forked from 7enTropy7/Rock_Pop
-
Notifications
You must be signed in to change notification settings - Fork 0
/
agent.py
158 lines (129 loc) · 6.73 KB
/
agent.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
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import copy
from collections import namedtuple, deque
import random
import numpy as np
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
class Actor(nn.Module):
def __init__(self,state_space_size,action_space_size,seed,fully_conected_units=256):
super(Actor,self).__init__()
self.seed = torch.manual_seed(seed)
self.fully_connected_1 = nn.Linear(state_space_size,fully_conected_units)
self.fully_connected_2 = nn.Linear(fully_conected_units,action_space_size)
self.params_reset()
def forward(self,state):
x = F.relu(self.fully_connected_1(state))
return F.tanh(self.fully_connected_2(x))
def params_reset(self):
self.fully_connected_1.weight.data.uniform_(*limits(self.fully_connected_1))
self.fully_connected_2.weight.data.uniform_(-3e-3,3e-3)
class Critic(nn.Module):
def __init__(self,state_space_size,action_space_size,seed,hidden_1_units=256,hidden_2_units=256,hidden_3_units=128):
super(Critic,self).__init__()
self.seed = torch.manual_seed(seed)
self.hidden_1 = nn.Linear(state_space_size,hidden_1_units)
self.fully_connected_2 = nn.Linear(hidden_1_units+action_space_size,hidden_2_units)
self.fully_connected_3 = nn.Linear(hidden_2_units,hidden_3_units)
self.fully_connected_4 = nn.Linear(hidden_3_units,1)
self.params_reset()
def forward(self,state,action):
t = F.leaky_relu(self.hidden_1(state))
x = torch.cat((t,action),dim=1)
x = F.leaky_relu(self.fully_connected_2(x))
x = F.leaky_relu(self.fully_connected_3(x))
return self.fully_connected_4(x)
def params_reset(self):
self.hidden_1.weight.data.uniform_(*limits(self.hidden_1))
self.fully_connected_2.weight.data.uniform_(*limits(self.fully_connected_2))
self.fully_connected_3.weight.data.uniform_(*limits(self.fully_connected_3))
self.fully_connected_4.weight.data.uniform_(-3e-3,3e-3)
class Memory:
def __init__(self,action_space_size,buffer_size,batch_size,seed):
self.action_space_size = action_space_size
self.memory = deque(maxlen=buffer_size)
self.batch_size = batch_size
self.experience = namedtuple("Experience", field_names=["state", "action", "reward", "next_state", "done"])
self.seed = random.seed(seed)
def add_exp(self, state, action, reward, next_state, done):
exp = self.experience(state, action, reward, next_state, done)
self.memory.append(exp)
def memory_sample(self):
experiences = random.sample(self.memory, k=self.batch_size)
states = torch.from_numpy(np.vstack([exp.state for exp in experiences if exp is not None])).float().to(device)
actions = torch.from_numpy(np.vstack([exp.action for exp in experiences if exp is not None])).float().to(device)
rewards = torch.from_numpy(np.vstack([exp.reward for exp in experiences if exp is not None])).float().to(device)
next_states = torch.from_numpy(np.vstack([exp.next_state for exp in experiences if exp is not None])).float().to(device)
dones = torch.from_numpy(np.vstack([exp.done for exp in experiences if exp is not None]).astype(np.uint8)).float().to(device)
return (states, actions, rewards, next_states, dones)
def __len__(self):
return len(self.memory)
class Noise:
def __init__(self,size,seed,mu=0.,theta=0.15,sigma=0.2):
self.theta = theta
self.sigma = sigma
self.seed = random.seed(seed)
self.mu = mu * np.ones(size)
self.reset()
def reset(self):
self.state = copy.copy(self.mu)
def noise_sample(self):
x = self.state
dx = self.theta*(self.mu-x)+self.sigma*np.array([random.random() for i in range(len(x))])
self.state = x+dx
return self.state
def limits(layer):
inp = layer.weight.data.size()[0]
limit = 1./np.sqrt(inp)
return (-limit, limit)
class DDPG_Agent():
def __init__(self,state_space_size,action_space_size,random_seed):
self.state_space_size = state_space_size
self.action_space_size = action_space_size
self.seed = random.seed(random_seed)
self.actor_local = Actor(state_space_size, action_space_size, random_seed).to(device)
self.actor_target = Actor(state_space_size, action_space_size, random_seed).to(device)
self.actor_optimizer = optim.Adam(self.actor_local.parameters(), lr=1e-4)
self.critic_local = Critic(state_space_size, action_space_size, random_seed).to(device)
self.critic_target = Critic(state_space_size, action_space_size, random_seed).to(device)
self.critic_optimizer = optim.Adam(self.critic_local.parameters(), lr=3e-4, weight_decay=0.0001)
self.noise = Noise(action_space_size, random_seed)
self.memory = Memory(action_space_size, int(1e6), 128, random_seed)
def reset(self):
self.noise.reset()
def soft_update(self, local_model, target_model, tau):
for target_param, local_param in zip(target_model.parameters(), local_model.parameters()):
target_param.data.copy_(tau*local_param.data + (1.0-tau)*target_param.data)
def learn(self, experiences, gamma):
states, actions, rewards, next_states, dones = experiences
actions_next = self.actor_target(next_states)
Q_targets_next = self.critic_target(next_states, actions_next)
Q_targets = rewards+(gamma*Q_targets_next*(1-dones))
Q_expected = self.critic_local(states,actions)
critic_loss = F.mse_loss(Q_expected,Q_targets)
self.critic_optimizer.zero_grad()
critic_loss.backward()
self.critic_optimizer.step()
actions_pred = self.actor_local(states)
actor_loss = -self.critic_local(states, actions_pred).mean()
self.actor_optimizer.zero_grad()
actor_loss.backward()
self.actor_optimizer.step()
self.soft_update(self.critic_local, self.critic_target, 1e-3)
self.soft_update(self.actor_local, self.actor_target, 1e-3)
def current_action(self, state, add_noise=True):
state = torch.from_numpy(state).float().to(device)
self.actor_local.eval()
with torch.no_grad():
action = self.actor_local(state).cpu().data.numpy()
self.actor_local.train()
if add_noise:
action += self.noise.noise_sample()
return np.clip(action, -1, 1)
def step(self, state, action, reward, next_state, done):
self.memory.add_exp(state, action, reward, next_state, done)
if len(self.memory) > 128:
experiences = self.memory.memory_sample()
self.learn(experiences, 0.99)