-
Notifications
You must be signed in to change notification settings - Fork 2
/
TD3.py
176 lines (129 loc) · 6.58 KB
/
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
import numpy as np
import torch
import torch.nn as nn
from torch.autograd import Variable
import torch.nn.functional as F
import utils
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Implementation of Twin Delayed Deep Deterministic Policy Gradients (TD3)
# Paper: https://arxiv.org/abs/1802.09477
class Actor(nn.Module):
def __init__(self, state_dim, action_dim, max_action):
super(Actor, self).__init__()
self.l1 = nn.Linear(state_dim, 400)
self.l2 = nn.Linear(400, 300)
self.l3 = nn.Linear(300, action_dim)
self.max_action = max_action
def forward(self, x):
x = F.relu(self.l1(x))
x = F.relu(self.l2(x))
x = self.max_action * torch.tanh(self.l3(x))
return x
class Critic(nn.Module):
def __init__(self, state_dim, action_dim):
super(Critic, self).__init__()
# Q1 architecture
self.l1 = nn.Linear(state_dim + action_dim, 400)
self.l2 = nn.Linear(400, 300)
self.l3 = nn.Linear(300, 1)
# Q2 architecture
self.l4 = nn.Linear(state_dim + action_dim, 400)
self.l5 = nn.Linear(400, 300)
self.l6 = nn.Linear(300, 1)
def forward(self, x, u):
xu = torch.cat([x, u], 1)
x1 = F.relu(self.l1(xu))
x1 = F.relu(self.l2(x1))
x1 = self.l3(x1)
x2 = F.relu(self.l4(xu))
x2 = F.relu(self.l5(x2))
x2 = self.l6(x2)
return x1, x2
def Q1(self, x, u):
xu = torch.cat([x, u], 1)
x1 = F.relu(self.l1(xu))
x1 = F.relu(self.l2(x1))
x1 = self.l3(x1)
return x1
class TD3(object):
def __init__(self, state_dim, action_dim, max_action, transfer_env, transfer_state_dim, transfer_action_dim):
#setup actor
self.actor = Actor(state_dim, action_dim, max_action).to(device)
self.actor_target = Actor(state_dim, action_dim, max_action).to(device)
if transfer_env is not None:
print("training model via transfer")
#load pretrained model
self.actor_transfer = Actor(transfer_state_dim, transfer_action_dim, max_action).to(device)
self.actor_transfer.load_state_dict(torch.load('%s/%s_actor.pth' % ('./pytorch_models',transfer_env),map_location=device))
#replace first and last layers to avoid issues with state and action space size differences between environments
self.actor_transfer.l1 = nn.Linear(state_dim, 400)
self.actor_transfer.l3 = nn.Linear(300, action_dim)
self.actor.load_state_dict(self.actor_transfer.state_dict())
else:
print("training model from scratch")
#copy actor to actor target so their initialization is the same
self.actor_target.load_state_dict(self.actor.state_dict())
self.actor_optimizer = torch.optim.Adam(self.actor.parameters())
#setup critic
self.critic = Critic(state_dim, action_dim).to(device)
self.critic_target = Critic(state_dim, action_dim).to(device)
if transfer_env is not None:
#load pretrained model
self.critic_transfer = Critic(transfer_state_dim, transfer_action_dim).to(device)
self.critic_transfer.load_state_dict(torch.load('%s/%s_critic.pth' % ('./pytorch_models',transfer_env),map_location=device))
#replace first and last layers to avoid issues with state and action space size differences between environments
self.critic_transfer.l1 = nn.Linear(state_dim + action_dim, 400)
self.critic_transfer.l4 = nn.Linear(state_dim + action_dim, 400)
self.critic_transfer.l3 = nn.Linear(300, 1)
self.critic_transfer.l6 = nn.Linear(300, 1)
self.critic.load_state_dict(self.critic_transfer.state_dict())
self.critic_target.load_state_dict(self.critic.state_dict())
self.critic_optimizer = torch.optim.Adam(self.critic.parameters())
self.max_action = max_action
def select_action(self, state):
state = torch.FloatTensor(state.reshape(1, -1)).to(device)
return self.actor(state).cpu().data.numpy().flatten()
def train(self, replay_buffer, iterations, batch_size=100, discount=0.99, tau=0.005, policy_noise=0.2, noise_clip=0.5, policy_freq=2):
for it in range(iterations):
# Sample replay buffer
x, y, u, r, d = replay_buffer.sample(batch_size)
state = torch.FloatTensor(x).to(device)
action = torch.FloatTensor(u).to(device)
next_state = torch.FloatTensor(y).to(device)
done = torch.FloatTensor(1 - d).to(device)
reward = torch.FloatTensor(r).to(device)
# Select action according to policy and add clipped noise
noise = torch.FloatTensor(u).data.normal_(0, policy_noise).to(device)
noise = noise.clamp(-noise_clip, noise_clip)
next_action = (self.actor_target(next_state) + noise).clamp(-self.max_action, self.max_action)
# Compute the target Q value
target_Q1, target_Q2 = self.critic_target(next_state, next_action)
target_Q = torch.min(target_Q1, target_Q2)
target_Q = reward + (done * discount * target_Q).detach()
# Get current Q estimates
current_Q1, current_Q2 = self.critic(state, action)
# Compute critic loss
critic_loss = F.mse_loss(current_Q1, target_Q) + F.mse_loss(current_Q2, target_Q)
# Optimize the critic
self.critic_optimizer.zero_grad()
critic_loss.backward()
self.critic_optimizer.step()
# Delayed policy updates
if it % policy_freq == 0:
# Compute actor loss
actor_loss = -self.critic.Q1(state, self.actor(state)).mean()
# Optimize the actor
self.actor_optimizer.zero_grad()
actor_loss.backward()
self.actor_optimizer.step()
# Update the frozen target models
for param, target_param in zip(self.critic.parameters(), self.critic_target.parameters()):
target_param.data.copy_(tau * param.data + (1 - tau) * target_param.data)
for param, target_param in zip(self.actor.parameters(), self.actor_target.parameters()):
target_param.data.copy_(tau * param.data + (1 - tau) * target_param.data)
def save(self, filename, directory):
torch.save(self.actor.state_dict(), '%s/%s_actor.pth' % (directory, filename))
torch.save(self.critic.state_dict(), '%s/%s_critic.pth' % (directory, filename))
def load(self, filename, directory):
self.actor.load_state_dict(torch.load('%s/%s_actor.pth' % (directory, filename),map_location=device))
self.critic.load_state_dict(torch.load('%s/%s_critic.pth' % (directory, filename),map_location=device))