【转载】 强化学习（九）Deep Q-Learning进阶之Nature DQN

原文地址：

https://www.cnblogs.com/pinard/p/9756075.html

-------------------------------------------------------------------------------------------------------

在强化学习（八）价值函数的近似表示与Deep Q-Learning中，我们讲到了Deep Q-Learning（NIPS 2013）的算法和代码，在这个算法基础上，有很多Deep Q-Learning(以下简称DQN)的改进版，今天我们来讨论DQN的第一个改进版Nature DQN(NIPS 2015)。

　　　　本章内容主要参考了ICML 2016的deep RL tutorial和Nature DQN的论文。

1. DQN(NIPS 2013)的问题

在上一篇我们已经讨论了DQN(NIPS 2013)的算法原理和代码实现，虽然它可以训练像CartPole这样的简单游戏，但是有很多问题。这里我们先讨论第一个问题。

注意到DQN(NIPS 2013)里面，我们使用的目标Q值的计算方式：

因此，一个改进版的DQN: Nature DQN尝试用两个Q网络来减少目标Q值计算和要更新Q网络参数之间的依赖关系。下面我们来看看Nature DQN是怎么做的。

2. Nature DQN的建模

        Nature DQN使用了两个Q网络，一个当前Q网络QQQQ用来选择动作，更新模型参数，另一个目标Q网络Q′用于计算目标Q值。目标Q网络的网络参数不需要迭代更新，而是每隔一段时间从当前Q网络QQ复制过来，即延时更新，这样可以减少   目标Q值和当前的Q值  相关性。

　　　　要注意的是，两个Q网络的结构是一模一样的。这样才可以复制网络参数。

Nature DQN和上一篇的DQN相比，除了用一个新的相同结构的目标Q网络来计算目标Q值以外，其余部分基本是完全相同的。

3. Nature DQN的算法流程

　　　　下面我们来总结下Nature DQN的算法流程， 基于DQN NIPS 2015：

注意，上述第二步的f步和g步的Q值计算也都需要通过Q网络计算得到。另外，实际应用中，为了算法较好的收敛，探索率εϵ需要随着迭代的进行而变小。

4. Nature DQN算法实例

下面我们用一个具体的例子来演示DQN的应用。仍然使用了OpenAI Gym中的CartPole-v0游戏来作为我们算法应用。CartPole-v0游戏的介绍参见这里。它比较简单，基本要求就是控制下面的cart移动使连接在上面的pole保持垂直不倒。这个任务只有两个离散动作，要么向左用力，要么向右用力。而state状态就是这个cart的位置和速度， pole的角度和角速度，4维的特征。坚持到200分的奖励则为过关。

　　　　完整的代码参见我的github: https://github.com/ljpzzz/machinelearning/blob/master/reinforcement-learning/nature_dqn.py

　　　　这里我们重点关注Nature DQN和上一节的NIPS 2013 DQN的代码的不同之处。

　　　　首先是Q网络，上一篇的DQN是一个三层的神经网络，而这里我们有两个一样的三层神经网络，一个是当前Q网络，一个是目标Q网络，网络的定义部分如下：

def create_Q_network(self):
    # input layer
    self.state_input = tf.placeholder("float", [None, self.state_dim])
    # network weights
    with tf.variable_scope('current_net'):
        W1 = self.weight_variable([self.state_dim,20])
        b1 = self.bias_variable([20])
        W2 = self.weight_variable([20,self.action_dim])
        b2 = self.bias_variable([self.action_dim])

        # hidden layers
        h_layer = tf.nn.relu(tf.matmul(self.state_input,W1) + b1)
        # Q Value layer
        self.Q_value = tf.matmul(h_layer,W2) + b2

    with tf.variable_scope('target_net'):
        W1t = self.weight_variable([self.state_dim,20])
        b1t = self.bias_variable([20])
        W2t = self.weight_variable([20,self.action_dim])
        b2t = self.bias_variable([self.action_dim])

        # hidden layers
        h_layer_t = tf.nn.relu(tf.matmul(self.state_input,W1t) + b1t)
        # Q Value layer
        self.target_Q_value = tf.matmul(h_layer,W2t) + b2t

　　　　对于定期将目标Q网络的参数更新的代码如下面两部分：

    t_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='target_net')
    e_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='current_net')

    with tf.variable_scope('soft_replacement'):
        self.target_replace_op = [tf.assign(t, e) for t, e in zip(t_params, e_params)]

  def update_target_q_network(self, episode):
    # update target Q netowrk
    if episode % REPLACE_TARGET_FREQ == 0:
        self.session.run(self.target_replace_op)
        #print('episode '+str(episode) +', target Q network params replaced!')

