-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathSAC.py
190 lines (155 loc) · 8.51 KB
/
SAC.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
import os
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Adam
from RL_SAC_utils import soft_update, hard_update
from RL_SAC_model import QNetwork, DeterministicPolicy, GaussianPolicy
# # Environment
class Environment(object):
def __init__(self, state0, num_epis, num_prod, num_inj, my_rom):
super(Environment, self).__init__()
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
self.state = state0
self.nsteps = num_epis
self.num_prod = num_prod
self.num_inj = num_inj
self.istep = 0
self.dt = torch.tensor(np.ones((1,1)), dtype=torch.float32).to(device=self.device) # dt=20days, normalized to 1
self.rom = my_rom
self.noise = torch.Tensor(state0.shape[-1]).to(self.device)
self.scale_bhp = (2500-2200)/(4069.2-2200)
self.bias_bhp = (2200-2200)/(4069.2-2200)
self.scale_rate = (1.0e6-1.0e5)/(1.2e6-0)
self.bias_rate = (1.0e5-0)/(1.2e6-0)
self.Qdiff_w = 3151.0 - 0
self.Qdiff_g = 1.2e6 - 0
def step(self, action):
self.istep +=1
self.state, yobs = self.rom.predict_latent(self.state, self.dt, action)
yobs[:, :self.num_prod] = yobs[:, :self.num_prod]*self.Qdiff_w
yobs[:, self.num_prod:self.num_prod*2] = yobs[:, self.num_prod:self.num_prod*2]*self.Qdiff_g
# self.state += action
reward = reward_fun(yobs, action, self.num_prod, self.num_inj)
done = self.istep == self.nsteps
return self.state, reward, done
def reset(self, z0):
self.istep =0
# noise = self.noise.normal_(0., std=0.10)
# z00 = z0 + noise
z00 = z0
self.state = z00
return z00
def sample_action(self):
# action_bhp = torch.FloatTensor(2000+(2500-2000)*torch.rand(self.num_prod)).to(self.device).unsqueeze(0) ## bhp_min +(bhp_max-bhp_min)*sigma
# action_bhp_norm = (action_bhp-2000)/(3322.3*1.25-2000)
# action_rate = torch.FloatTensor(500+(1000 - 500)*torch.rand(self.num_inj)).to(self.device).unsqueeze(0) ## q_min +(q_max-q_min)*sigma
# action_rate_norm = (action_rate-0)/(1000*1.2-0)
action_bhp = torch.FloatTensor(torch.rand(self.num_prod)).to(self.device).unsqueeze(0)* self.scale_bhp + self.bias_bhp ## bhp_min +(bhp_max-bhp_min)*sigma
action_rate = torch.FloatTensor(torch.rand(self.num_inj)).to(self.device).unsqueeze(0)* self.scale_rate + self.bias_rate ## q_min +(q_max-q_min)*sigma
# action = torch.cat((action_bhp_norm,action_rate_norm),dim=1)
action = torch.cat((action_bhp,action_rate),dim=1)
return action
def reward_fun(yobs, action, num_prod, num_inj):
lf3toton =0.1167*4.536e-4 # convert lf^3 to ton
PV = ((50-10)*lf3toton*torch.sum(action[:,num_prod:], dim=1) - 5.0*torch.sum(yobs[:, :num_prod],dim=1) - 50.0*lf3toton*torch.sum(yobs[:, num_prod:num_prod*2], dim=1))/1000
return PV
class SAC(object):
def __init__(self, num_inputs, u_dim):
super(SAC, self).__init__()
self.gamma = 0.986
self.tau = 0.005
# self.alpha = 0.20
self.alpha = 0.00
# self.policy_type = Deterministic
self.target_update_interval = 1
self.automatic_entropy_tuning = False
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
self.critic = QNetwork(num_inputs, u_dim, 200).to(device=self.device)
self.critic_optim = Adam(self.critic.parameters(), lr=0.0001)
self.critic_target = QNetwork(num_inputs, u_dim, 200).to(self.device)
hard_update(self.critic_target, self.critic)
self.alpha = 0
self.automatic_entropy_tuning = False
self.policy = DeterministicPolicy(num_inputs, u_dim, 200).to(device=self.device)
# self.policy = GaussianPolicy(num_inputs, u_dim, 200).to(device=self.device)
self.policy_optim = Adam(self.policy.parameters(), lr=0.0001)
def select_action(self, state, evaluate=False):
# state = torch.FloatTensor(state).to(self.device).unsqueeze(0)
if evaluate is False:
