bacow/easy-rl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update

Browse Source
This commit is contained in:
JohnJim0816
2021-05-06 02:07:56 +08:00
parent 747f3238c0
commit b17c8f4e41
107 changed files with 1439 additions and 987 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
codes/DoubleDQN/agent.py
View File

@@ -5,7 +5,7 @@
@Email: johnjim0816@gmail.com
@Date: 2020-06-12 00:50:49
@LastEditor: John
LastEditTime: 2021-05-04 15:04:45
LastEditTime: 2021-05-04 22:28:06
@Discription:
@Environment: python 3.7.7
'''
@@ -35,9 +35,10 @@ class DoubleDQN:
self.batch_size = cfg.batch_size
self.policy_net = MLP(state_dim, action_dim,hidden_dim=cfg.hidden_dim).to(self.device)
self.target_net = MLP(state_dim, action_dim,hidden_dim=cfg.hidden_dim).to(self.device)
# target_net的初始模型参数完全复制policy_net
self.target_net.load_state_dict(self.policy_net.state_dict())
self.target_net.eval() # 不启用 BatchNormalization 和 Dropout
# target_net copy from policy_net
for target_param, param in zip(self.target_net.parameters(), self.policy_net.parameters()):
target_param.data.copy_(param.data)
# self.target_net.eval() # 不启用 BatchNormalization 和 Dropout
# 可查parameters()与state_dict()的区别，前者require_grad=True
self.optimizer = optim.Adam(self.policy_net.parameters(), lr=cfg.lr)
self.loss = 0
@@ -58,9 +59,9 @@ class DoubleDQN:
def choose_action(self, state):
'''选择动作
'''
self.actions_count += 1
self.epsilon = self.epsilon_end + (self.epsilon_start - self.epsilon_end) * \
math.exp(-1. * self.actions_count / self.epsilon_decay)
self.actions_count += 1
if random.random() > self.epsilon:
action = self.predict(state)
else:
@@ -73,7 +74,7 @@ class DoubleDQN:
# 从memory中随机采样transition
state_batch, action_batch, reward_batch, next_state_batch, done_batch = self.memory.sample(
self.batch_size)
### 转为张量 ###
# convert to tensor
state_batch = torch.tensor(
state_batch, device=self.device, dtype=torch.float)
action_batch = torch.tensor(action_batch, device=self.device).unsqueeze(
@@ -84,8 +85,7 @@ class DoubleDQN:
next_state_batch, device=self.device, dtype=torch.float)
done_batch = torch.tensor(np.float32(
done_batch), device=self.device).unsqueeze(1) # 将bool转为float然后转为张量
done_batch), device=self.device) # 将bool转为float然后转为张量
# 计算当前(s_t,a)对应的Q(s_t, a)
q_values = self.policy_net(state_batch)
next_q_values = self.policy_net(next_state_batch)
@@ -104,7 +104,7 @@ class DoubleDQN:
next_state_batch)
# 选出Q(s_t‘, a)对应的action，代入到next_target_values获得target net对应的next_q_value，即Q’(s_t|a=argmax Q(s_t‘, a))
next_target_q_value = next_target_values.gather(1, torch.max(next_q_values, 1)[1].unsqueeze(1)).squeeze(1)
q_target = reward_batch + self.gamma * next_target_q_value * (1-done_batch[0])
q_target = reward_batch + self.gamma * next_target_q_value * (1-done_batch)
self.loss = nn.MSELoss()(q_value, q_target.unsqueeze(1)) # 计算 均方误差loss
# 优化模型
self.optimizer.zero_grad() # zero_grad清除上一步所有旧的gradients from the last step