class DQN: def __init__(self, state_dim, action_dim, cfg): self.action_dim = action_dim self.state_dim = state_dim self.loss = 0 self.gamma = cfg.gamma self.frame_idx = 0 # 用于epsilon的衰减计数 self.epsilon = lambda frame_idx: cfg.epsilon_end + (cfg.epsilon_start - cfg.epsilon_end) * math.exp(-1. * frame_idx / cfg.epsilon_decay) self.batch_size = cfg.batch_size self.device = cfg.device self.policy_net = MLP(state_dim, action_dim, cfg.hidden_dim).to(cfg.device) self.target_net = MLP(state_dim, action_dim, cfg.hidden_dim).to(cfg.device) self.optimizer = optim.Adam(self.policy_net.parameters(), lr=cfg.lr) self.memory = ReplayBuffer(cfg.memory_capacity) def choose_action(self, state): '''policy_net负责与环境进行互动并产生相关动作存放到经验池中,因为后边会采样经验池中的数据来重新生成相关Q值,所以此处不进行梯度的更新''' self.frame_idx += 1 if random.random() > self.epsilon(self.frame_idx): with torch.no_grad(): # 使用该语句,使policy_net网络不会进行更新 state = torch.tensor([state], device=self.device, dtype=torch.float32) q_value = self.policy_net(state) action = q_value.max(1)[1].item() # tensor.max(1)[1]返回最大值对应的下标,即action else: action = random.randrange(self.action_dim) return action def update(self): if len(self.memory) < self.batch_size: return state_batch, action_batch, reward_batch, next_state_batch, done_batch = self.memory.sample(self.batch_size) ''' 转换为Tensor ''' state_batch = torch.tensor(state_batch, device=self.device, dtype=torch.float) action_batch = torch.tensor(action_batch, device=self.device).unsqueeze(1) reward_batch = torch.tensor(reward_batch, device=self.device, dtype=torch.float) next_state_batch = torch.tensor(next_state_batch, device=self.device, dtype=torch.float) done_batch = torch.tensor(np.float32(done_batch), device=self.device) # 计算当前(s_t, a)对应的Q值,此处的Q值用来训练,所以要求计算梯度; # 其实也可以在choose_action时将q值存到经验池中,就可以不同进行下一步的计算了 q_values = self.policy_net(state_batch).gather(dim=1, index=action_batch) # 计算s_t+1状态下target_net网络的最大值 next_q_values = self.target_net(next_state_batch).max(1)[0].detach() # 由target_net输出的值不会参与到梯度的计算中 # 对于终止状态,此时done_batch[0]=1, 对应的expected_q_value等于reward expected_q_values = reward_batch + self.gamma * next_q_values * (1-done_batch) self.loss = nn.MSELoss()(q_values, expected_q_values.unsqueeze(1)) self.optimizer.zero_grad() self.loss.backward() self.optimizer.step() def save(self, path): torch.save(self.target_net.state_dict(), path+'DQN_CheckPoint.pth') def load(self, path): self.target_net.load_state_dict(torch.load(path+'DQN_CheckPoint.pth'))
class DoubleDQN: def __init__(self, state_dim, action_dim, cfg): self.action_dim = action_dim # 总的动作个数 self.device = cfg.device # 设备,cpu或gpu等 self.gamma = cfg.gamma # e-greedy策略相关参数 self.actions_count = 0 self.epsilon_start = cfg.epsilon_start self.epsilon_end = cfg.epsilon_end self.epsilon_decay = cfg.epsilon_decay 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 # 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 self.memory = ReplayBuffer(cfg.memory_capacity) def predict(self, state): with torch.no_grad(): # 先转为张量便于丢给神经网络,state元素数据原本为float64 # 注意state=torch.tensor(state).unsqueeze(0)跟state=torch.tensor([state])等价 state = torch.tensor([state], device=self.device, dtype=torch.float32) # 如tensor([[-0.0798, -0.0079]], grad_fn=<AddmmBackward>) q_value = self.policy_net(state) # tensor.max(1)返回每行的最大值以及对应的下标, # 如torch.return_types.max(values=tensor([10.3587]),indices=tensor([0])) # 所以tensor.max(1)[1]返回最大值对应的下标,即action action = q_value.max(1)[1].item() return action def choose_action(self, state): '''选择动作 ''' 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: action = random.randrange(self.action_dim) return action def update(self): if len(self.memory) < self.batch_size: return # 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 ### state_batch = torch.tensor(state_batch, device=self.device, dtype=torch.float) action_batch = torch.tensor(action_batch, device=self.device).unsqueeze( 1) # 例如tensor([[1],...,[0]]) reward_batch = torch.tensor( reward_batch, device=self.device, dtype=torch.float) # tensor([1., 1.,...,1]) next_state_batch = torch.tensor(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 # 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) # 代入当前选择的action,得到Q(s_t|a=a_t) q_value = q_values.gather(dim=1, index=action_batch) '''以下是Nature DQN的q_target计算方式 # 计算所有next states的Q'(s_{t+1})的最大值,Q'为目标网络的q函数 next_q_state_value = self.target_net( next_state_batch).max(1)[0].detach() # 比如tensor([ 0.0060, -0.0171,...,]) # 计算 q_target # 对于终止状态,此时done_batch[0]=1, 对应的expected_q_value等于reward q_target = reward_batch + self.gamma * next_q_state_value * (1-done_batch[0]) ''' '''以下是Double DQN q_target计算方式,与NatureDQN稍有不同''' next_target_values = self.target_net(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]) 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 # loss.backward()使用backpropagation计算loss相对于所有parameters(需要gradients)的微分 self.loss.backward() for param in self.policy_net.parameters(): # clip防止梯度爆炸 param.grad.data.clamp_(-1, 1) self.optimizer.step() # 更新模型 def save(self, path): torch.save(self.target_net.state_dict(), path + 'checkpoint.pth') def load(self, path): self.target_net.load_state_dict(torch.load(path + 'checkpoint.pth')) for target_param, param in zip(self.target_net.parameters(), self.policy_net.parameters()): param.data.copy_(target_param.data)
class HierarchicalDQN: def __init__(self, state_dim, action_dim, cfg): self.state_dim = state_dim self.action_dim = action_dim self.gamma = cfg.gamma self.device = cfg.device self.batch_size = cfg.batch_size self.frame_idx = 0 self.epsilon = lambda frame_idx: cfg.epsilon_end + ( cfg.epsilon_start - cfg.epsilon_end) * math.exp(-1. * frame_idx / cfg.epsilon_decay) self.policy_net = MLP(2 * state_dim, action_dim, cfg.hidden_dim).to(self.device) self.meta_policy_net = MLP(state_dim, state_dim, cfg.hidden_dim).to(self.device) self.optimizer = optim.Adam(self.policy_net.parameters(), lr=cfg.lr) self.meta_optimizer = optim.Adam(self.meta_policy_net.parameters(), lr=cfg.lr) self.memory = ReplayBuffer(cfg.memory_capacity) self.meta_memory = ReplayBuffer(cfg.memory_capacity) self.loss_numpy = 0 self.meta_loss_numpy = 0 self.losses = [] self.meta_losses = [] def to_onehot(self, x): oh = np.zeros(self.state_dim) oh[x - 1] = 1. return oh def set_goal(self, state): if random.random() > self.epsilon(self.frame_idx): with torch.no_grad(): state = torch.tensor(state, device=self.device, dtype=torch.float32).unsqueeze(0) goal = self.meta_policy_net(state).max(1)[1].item() else: goal = random.randrange(self.state_dim) return goal def choose_action(self, state): self.frame_idx += 1 if random.random() > self.epsilon(self.frame_idx): with torch.no_grad(): state = torch.tensor(state, device=self.device, dtype=torch.float32).unsqueeze(0) q_value = self.policy_net(state) action = q_value.max(1)[1].item() else: action = random.randrange(self.action_dim) return action def update(self): self.update_policy() self.update_meta() def update_policy(self): if self.batch_size > len(self.memory): return state_batch, action_batch, reward_batch, next_state_batch, done_batch = self.memory.sample( self.batch_size) state_batch = torch.tensor(state_batch, dtype=torch.float) action_batch = torch.tensor(action_batch, dtype=torch.int64).unsqueeze(1) reward_batch = torch.tensor(reward_batch, dtype=torch.float) next_state_batch = torch.tensor(next_state_batch, dtype=torch.float) done_batch = torch.tensor(np.float32(done_batch)) q_values = self.policy_net(state_batch).gather( dim=1, index=action_batch).squeeze(1) next_state_values = self.policy_net(next_state_batch).max( 1)[0].detach() expected_q_values = reward_batch + 0.99 * next_state_values * ( 1 - done_batch) loss = nn.MSELoss()(q_values, expected_q_values) self.optimizer.zero_grad() loss.backward() for param in self.policy_net.parameters(): # clip防止梯度爆炸 param.grad.data.clamp_(-1, 1) self.optimizer.step() self.loss_numpy = loss.detach().numpy() self.losses.append(self.loss_numpy) def update_meta(self): if self.batch_size > len(self.meta_memory): return state_batch, action_batch, reward_batch, next_state_batch, done_batch = self.meta_memory.sample( self.batch_size) state_batch = torch.tensor(state_batch, dtype=torch.float) action_batch = torch.tensor(action_batch, dtype=torch.int64).unsqueeze(1) reward_batch = torch.tensor(reward_batch, dtype=torch.float) next_state_batch = torch.tensor(next_state_batch, dtype=torch.float) done_batch = torch.tensor(np.float32(done_batch)) q_values = self.meta_policy_net(state_batch).gather( dim=1, index=action_batch).squeeze(1) next_state_values = self.meta_policy_net(next_state_batch).max( 1)[0].detach() expected_q_values = reward_batch + 0.99 * next_state_values * ( 1 - done_batch) meta_loss = nn.MSELoss()(q_values, expected_q_values) self.meta_optimizer.zero_grad() meta_loss.backward() for param in self.meta_policy_net.parameters(): # clip防止梯度爆炸 param.grad.data.clamp_(-1, 1) self.meta_optimizer.step() self.meta_loss_numpy = meta_loss.detach().numpy() self.meta_losses.append(self.meta_loss_numpy) def save(self, path): torch.save(self.policy_net.state_dict(), path + 'policy_checkpoint.pth') torch.save(self.meta_policy_net.state_dict(), path + 'meta_checkpoint.pth') def load(self, path): self.policy_net.load_state_dict( torch.load(path + 'policy_checkpoint.pth')) self.meta_policy_net.load_state_dict( torch.load(path + 'meta_checkpoint.pth'))
class DQN: def __init__(self, state_dim, action_dim, cfg): self.action_dim = action_dim # 总的动作个数 self.device = cfg.device # 设备,cpu或gpu等 self.gamma = cfg.gamma # 奖励的折扣因子 # e-greedy策略相关参数 self.sample_count = 0 # 用于epsilon的衰减计数 self.epsilon = 0 self.epsilon_start = cfg.epsilon_start self.epsilon_end = cfg.epsilon_end self.epsilon_decay = cfg.epsilon_decay 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 # 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 self.memory = ReplayBuffer(cfg.memory_capacity) def choose_action(self, state, train=True): '''选择动作 ''' if train: self.epsilon = self.epsilon_end + (self.epsilon_start - self.epsilon_end) * \ math.exp(-1. * self.sample_count / self.epsilon_decay) self.sample_count += 1 if random.random() > self.epsilon: with torch.no_grad(): # 先转为张量便于丢给神经网络,state元素数据原本为float64 # 注意state=torch.tensor(state).unsqueeze(0)跟state=torch.tensor([state])等价 state = torch.tensor([state], device=self.device, dtype=torch.float32) # 如tensor([[-0.0798, -0.0079]], grad_fn=<AddmmBackward>) q_value = self.policy_net(state) # tensor.max(1)返回每行的最大值以及对应的下标, # 如torch.return_types.max(values=tensor([10.3587]),indices=tensor([0])) # 所以tensor.max(1)[1]返回最大值对应的下标,即action action = q_value.max(1)[1].item() else: action = random.randrange(self.action_dim) return action else: with torch.no_grad(): # 取消保存梯度 # 先转为张量便于丢给神经网络,state元素数据原本为float64 # 注意state=torch.tensor(state).unsqueeze(0)跟state=torch.tensor([state])等价 state = torch.tensor( [state], device='cpu', dtype=torch.float32 ) # 如tensor([[-0.0798, -0.0079]], grad_fn=<AddmmBackward>) q_value = self.target_net(state) # tensor.max(1)返回每行的最大值以及对应的下标, # 如torch.return_types.max(values=tensor([10.3587]),indices=tensor([0])) # 所以tensor.max(1)[1]返回最大值对应的下标,即action action = q_value.max(1)[1].item() return action def update(self): if len(self.memory) < self.batch_size: return # 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 例如tensor([[-4.5543e-02, -2.3910e-01, 1.8344e-02, 2.3158e-01],...,[-1.8615e-02, -2.3921e-01, -1.1791e-02, 2.3400e-01]])''' state_batch = torch.tensor(state_batch, device=self.device, dtype=torch.float) action_batch = torch.tensor(action_batch, device=self.device).unsqueeze( 1) # 例如tensor([[1],...,[0]]) reward_batch = torch.tensor( reward_batch, device=self.device, dtype=torch.float) # tensor([1., 1.,...,1]) next_state_batch = torch.tensor(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 '''计算当前(s_t,a)对应的Q(s_t, a)''' '''torch.gather:对于a=torch.Tensor([[1,2],[3,4]]),那么a.gather(1,torch.Tensor([[0],[1]]))=torch.Tensor([[1],[3]])''' q_values = self.policy_net(state_batch).gather( dim=1, index=action_batch) # 等价于self.forward # 计算所有next states的V(s_{t+1}),即通过target_net中选取reward最大的对应states next_state_values = self.target_net(next_state_batch).max( 1)[0].detach() # 比如tensor([ 0.0060, -0.0171,...,]) # 计算 expected_q_value # 对于终止状态,此时done_batch[0]=1, 对应的expected_q_value等于reward expected_q_values = reward_batch + self.gamma * \ next_state_values * (1-done_batch[0]) # self.loss = F.smooth_l1_loss(q_values,expected_q_values.unsqueeze(1)) # 计算 Huber loss self.loss = nn.MSELoss()(q_values, expected_q_values.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 # loss.backward()使用backpropagation计算loss相对于所有parameters(需要gradients)的微分 self.loss.backward() for param in self.policy_net.parameters(): # clip防止梯度爆炸 param.grad.data.clamp_(-1, 1) self.optimizer.step() # 更新模型 def save(self, path): torch.save(self.target_net.state_dict(), path + 'dqn_checkpoint.pth') def load(self, path): self.target_net.load_state_dict(torch.load(path + 'dqn_checkpoint.pth'))
class DoubleDQN: def __init__(self, state_dim, action_dim, cfg): self.action_dim = action_dim self.state_dim = state_dim self.gamma = cfg.gamma self.policy_net = MLP(state_dim, action_dim, cfg.hidden_dim).to(cfg.device) self.target_net = MLP(state_dim, action_dim, cfg.hidden_dim).to(cfg.device) self.target_net.load_state_dict(self.policy_net.state_dict()) self.target_net.eval() # 不启用 BatchNormalization 和 Dropout self.optim = optim.Adam(self.policy_net.parameters(), lr=cfg.lr) self.device = cfg.device self.frame_idx = 0 self.epsilon = lambda frame_idx: cfg.epsilon_end + ( cfg.epsilon_start - cfg.epsilon_end) * math.exp(-1. * frame_idx / cfg.epsilon_decay) self.memory = ReplayBuffer(cfg.memory_capacity) self.batch_size = cfg.batch_size self.loss = 0 def choose_action(self, state): self.frame_idx += 1 state = torch.tensor([state], device=self.device, dtype=torch.float32) if random.random() > self.epsilon(self.frame_idx): with torch.no_grad(): # 此处不进行梯度传播 q_values = self.policy_net(state) action = q_values.max(1)[1].item() else: action = random.randrange(self.action_dim) return action def update(self): if len(self.memory) < self.batch_size: return # 抽样数据 state_batch, action_batch, reward_batch, next_state_batch, done_batch = self.memory.sample( self.batch_size) # 将数据转换为Tensor并推送到GPU state_batch = torch.tensor(state_batch, device=self.device, dtype=torch.float32) action_batch = torch.tensor(action_batch, device=self.device).unsqueeze(1) reward_batch = torch.tensor(reward_batch, device=self.device) next_state_batch = torch.tensor(next_state_batch, device=self.device, dtype=torch.float32) done_batch = torch.tensor(done_batch, device=self.device) # 产生(s_t,a)下的q值 q_values = self.policy_net(state_batch).gather(dim=1, index=action_batch) # 计算next_q_values next_action_batch = self.policy_net(next_state_batch).max( 1)[1].unsqueeze(1) # 