　　　　此外，注意下我们计算目标Q值的部分，这里使用的目标Q网络的参数，而不是当前Q网络的参数：

# Step 2: calculate y
    y_batch = []
    Q_value_batch = self.target_Q_value.eval(feed_dict={self.state_input:next_state_batch})
    for i in range(0,BATCH_SIZE):
      done = minibatch[i][4]
      if done:
        y_batch.append(reward_batch[i])
      else :
        y_batch.append(reward_batch[i] + GAMMA * np.max(Q_value_batch[i]))

　　　　其余部分基本和上一篇DQN的代码相同。这里给出我跑的某一次的结果:

episode: 0 Evaluation Average Reward: 9.8
episode: 100 Evaluation Average Reward: 9.8
episode: 200 Evaluation Average Reward: 9.6
episode: 300 Evaluation Average Reward: 10.0
episode: 400 Evaluation Average Reward: 34.8
episode: 500 Evaluation Average Reward: 177.4
episode: 600 Evaluation Average Reward: 200.0
episode: 700 Evaluation Average Reward: 200.0
episode: 800 Evaluation Average Reward: 200.0
episode: 900 Evaluation Average Reward: 198.4
episode: 1000 Evaluation Average Reward: 200.0
episode: 1100 Evaluation Average Reward: 193.2
episode: 1200 Evaluation Average Reward: 200.0
episode: 1300 Evaluation Average Reward: 200.0
episode: 1400 Evaluation Average Reward: 200.0
episode: 1500 Evaluation Average Reward: 200.0
episode: 1600 Evaluation Average Reward: 200.0
episode: 1700 Evaluation Average Reward: 200.0
episode: 1800 Evaluation Average Reward: 200.0
episode: 1900 Evaluation Average Reward: 200.0
episode: 2000 Evaluation Average Reward: 200.0
episode: 2100 Evaluation Average Reward: 200.0
episode: 2200 Evaluation Average Reward: 200.0
episode: 2300 Evaluation Average Reward: 200.0
episode: 2400 Evaluation Average Reward: 200.0
episode: 2500 Evaluation Average Reward: 200.0
episode: 2600 Evaluation Average Reward: 200.0
episode: 2700 Evaluation Average Reward: 200.0
episode: 2800 Evaluation Average Reward: 200.0
episode: 2900 Evaluation Average Reward: 200.0

注意，由于DQN不保证稳定的收敛，所以每次跑的结果会不同，如果你跑的结果后面仍然收敛的不好，可以把代码多跑几次，选择一个最好的训练结果。

5. Nature DQN总结

Nature DQN对DQN NIPS 2013做了相关性方面的改进，这个改进虽然不错，但是仍然没有解决DQN的 很多问题，比如：

　　　　1） 目标Q值的计算是否准确？全部通过max Q来计算有没有问题？

　　　　2） 随机采样的方法好吗？按道理不同样本的重要性是不一样的。

　　　　3） Q值代表状态，动作的价值，那么单独动作价值的评估会不会更准确？

第一个问题对应的改进是Double DQN, 第二个问题的改进是Prioritised Replay DQN，第三个问题的改进是Dueling DQN，这三个DQN的改进版我们在下一篇来讨论。

（欢迎转载，转载请注明出处。欢迎沟通交流： liujianping-ok@163.com）

------------------------------------------------------------------------------------------------------

#######################################################################
# Copyright (C)                                                       #
# 2016 - 2019 Pinard Liu(liujianping-ok@163.com)                      #
# https://www.cnblogs.com/pinard                                      #
# Permission given to modify the code as long as you keep this        #
# declaration at the top                                              #
#######################################################################
##https://www.cnblogs.com/pinard/p/9756075.html##
##强化学习（九）Deep Q-Learning进阶之Nature DQN##

import gym
import tensorflow as tf
import numpy as np
import random
from collections import deque

# Hyper Parameters for DQN
GAMMA = 0.9 # discount factor for target Q
INITIAL_EPSILON = 0.5 # starting value of epsilon
FINAL_EPSILON = 0.01 # final value of epsilon
REPLAY_SIZE = 10000 # experience replay buffer size
BATCH_SIZE = 32 # size of minibatch
REPLACE_TARGET_FREQ = 10 # frequency to update target Q network

class DQN():
  # DQN Agent
  def __init__(self, env):
    # init experience replay
    self.replay_buffer = deque()
    # init some parameters
    self.time_step = 0
    self.epsilon = INITIAL_EPSILON
    self.state_dim = env.observation_space.shape[0]
    self.action_dim = env.action_space.n

    self.create_Q_network()
    self.create_training_method()

    # Init session
    self.session = tf.InteractiveSession()
    self.session.run(tf.global_variables_initializer())

  def create_Q_network(self):
    # input layer
    self.state_input = tf.placeholder("float", [None, self.state_dim])
    # network weights
    with tf.variable_scope('current_net'):
        W1 = self.weight_variable([self.state_dim,20])
        b1 = self.bias_variable([20])
        W2 = self.weight_variable([20,self.action_dim])
        b2 = self.bias_variable([self.action_dim])

