add_argument("--gamma", type=float, default=0.97)
parser.add_argument("--tau", type=float, default=0.001)
parser.add_argument("--hidden_size", type=int, default=128)
parser.add_argument("--replay_capa", type=int, default=20000)
parser.add_argument("--batch_size", type=int, default=64)
parser.add_argument("--policy_freq", type=int, default=2)
parser.add_argument("--repeat_num", type=int, default=10)
parser.add_argument(
"--episode_length", type=int, default=1000, help="the length of an episode (sec)"
)
parser.add_argument("--save_model_freq", type=int, default=25)
parser.add_argument(
"--load_ep",
type=int,
default=575,
help="Load model ep. Make sure this number is divisible by save_model_freq",
)
# ------ env paras ------
parser.add_argument(
"--R_dc",
type=float,
default=6.0,
metavar="R_DC",
help="the radius of data collection",
)
parser.add_argument("--border_x", type=float, default=200.0, help="Area x size")
parser.add_argument("--border_y", type=float, default=200.0, help="Area y size")
parser.add_argument("--n_s", type=int, default=30, help="The number of SNs")
parser.add_argument("--N_AUV", type=int, default=2, help="The number of AUVs")
parser.add_argument("--Q", type=float, default=2, help="Capacity of SNs (Mbits)")
parser.add_argument(
"--alpha", type=float, default=0.05, help="SNs choosing distance priority"
)
args = parser.parse_args()
BASE_PATH = os.path.dirname(os.path.abspath(__file__))
SAVE_PATH = BASE_PATH + "/models_ddpg/"
RES_PATH = BASE_PATH + "/results"
if not os.path.exists(RES_PATH):
os.makedirs(RES_PATH)
def eval():
for ep in range(args.repeat_num):
state_c = env.reset()
state = copy.deepcopy(state_c)
# record trajs
x_auv = [[env.xy[i][0]] for i in range(N_AUV)]
y_auv = [[env.xy[i][1]] for i in range(N_AUV)]
x_usv = [env.usv_xy[0]]
y_usv = [env.usv_xy[1]]
tracking_error = [
[np.linalg.norm(env.obs_xy[i] - env.xy[i])] for i in range(N_AUV)
]
# hovering point
sx = [[] for i in range(N_AUV)]
sy = [[] for i in range(N_AUV)]
ep_r = 0
idu = 0
N_DO = 0
DQ = 0
FX = [0] * N_AUV
sum_rate = 0
Ec = [0] * N_AUV
Ht = [0] * N_AUV
Ft = 0
crash = 0
mode = [0] * N_AUV
ht = [0] * N_AUV
hovers = [False] * N_AUV # flags
ep_reward = 0
while True:
act = []
for i in range(N_AUV):
iact = agents[i].select_action(state[i])
act.append(iact)
env.posit_change(act, hovers)
state_, rewards, Done, data_rate, ec, cs = env.step_move(hovers)
# add_posits
crash += cs
ep_reward += np.sum(rewards) / 1000
for i in range(N_AUV):
# append positions to traj
x_auv[i].append(env.xy[i][0])
y_auv[i].append(env.xy[i][1])
x_usv.append(env.usv_xy[0])
y_usv.append(env.usv_xy[1])
# add the tracking error
tracking_error[i].append(np.linalg.norm(env.obs_xy[i] - env.xy[i]))
if mode[i] == 0:
state[i] = copy.deepcopy(state_[i])
if Done[i] == True: # SN serving
idu += 1
ht[i] = args.Q * env.updata[i] / data_rate[i]
mode[i] += math.ceil(ht[i])
# add the hovering point
sx[i].append(env.xy[i][0])
sy[i].append(env.xy[i][1])
hovers[i] = True
sum_rate += data_rate[i]
else:
mode[i] -= 1
Ht[i] += 1
if mode[i] == 0:
hovers[i] = False
Ht[i] -= math.ceil(ht[i]) - ht[i]
state[i] = env.CHOOSE_AIM(idx=i, lamda=args.alpha)
Ft += 1
env.Ft = Ft
N_DO += env.N_DO
FX = np.array(FX) + np.array(env.FX)
DQ += sum(env.b_S / env.Fully_buffer)
Ec = np.array(Ec) + np.array(ec)
if Ft > args.episode_length:
N_DO /= Ft
DQ /= Ft
DQ /= env.N_POI
Ec = np.sum(np.array(Ec) / (Ft - np.array(Ht))) / N_AUV
print(
"EP:{:.0f} | ep_r {:.0f} | L_data {:.2f} | sum_rate {:.2f} | idu {:.2f} | ec {:.2f} | N_D {:.0f} | CS {} | FX {}".format(
ep, ep_reward, DQ, sum_rate, idu, Ec, N_DO, crash, FX
)
)
# save the file, and set the number index
traj_start_idx = len(
[f for f in os.listdir(RES_PATH) if f.lower().count("traj") != 0]
)
terror_start_idx = len(
[
f
for f in os.listdir(RES_PATH)
if f.lower().count("tracking_error") != 0
]
)
with open(f"{RES_PATH}/traj_{traj_start_idx}.pkl", "wb") as f:
pickle.dump(
[x_auv, y_auv, x_usv, y_usv, sx, sy, env.SoPcenter, env.lda], f
)
with open(
f"{RES_PATH}/tracking_error_{terror_start_idx}.pkl", "wb"
) as f:
pickle.dump(
[tracking_error, x_auv, y_auv, [env.X_max, env.Y_max, env.H]], f
)
break
if __name__ == "__main__":
env = Env(args)
N_AUV = args.N_AUV
state_dim = env.state_dim
action_dim = 2
agents = [TD3(state_dim, action_dim) for _ in range(N_AUV)]
# load models
for i in range(N_AUV):
agents[i].load(SAVE_PATH, args.load_ep, idx=i)
eval()