action, _, _ = self.policy.sample(state)
else:
_, _, action = self.policy.sample(state)
return action
def update_parameters(self, memory, batch_size, updates):
# Sample a batch from memory
# state_batch, action_batch, reward_batch, next_state_batch, mask_batch = memory.sample(batch_size=batch_size)
state_batch, action_batch, reward_batch, next_state_batch = memory.sample(batch_size=batch_size)
with torch.no_grad():
next_state_action, next_state_log_pi, _ = self.policy.sample(next_state_batch)
qf1_next_target, qf2_next_target = self.critic_target(next_state_batch, next_state_action)
min_qf_next_target = torch.min(qf1_next_target, qf2_next_target) - self.alpha * next_state_log_pi
# next_q_value = reward_batch + mask_batch * self.gamma * (min_qf_next_target)
next_q_value = reward_batch + self.gamma * (min_qf_next_target)
# print(state_batch)
# print(action_batch)
qf1, qf2 = self.critic(state_batch, action_batch) # Two Q-functions to mitigate positive bias in the policy improvement step
qf1_loss = F.mse_loss(qf1, next_q_value) # JQ = 𝔼(st,at)~D[0.5(Q1(st,at) - r(st,at) - γ(𝔼st+1~p[V(st+1)]))^2]
qf2_loss = F.mse_loss(qf2, next_q_value) # JQ = 𝔼(st,at)~D[0.5(Q1(st,at) - r(st,at) - γ(𝔼st+1~p[V(st+1)]))^2]
qf_loss = qf1_loss + qf2_loss
self.critic_optim.zero_grad()
# qf_loss.backward()
qf_loss.backward(retain_graph=True)
self.critic_optim.step()
pi, log_pi, _ = self.policy.sample(state_batch)
qf1_pi, qf2_pi = self.critic(state_batch, pi)
min_qf_pi = torch.min(qf1_pi, qf2_pi)
policy_loss = ((self.alpha * log_pi) - min_qf_pi).mean() # Jπ = 𝔼st∼D,εt∼N[α * logπ(f(εt;st)|st) − Q(st,f(εt;st))]
# print(policy_loss)
self.policy_optim.zero_grad()
policy_loss.backward(retain_graph=True)
self.policy_optim.step()
# if self.automatic_entropy_tuning:
# alpha_loss = -(self.log_alpha * (log_pi + self.target_entropy).detach()).mean()
# self.alpha_optim.zero_grad()
# alpha_loss.backward()
# self.alpha_optim.step()
# self.alpha = self.log_alpha.exp()
# alpha_tlogs = self.alpha.clone() # For TensorboardX logs
# else:
alpha_loss = torch.tensor(0.).to(self.device)
alpha_tlogs = torch.tensor(self.alpha) # For TensorboardX logs
if updates % self.target_update_interval == 0:
soft_update(self.critic_target, self.critic, self.tau)
return qf1_loss.item(), qf2_loss.item(), policy_loss.item(), alpha_loss.item(), alpha_tlogs.item()
# Save model parameters
def save_checkpoint(self, env_name, suffix="", ckpt_path=None):
if not os.path.exists('checkpoints/'):
os.makedirs('checkpoints/')
if ckpt_path is None:
ckpt_path = "checkpoints/sac_checkpoint_{}_{}".format(env_name, suffix)
# print('Saving models to {}'.format(ckpt_path))
torch.save({'policy_state_dict': self.policy.state_dict(),
'critic_state_dict': self.critic.state_dict(),
'critic_target_state_dict': self.critic_target.state_dict(),
'critic_optimizer_state_dict': self.critic_optim.state_dict(),
'policy_optimizer_state_dict': self.policy_optim.state_dict()}, ckpt_path)
# Load model parameters
def load_checkpoint(self, ckpt_path, evaluate=False):
print('Loading models from {}'.format(ckpt_path))
if ckpt_path is not None:
checkpoint = torch.load(ckpt_path)
self.policy.load_state_dict(checkpoint['policy_state_dict'])
self.critic.load_state_dict(checkpoint['critic_state_dict'])
self.critic_target.load_state_dict(checkpoint['critic_target_state_dict'])
self.critic_optim.load_state_dict(checkpoint['critic_optimizer_state_dict'])
self.policy_optim.load_state_dict(checkpoint['policy_optimizer_state_dict'])
if evaluate:
self.policy.eval()
self.critic.eval()
self.critic_target.eval()
else:
self.policy.train()
self.critic.train()
self.critic_target.train()