-
Notifications
You must be signed in to change notification settings - Fork 0
/
reinforce_continuous.py
222 lines (176 loc) · 7.99 KB
/
reinforce_continuous.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
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
'''
@Author: Wenhao Ding
@Email: [email protected]
@Date: 2020-01-24 13:52:10
LastEditTime: 2022-04-10 22:03:33
@Description:
'''
import os
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torch.nn.utils as utils
from torch.autograd import Variable
import torch.nn.init as init
def CUDA(var):
return var.cuda() if torch.cuda.is_available() else var
def kaiming_init(m):
if isinstance(m, nn.Linear):
init.xavier_normal_(m.weight)
if m.bias is not None:
m.bias.data.fill_(0)
elif isinstance(m, nn.Conv2d) or isinstance(m, nn.ConvTranspose2d):
init.kaiming_normal_(m.weight)
if m.bias is not None:
m.bias.data.fill_(0)
elif isinstance(m, (nn.BatchNorm1d, nn.BatchNorm2d)):
m.weight.data.fill_(1) # ??
if m.bias is not None:
m.bias.data.fill_(0)
pi = CUDA(Variable(torch.FloatTensor([math.pi])))
def normal(x, mu, sigma_sq):
a = (-1*(CUDA(Variable(x))-mu).pow(2)/(2*sigma_sq)).exp()
b = 1/(2*sigma_sq*pi.expand_as(sigma_sq)).sqrt()
return a*b
class HD_Autoregressive_Policy(nn.Module):
def __init__(self):
super(HD_Autoregressive_Policy, self).__init__()
input_size = 30*2+1 ## ？
hidden_size_1 = 32 # ？higher hidden dim ?
self.a_os = 1
self.b_os = 1
self.c_os = 1
self.d_os = 1
self.fc_input = nn.Sequential(
nn.Linear(input_size, hidden_size_1),
nn.ReLU(inplace=True)
)
self.fc_action_a = nn.Sequential(
nn.Linear(hidden_size_1, self.a_os*2),
)
self.fc_action_b = nn.Sequential(
nn.Linear(1+hidden_size_1, self.b_os*2),
)
self.fc_action_c = nn.Sequential(
nn.Linear(1+1+hidden_size_1, self.c_os*2),
)
self.fc_action_d = nn.Sequential(
nn.Linear(1+1+1+hidden_size_1, self.d_os*2),
)
def sample_action(self, normal_action, action_os):
# get the mu and sigma
#mu = torch.tanh(normal_action[:, :action_os])
mu = normal_action[:, :action_os] # concatenated
sigma = F.softplus(normal_action[:, action_os:])
# calculate the probability by mu and sigma of normal distribution
eps = CUDA(Variable(torch.randn(mu.size())))
action = (mu + sigma.sqrt()*eps) # reparameterization
return action, mu, sigma
def forward(self, x):
# p(s)
x = x.view(1, -1)
s = self.fc_input(x)
# p(a|s)
normal_a = self.fc_action_a(s)
action_a, mu_a, sigma_a = self.sample_action(normal_a, self.a_os)
# p(b|a,s)
state_sample_a = torch.cat((s, action_a), dim=1)
normal_b = self.fc_action_b(state_sample_a)
action_b, mu_b, sigma_b = self.sample_action(normal_b, self.b_os)
# p(c|a,b,s)
state_sample_a_b = torch.cat((s, action_a, action_b), dim=1)
normal_c = self.fc_action_c(state_sample_a_b)
action_c, mu_c, sigma_c = self.sample_action(normal_c, self.c_os)
# p(d|a,b,c,s)
state_sample_a_b_c = torch.cat((s, action_a, action_b, action_c), dim=1)
normal_d = self.fc_action_d(state_sample_a_b_c)
action_d, mu_d, sigma_d = self.sample_action(normal_d, self.d_os)
return [mu_a, mu_b, mu_c, mu_d], [sigma_a, sigma_b, sigma_c, sigma_d], [action_a[0], action_b[0], action_c[0], action_d[0]]
# deterministic output
def deterministic_forward(self, x):
# p(s)
x = x.view(1, -1)
s = self.fc_input(x)
# p(a|s)
normal_a = self.fc_action_a(s)
_, mu_a, sigma_a = self.sample_action(normal_a, self.a_os)
# p(b|a,s)
state_sample_a = torch.cat((s, mu_a), dim=1)
normal_b = self.fc_action_b(state_sample_a)
_, mu_b, sigma_b = self.sample_action(normal_b, self.b_os)
# p(c|a,b,s)
state_sample_a_b = torch.cat((s, mu_a, mu_b), dim=1)