        # hidden layers
        h_layer = tf.nn.relu(tf.matmul(self.state_input,W1) + b1)
        # Q Value layer
        self.Q_value = tf.matmul(h_layer,W2) + b2

    with tf.variable_scope('target_net'):
        W1t = self.weight_variable([self.state_dim,20])
        b1t = self.bias_variable([20])
        W2t = self.weight_variable([20,self.action_dim])
        b2t = self.bias_variable([self.action_dim])

        # hidden layers
        h_layer_t = tf.nn.relu(tf.matmul(self.state_input,W1t) + b1t)
        # Q Value layer
        self.target_Q_value = tf.matmul(h_layer,W2t) + b2t

    t_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='target_net')
    e_params = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='current_net')

    with tf.variable_scope('soft_replacement'):
        self.target_replace_op = [tf.assign(t, e) for t, e in zip(t_params, e_params)]

  def create_training_method(self):
    self.action_input = tf.placeholder("float",[None,self.action_dim]) # one hot presentation
    self.y_input = tf.placeholder("float",[None])
    Q_action = tf.reduce_sum(tf.multiply(self.Q_value,self.action_input),reduction_indices = 1)
    self.cost = tf.reduce_mean(tf.square(self.y_input - Q_action))
    self.optimizer = tf.train.AdamOptimizer(0.0001).minimize(self.cost)

  def perceive(self,state,action,reward,next_state,done):
    one_hot_action = np.zeros(self.action_dim)
    one_hot_action[action] = 1
    self.replay_buffer.append((state,one_hot_action,reward,next_state,done))
    if len(self.replay_buffer) > REPLAY_SIZE:
      self.replay_buffer.popleft()

    if len(self.replay_buffer) > BATCH_SIZE:
      self.train_Q_network()

  def train_Q_network(self):
    self.time_step += 1
    # Step 1: obtain random minibatch from replay memory
    minibatch = random.sample(self.replay_buffer,BATCH_SIZE)
    state_batch = [data[0] for data in minibatch]
    action_batch = [data[1] for data in minibatch]
    reward_batch = [data[2] for data in minibatch]
    next_state_batch = [data[3] for data in minibatch]

    # Step 2: calculate y
    y_batch = []
    Q_value_batch = self.target_Q_value.eval(feed_dict={self.state_input:next_state_batch})
    for i in range(0,BATCH_SIZE):
      done = minibatch[i][4]
      if done:
        y_batch.append(reward_batch[i])
      else :
        y_batch.append(reward_batch[i] + GAMMA * np.max(Q_value_batch[i]))

    self.optimizer.run(feed_dict={
      self.y_input:y_batch,
      self.action_input:action_batch,
      self.state_input:state_batch
      })

  def egreedy_action(self,state):
    Q_value = self.Q_value.eval(feed_dict = {
      self.state_input:[state]
      })[0]
    if random.random() <= self.epsilon:
        self.epsilon -= (INITIAL_EPSILON - FINAL_EPSILON) / 10000
        return random.randint(0,self.action_dim - 1)
    else:
        self.epsilon -= (INITIAL_EPSILON - FINAL_EPSILON) / 10000
        return np.argmax(Q_value)

  def action(self,state):
    return np.argmax(self.Q_value.eval(feed_dict = {
      self.state_input:[state]
      })[0])

  def update_target_q_network(self, episode):
    # update target Q netowrk
    if episode % REPLACE_TARGET_FREQ == 0:
        self.session.run(self.target_replace_op)
        #print('episode '+str(episode) +', target Q network params replaced!')

  def weight_variable(self,shape):
    initial = tf.truncated_normal(shape)
    return tf.Variable(initial)

  def bias_variable(self,shape):
    initial = tf.constant(0.01, shape = shape)
    return tf.Variable(initial)
# ---------------------------------------------------------
# Hyper Parameters
ENV_NAME = 'CartPole-v0'
EPISODE = 3000 # Episode limitation
STEP = 300 # Step limitation in an episode
TEST = 5 # The number of experiment test every 100 episode

def main():
  # initialize OpenAI Gym env and dqn agent
  env = gym.make(ENV_NAME)
  agent = DQN(env)

  for episode in range(EPISODE):
    # initialize task
    state = env.reset()
    # Train
    for step in range(STEP):
      action = agent.egreedy_action(state) # e-greedy action for train
      next_state,reward,done,_ = env.step(action)
      # Define reward for agent
      reward = -1 if done else 0.1
      agent.perceive(state,action,reward,next_state,done)
      state = next_state
      if done:
        break
    # Test every 100 episodes
    if episode % 100 == 0:
      total_reward = 0
      for i in range(TEST):
        state = env.reset()
        for j in range(STEP):
          env.render()
          action = agent.action(state) # direct action for test
          state,reward,done,_ = env.step(action)
          total_reward += reward
          if done:
            break
      ave_reward = total_reward/TEST
      print ('episode: ',episode,'Evaluation Average Reward:',ave_reward)
    agent.update_target_q_network(episode)

if __name__ == '__main__':
    main()