SQL编程(四)

十一放假,回来继续更新........

--IF 语句
DECLARE
v_job VARCHAR2(10);
v_sal NUMBER(6,2);
BEGIN
  SELECT job,sal INTO v_job,v_sal
  FROM emp WHERE empno=&no;
  IF v_job ='job'THEN
     UPDATE emp SET sal=v_sal +1000 WHERE dempno=&no;
  ELSE IF v_job='Man' THEN
     UPDATE emp SET sal = v_sal +2000 WHERE dempno=&no;
  ELSE
     UPDATE emp SET sal = v_sal +3000 WHERE dempno=&no;
  END IF;
END; 
               
--CASE 语句
DECLARE
  v_sal emp.sal%TYPE;
  v_ename emp.ename%TYPE;
BEGIN
  SELECT ename,sal INTO v_ename,v_sal
  FROM emp WHERE empno =&no;
  CASE
  WHEN v_sal <1000 THEN
     UPDATE emp SET comm=50 WHERE ename=v_ename;
  WHEN v_sal <2000 THEN
  UPDATE emp SET comm =80 WHERE ename =v_ename;
  WHEN v_sal <6000 THEN
  UPDATE emp SET comm=30 WHERE ename =v_ename;
 END CASE;
END;               

--LOOP循环

DECLARE
  i INT:=1;
BEGIN
  LOOP
  INSERT INTO temp values(i);
  EXIT WHEN i=10;
  i:=i+1;
  END LOOP;
END;

--WHILE 循环
DECLARE
  i INT:=1;
BEGIN
  WHILE i<=10 LOOP
  INSERT INTO temp values(i);
  i:=i+1;
  END LOOP;
END;
--FOR循环
DECLARE
  i INT:=1
BEGIN
   FOR i IN REVERSE 1..10 LOOP
     INSERT INTO temp VALUES(i);
    END LOOP;
END;

-- 嵌套循环
DECLARE
 result INT;
BEGIN
 <<outer>>
 FOR i IN 1..100 LOOP
 <<inner>>
 FOR j IN 1..100 LOOP
 result:=i*j;
 EXIT outer WHEN result =1000;
 EXIT WHEN result =500;
 END LOOP inner;
 END LOOP outer;
END;
--%ROWTYPE 定义记录变量
DECLARE
dept_record dept%ROWTYPE;
BEGIN
 dept_record.dno:='50';
 dept_record.dname:='admin';
 INSERT INTO dept values dept_record;
END;
--索引表
DECLARE
  TYPE a_table_type AS TABLE OF emp.ename%TYPE
       INDEX BY VARCHAR2(10);
       a_table a_table_type;
BEGIN
  a_table('沈阳'):=1;
END;
--嵌套表
DECLARE
TYPE ename_table_type IS TABLE OF emp.ename%TYPE;
   ename_table ename_table_type;
BEGIN
   ename_table:=ename_table_type('1','2','3');--构造函数
SELECT ename INTO ename_table(2) FROM emp
  WHERE empno='&no';
END;
--嵌套表插入数据
BEGIN
 INSERT INTO employee VALUES(1,'fei'.7000,phone_type('123456','9877665'));
EBD;
--检索嵌套表
DECLARE
 p_table phone_table_type;
 BEGIN
   SELECT phone INTO p_table
     FROM employee WHERE id =1;
    FOR i IN 1..p_table.count loop
     END LOOP;
 END;
--VARRAY()实现多维数组
DECLARE
   TYPE al_varray_type IS VARRAY(10) OF INT;
   TYPE nal_varray_type IS VARRAY(10) OF al_varray_type;
   nvl nal_varray_type:=nal_varray_type(
   al_varray_type(58,100,102),
   al_varray_type(44,22,33)
   );
BEGIN
  FOR i IN 1..nvl.count LOOP
   FOR j IN 1..nvl(i).count LOOP
   END LOOP;
  END LOOP;
END;
--批量绑定插入
DECLARE
 TYPE id_table_type AS TABLE OF NUMBER(6)
 INDEX BY BINARY_INTEGER;
 TYPE name_table_type AS TABLE OF VARCHAR2(10)
 INDEX BY BINARY_INTEGER;
 id_table id_table_type;
 name_table name_table_type;
 start_time NUMBER(10);
 end_time NUMBER(10);
BEGIN
 FOR i IN 1..5000 LOOP
  id_table(i):=i;
  name_table(i):='name'||to_char(i);
 END LOOP;
 FORALL i IN 1..id_table.count
   INSERT INTO demo VALUES (id_table(i),name_table(i));
END;