此处就是DoubleDQN的关键,动作的选取是通过policy_net的 next_q_values = self.target_net(next_state_batch).gather( dim=1, index=next_action_batch).detach().squeeze(1) # q值是target_net输出的 expected_q_values = reward_batch + self.gamma * next_q_values * ( ~done_batch) self.loss = nn.MSELoss()(q_values, expected_q_values.unsqueeze(1)) self.optim.zero_grad() self.loss.backward() for param in self.policy_net.parameters(): # clip防止梯度爆炸 param.grad.data.clamp_(-1, 1) self.optim.step() def save(self, path): torch.save(self.target_net.state_dict(), path + 'DQN_CheckPoint.pth') def load(self, path): self.target_net.load_state_dict(torch.load(path + 'DQN_CheckPoint.pth'))
class DQN: def __init__(self, state_dim, action_dim, cfg): self.action_dim = action_dim # 总的动作个数 self.device = cfg.device # 设备,cpu或gpu等 self.gamma = cfg.gamma # 奖励的折扣因子 # e-greedy策略相关参数 self.frame_idx = 0 # 用于epsilon的衰减计数 self.epsilon = lambda frame_idx: cfg.epsilon_end + \ (cfg.epsilon_start - cfg.epsilon_end) * \ math.exp(-1. * frame_idx / cfg.epsilon_decay) 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) for target_param, param in zip(self.target_net.parameters(),self.policy_net.parameters()): # copy params from policy net target_param.data.copy_(param.data) self.optimizer = optim.Adam(self.policy_net.parameters(), lr=cfg.lr) self.memory = ReplayBuffer(cfg.memory_capacity) def choose_action(self, state): '''选择动作 ''' self.frame_idx += 1 if random.random() > self.epsilon(self.frame_idx): action = self.predict(state) else: action = random.randrange(self.action_dim) return action def predict(self,state): with torch.no_grad(): state = torch.tensor([state], device=self.device, dtype=torch.float32) q_values = self.policy_net(state) action = q_values.max(1)[1].item() return action def update(self): if len(self.memory) < self.batch_size: return # 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 例如tensor([[-4.5543e-02, -2.3910e-01, 1.8344e-02, 2.3158e-01],...,[-1.8615e-02, -2.3921e-01, -1.1791e-02, 2.3400e-01]])''' state_batch = torch.tensor( state_batch, device=self.device, dtype=torch.float) action_batch = torch.tensor(action_batch, device=self.device).unsqueeze( 1) # 例如tensor([[1],...,[0]]) reward_batch = torch.tensor( reward_batch, device=self.device, dtype=torch.float) # tensor([1., 1.,...,1]) next_state_batch = torch.tensor( next_state_batch, device=self.device, dtype=torch.float) done_batch = torch.tensor(np.float32( done_batch), device=self.device) '''计算当前(s_t,a)对应的Q(s_t, a)''' '''torch.gather:对于a=torch.Tensor([[1,2],[3,4]]),那么a.gather(1,torch.Tensor([[0],[1]]))=torch.Tensor([[1],[3]])''' q_values = self.policy_net(state_batch).gather( dim=1, index=action_batch) # 等价于self.forward # 计算所有next states的V(s_{t+1}),即通过target_net中选取reward最大的对应states next_q_values = self.target_net(next_state_batch).max( 1)[0].detach() # 比如tensor([ 0.0060, -0.0171,...,]) # 计算 expected_q_value # 对于终止状态,此时done_batch[0]=1, 对应的expected_q_value等于reward expected_q_values = reward_batch + \ self.gamma * next_q_values * (1-done_batch) # self.loss = F.smooth_l1_loss(q_values,expected_q_values.unsqueeze(1)) # 计算 Huber loss loss = nn.MSELoss()(q_values, expected_q_values.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 # loss.backward()使用backpropagation计算loss相对于所有parameters(需要gradients)的微分 loss.backward() # for param in self.policy_net.parameters(): # clip防止梯度爆炸 # param.grad.data.clamp_(-1, 1) self.optimizer.step() # 更新模型 def save(self, path): torch.save(self.target_net.state_dict(), path+'dqn_checkpoint.pth') def load(self, path): self.target_net.load_state_dict(torch.load(path+'dqn_checkpoint.pth')) for target_param, param in zip(self.target_net.parameters(), self.policy_net.parameters()): param.data.copy_(target_param.data)