-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathQuadraKill.py
189 lines (151 loc) · 6.46 KB
/
QuadraKill.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
177
178
179
180
181
182
183
184
185
186
187
188
189
import os
import sys
from Environment_V2 import environment
import torch
import numpy as np
import torch
import torch.optim as optim
from torch.autograd import Variable
import matplotlib.pyplot as plt
import torch.nn.functional as F
import random
from Agent.ac import Actor, Critic
actor = Actor(A_DIM=8).cuda()
critic = Critic().cuda()
a_opt = optim.Adam(actor.parameters(), lr=0.001)
c_opt = optim.Adam(critic.parameters(), lr=0.001)
ENV = environment.env([21, 14], [45, 87], 999, plot=False)
action_dic = ['up', 'upright', 'right', 'rightdown', 'down', 'downleft', 'left', 'leftup']
saved_dict = "saved_model"
saved_fig = "saved_figure"
GAMMA = 0.99
TAU = 1.0
EnCOEF = 0.01
max_times = 100
if __name__ == '__main__':
actor.train()
critic.train()
value_point = ENV.data_base.value_point
length = len(value_point)
episode = 0
np.random.seed(1)
p_loss = []
v_loss = []
reward_record = []
p_loss_tmp = []
v_loss_tmp = []
reward_record_tmp = []
while True:
a = np.random.randint(0, length)
b = np.random.randint(0, length)
Time = np.random.randint(0, 10000)
if a==b:
print('Zhao equals SillyB')
continue
s, valid_action = ENV.reset(start_loc=value_point[a], target=value_point[b], time=10)
if np.sum(valid_action) == 0:
print("shibai")
continue
a_cx = Variable(torch.zeros(1, 256)).cuda()
a_hx = Variable(torch.zeros(1, 256)).cuda()
c_cx = Variable(torch.zeros(1, 256)).cuda()
c_hx = Variable(torch.zeros(1, 256)).cuda()
all_rewards = []
all_values = []
all_entropies = []
all_lprobs = []
for step in range(max_times):
s = Variable(torch.from_numpy(np.array(s))).view(1, 3, 100, 100).float().cuda()
value, (c_hx, c_cx) = critic((s, (c_hx, c_cx)))
probs, (a_hx, a_cx) = actor((s, (a_hx, a_cx)))
mask = Variable(torch.from_numpy(valid_action)).cuda().view(1, 8)
masked_probs = probs * mask
action = masked_probs.multinomial(1)
lporbs = torch.log(probs)
log_prob = lporbs.gather(1, action)
entropy = -(log_prob * probs).sum(1)
real_action = action_dic[int(action.cpu().data.numpy())]
s_, r, done, [_, _, valid_action], success = ENV.step(real_action) # True: Read terminal
if np.sum(valid_action) == 0: # used to deal with the environment's dirty data.
break
all_rewards.append(r)
all_values.append(value)
all_entropies.append(entropy)
all_lprobs.append(log_prob)
s = s_
# print('Episode: {} Step: {} Aciton: {}'.format(episode, step, real_action))
# if success:
# if not os.path.exists('/home/exx/Lab/SmartST/model_saved_rl'):
# os.mkdir('/home/exx/Lab/SmartST/model_saved_rl')
# torch.save(actor.state_dict(), '/home/exx/Lab/SmartST/model_saved_rl/' + 'suc_model_{:d}.pth'.format(episode))
if done or step == max_times-1:
episode += 1
break
R = Variable(torch.zeros(1, 1)).cuda()
if not done:
s = Variable(torch.from_numpy(np.array(s))).view(1, 3, 100, 100).float().cuda()
value, (_, _) = critic((s, (c_hx, c_cx)))
R = value
all_values.append(R)
policy_loss = 0
value_loss = 0
gae = Variable(torch.zeros(1, 1)).cuda()
for i in reversed(range(len(all_rewards))):
R = GAMMA * R + all_rewards[i]
advantage = R - all_values[i]
value_loss += 0.5 * advantage.pow(2)
a = all_values[i+1]
# Generalized Advantage Estimation
delta_t = np.float(all_rewards[i]) + GAMMA * all_values[i+1] - all_values[i]
gae = gae * GAMMA * TAU + delta_t
policy_loss = policy_loss - all_lprobs[i] * gae - EnCOEF * all_entropies[i]
actor.zero_grad()
policy_loss.backward(retain_graph=True)
a_opt.step()
critic.zero_grad()
value_loss.backward(retain_graph=True)
c_opt.step()
p_loss_tmp.append(policy_loss.data)
v_loss_tmp.append(value_loss.data)
reward_record_tmp.append(sum(all_rewards)/all_rewards.__len__())
if (episode+1) % 50 == 0:
tmp_v_loss = sum(v_loss_tmp)/v_loss_tmp.__len__()
tmp_p_loss = sum(p_loss_tmp) / p_loss_tmp.__len__()
tmp_reward_record = sum(reward_record_tmp)/reward_record_tmp.__len__()
print("Episode: {}, In last 50 episode, average value loss:{},"
"average policy loss:{},average reward:{}".format(episode, tmp_v_loss,tmp_p_loss,tmp_reward_record))
v_loss.append(tmp_v_loss)
p_loss.append(tmp_p_loss)
reward_record.append(tmp_reward_record)
v_loss_tmp = []
p_loss_tmp = []
reward_record_tmp = []
if (episode + 1) % 20000 == 0:
if not os.path.exists(saved_dict):
os.mkdir(saved_dict)
torch.save(actor.state_dict(), os.path.join(os.path.join(os.getcwd(), saved_dict), 'actor_model_{:d}.pth'.format(episode)))
torch.save(critic.state_dict(), os.path.join(os.path.join(os.getcwd(), saved_dict), 'critic_model_{:d}.pth'.format(episode)))
if not os.path.exists(saved_fig):
os.mkdir(saved_fig)
path = os.path.join(os.getcwd(), saved_fig)
plt.title("Policy loss for {} episode".format(episode))
plt.plot(p_loss)
plt.xlabel("every 50 episode")
plt.ylabel("every 50 episode average policy loss")
plt.savefig(os.path.join(path, "policy_loss_{}".format(episode)))
plt.close()
p_loss = []
plt.title("value loss for {} episode".format(episode))
plt.plot(v_loss)
plt.xlabel("every 50 episode")
plt.ylabel("every 50 episode average value loss")
plt.savefig(os.path.join(path, "value_loss_{}".format(episode)))
plt.close()
v_loss = []
plt.title("reward for {} episode".format(episode))
plt.plot(reward_record)
plt.xlabel("every 50 episode")
plt.ylabel("every 50 episode average reward")
plt.savefig(os.path.join(path, "reward_record_{}".format(episode)))
plt.close()
reward_record = []