DQN(Deep Q-Network)是一种使用深度神经网络实现的强化学习算法,用于解决离散动作空间的问题。在PyTorch中实现DQN可以分为以下几个步骤: 1. 定义神经网络:使用PyTorch定义一个包含多个全连接层的神经网络,输入为状态空间的维度,输出为动作空间的维度。 ```python import torch.nn as nn import torch.nn.functional as F class QNet(nn.Module): def __init__(self, state_dim, action_dim): super(QNet, self).__init__() self.fc1 = nn.Linear(state_dim, 64) self.fc2 = nn.Linear(64, 64) self.fc3 = nn.Linear(64, action_dim) def forward(self, x): x = F.relu(self.fc1(x)) x = F.relu(self.fc2(x)) x = self.fc3(x) return x ``` 2. 定义经验回放缓存:包含多条经验,每条经验包含一个状态、一个动作、一个奖励和下一个状态。 ```python import random class ReplayBuffer(object): def __init__(self, max_size): self.buffer = [] self.max_size = max_size def push(self, state, action, reward, next_state): if len(self.buffer) < self.max_size: self.buffer.append((state, action, reward, next_state)) else: self.buffer.pop(0) self.buffer.append((state, action, reward, next_state)) def sample(self, batch_size): state, action, reward, next_state = zip(*random.sample(self.buffer, batch_size)) return torch.stack(state), torch.tensor(action), torch.tensor(reward), torch.stack(next_state) ``` 3. 定义DQN算法:使用PyTorch定义DQN算法,包含训练和预测两个方法。 ```python class DQN(object): def __init__(self, state_dim, action_dim, gamma, epsilon, lr): self.qnet = QNet(state_dim, action_dim) self.target_qnet = QNet(state_dim, action_dim) self.gamma = gamma self.epsilon = epsilon self.lr = lr self.optimizer = torch.optim.Adam(self.qnet.parameters(), lr=self.lr) self.buffer = ReplayBuffer(100000) self.loss_fn = nn.MSELoss() def act(self, state): if random.random() < self.epsilon: return random.randint(0, action_dim - 1) else: with torch.no_grad(): q_values = self.qnet(state) return q_values.argmax().item() def train(self, batch_size): state, action, reward, next_state = self.buffer.sample(batch_size) q_values = self.qnet(state).gather(1, action.unsqueeze(1)).squeeze(1) target_q_values = self.target_qnet(next_state).max(1)[0].detach() expected_q_values = reward + self.gamma * target_q_values loss = self.loss_fn(q_values, expected_q_values) self.optimizer.zero_grad() loss.backward() self.optimizer.step() def update_target_qnet(self): self.target_qnet.load_state_dict(self.qnet.state_dict()) ``` 4. 训练模型:使用DQN算法进行训练,并更新目标Q网络。 ```python dqn = DQN(state_dim, action_dim, gamma=0.99, epsilon=1.0, lr=0.001) for episode in range(num_episodes): state = env.reset() total_reward = 0 for step in range(max_steps): action = dqn.act(torch.tensor(state, dtype=torch.float32)) next_state, reward, done, _ = env.step(action) dqn.buffer.push(torch.tensor(state, dtype=torch.float32), action, reward, torch.tensor(next_state, dtype=torch.float32)) state = next_state total_reward += reward if len(dqn.buffer.buffer) > batch_size: dqn.train(batch_size) if step % target_update == 0: dqn.update_target_qnet() if done: break dqn.epsilon = max(0.01, dqn.epsilon * 0.995) ``` 5. 测试模型:使用训练好的模型进行测试。 ```python total_reward = 0 state = env.reset() while True: action = dqn.act(torch.tensor(state, dtype=torch.float32)) next_state, reward, done, _ = env.step(action) state = next_state total_reward += reward if done: break print("Total reward: {}".format(total_reward)) ``` 以上就是在PyTorch中实现DQN强化学习的基本步骤。需要注意的是,DQN算法中还有很多细节和超参数需要调整,具体实现过程需要根据具体问题进行调整。
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值