normal_c = self.fc_action_c(state_sample_a_b)
_, mu_c, sigma_c = self.sample_action(normal_c, self.c_os)
# p(d|a,b,c,s)
state_sample_a_b_c = torch.cat((s, mu_a, mu_b, mu_c), dim=1)
normal_d = self.fc_action_d(state_sample_a_b_c)
_, mu_d, sigma_d = self.sample_action(normal_d, self.d_os)
print('================================================== Test Action space ==================================================')
print('normal A: {} {}'.format(mu_a, sigma_a))
print('normal B: {} {}'.format(mu_b, sigma_b))
print('normal C: {} {}'.format(mu_c, sigma_c))
print('normal D: {} {}'.format(mu_d, sigma_d))
print('======================================================================================================================')
# output the mean value to be the deterministic action
return mu_a[0][0], mu_b[0][0], mu_c[0][0], mu_d[0][0]
class REINFORCE:
def __init__(self, lr, gamma, model_id=0, model_path='./model'):
self.model = CUDA(HD_Autoregressive_Policy())
self.model.apply(kaiming_init)
self.optimizer = optim.Adam(self.model.parameters(), lr=lr)
self.model.train()
self.model_id = model_id
self.model_path = model_path
self.gamma = gamma
def select_action(self, state):
state = CUDA(Variable(torch.from_numpy(state)))
mu_bag, sigma_bag, action_bag = self.model(state)
# calculate the probability that this distribution outputs this action
prob_a = normal(action_bag[0], mu_bag[0], sigma_bag[0])
prob_b = normal(action_bag[1], mu_bag[1], sigma_bag[1])
prob_c = normal(action_bag[2], mu_bag[2], sigma_bag[2])
prob_d = normal(action_bag[3], mu_bag[3], sigma_bag[3])
log_prob = prob_a.log() + prob_b.log() + prob_c.log() + prob_d.log()
# calculate the entropy
entropy_a = -0.5*((sigma_bag[0]+2*pi.expand_as(sigma_bag[0])).log()+1) # define of pi?
entropy_b = -0.5*((sigma_bag[1]+2*pi.expand_as(sigma_bag[1])).log()+1)
entropy_c = -0.5*((sigma_bag[2]+2*pi.expand_as(sigma_bag[2])).log()+1)
entropy_d = -0.5*((sigma_bag[2]+2*pi.expand_as(sigma_bag[2])).log()+1) # ? typo ?
entropy = entropy_a + entropy_b + entropy_c + entropy_d
a_1 = action_bag[0][0].detach().cpu().numpy()
a_2 = action_bag[1][0].detach().cpu().numpy()
a_3 = action_bag[2][0].detach().cpu().numpy()
a_4 = action_bag[3][0].detach().cpu().numpy()
return [a_1, a_2, a_3, a_4], log_prob, entropy
def deterministic_action(self, state):
with torch.no_grad():
state = CUDA(Variable(torch.from_numpy(state)))
action_a, action_b, action_c, action_d = self.model.deterministic_forward(state)
#print('score a: {}'.format(score_a.cpu().numpy()))
#print('score b: {}'.format(score_b.cpu().numpy()))
#print('score c: {}'.format(score_c.cpu().numpy()))
return [action_a.cpu().numpy(), action_b.cpu().numpy(), action_c.cpu().numpy(), action_d.cpu().numpy()]
def update_parameters(self, rewards, log_probs, entropies):
R = CUDA(torch.zeros(1, 1))
loss = 0
for i in reversed(range(len(rewards))):
R = self.gamma * R + rewards[i]
loss = loss - (log_probs[i]*(Variable(R).expand_as(log_probs[i]))).sum() # - (0.001*entropies[i]).sum()
loss = loss / len(rewards)
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
def save_model(self):
states = {'parameters': self.model.state_dict()}
filepath = os.path.join(self.model_path, 'model.'+str(self.model_id)+'.torch')
with open(filepath, 'wb+') as f:
torch.save(states, f)
def load_model(self):
filepath = os.path.join(self.model_path, 'model.'+str(self.model_id)+'.torch')
if os.path.isfile(filepath):
with open(filepath, 'rb') as f:
checkpoint = torch.load(f)
self.model.load_state_dict(checkpoint['parameters'])