update

codes/QLearning/README.md

@@ -0,0 +1,3 @@
+# Q-learning
+
+#TODO

codes/QLearning/agent.py

@@ -5,8 +5,8 @@ Author: John
 Email: johnjim0816@gmail.com
 Date: 2020-09-11 23:03:00
 LastEditor: John
-LastEditTime: 2021-03-26 16:51:01
-Discription: 
+LastEditTime: 2021-04-29 16:59:41
+Discription: use defaultdict to define Q table
 Environment: 
 '''
 import numpy as np
@@ -15,7 +15,7 @@ import torch
 from collections import defaultdict
 
 class QLearning(object):
-    def __init__(self,
+    def __init__(self,state_dim,
                  action_dim,cfg):
         self.action_dim = action_dim  # dimension of acgtion
         self.lr = cfg.lr  # learning rate
@@ -26,17 +26,20 @@ class QLearning(object):
         self.epsilon_end = cfg.epsilon_end
         self.epsilon_decay = cfg.epsilon_decay
         self.Q_table  = defaultdict(lambda: np.zeros(action_dim)) # A nested dictionary that maps state -> (action -> action-value)
+        
     def choose_action(self, state):
         self.sample_count += 1
         self.epsilon = self.epsilon_end + (self.epsilon_start - self.epsilon_end) * \
             math.exp(-1. * self.sample_count / self.epsilon_decay)
         # e-greedy policy
         if np.random.uniform(0, 1) > self.epsilon:
-            action = np.argmax(self.Q_table[str(state)])
+            action = self.predict(state)
         else:
             action = np.random.choice(self.action_dim) 
         return action
-            
+    def predict(self,state):
+        action = np.argmax(self.Q_table[str(state)])
+        return action
     def update(self, state, action, reward, next_state, done):
         Q_predict = self.Q_table[str(state)][action]
         if done:

codes/QLearning/agent1.py

@@ -0,0 +1,88 @@
+#!/usr/bin/env python
+# coding=utf-8
+'''
+Author: John
+Email: johnjim0816@gmail.com
+Date: 2020-09-11 23:03:00
+LastEditor: John
+LastEditTime: 2021-04-29 17:02:00
+Discription: 
+Environment: 
+'''
+#   Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import numpy as np
+import math
+#!/usr/bin/env python
+# coding=utf-8
+'''
+Author: John
+Email: johnjim0816@gmail.com
+Date: 2020-09-11 23:03:00
+LastEditor: John
+LastEditTime: 2021-04-29 16:45:33
+Discription: use np array to define Q table
+Environment: 
+'''
+import numpy as np
+import math
+
+class QLearning(object):
+    def __init__(self,
+                 state_dim,action_dim,cfg):
+        self.action_dim = action_dim  # dimension of acgtion
+        self.lr = cfg.lr  # learning rate
+        self.gamma = cfg.gamma  
+        self.epsilon = 0 
+        self.sample_count = 0  
+        self.epsilon_start = cfg.epsilon_start
+        self.epsilon_end = cfg.epsilon_end
+        self.epsilon_decay = cfg.epsilon_decay
+        self.Q_table = np.zeros((state_dim, action_dim)) # Q表
+        
+    def choose_action(self, state):
+        self.sample_count += 1
+        self.epsilon = self.epsilon_end + (self.epsilon_start - self.epsilon_end) * \
+            math.exp(-1. * self.sample_count / self.epsilon_decay)
+        if np.random.uniform(0, 1) > self.epsilon:  # 随机选取0-1之间的值，如果大于epsilon就按照贪心策略选取action，否则随机选取
+            action = self.predict(state)
+        else:
+            action = np.random.choice(self.action_dim)  #有一定概率随机探索选取一个动作
+        return action
+
+    def predict(self, state):
+        '''根据输入观测值，采样输出的动作值，带探索，测试模型时使用
+        '''
+        Q_list = self.Q_table[state, :]
+        Q_max = np.max(Q_list)
+        action_list = np.where(Q_list == Q_max)[0]  
+        action = np.random.choice(action_list) # Q_max可能对应多个 action ，可以随机抽取一个
+        return action
+            
+    def update(self, state, action, reward, next_state, done):
+        Q_predict = self.Q_table[state, action]
+        if done:
+            Q_target = reward  # 没有下一个状态了
+        else:
+            Q_target = reward + self.gamma * np.max(
+                self.Q_table[next_state, :])  # Q_table-learning
+        self.Q_table[state, action] += self.lr * (Q_target - Q_predict)  # 修正q
+    def save(self,path):
+        np.save(path+"Q_table.npy", self.Q_table)
+    def load(self, path):
+        self.Q_table = np.load(path+"Q_table.npy")
+
+

codes/QLearning/task0_eval.py

@@ -0,0 +1,84 @@
+#!/usr/bin/env python
+# coding=utf-8
+'''
+Author: John
+Email: johnjim0816@gmail.com
+Date: 2020-09-11 23:03:00
+LastEditor: John
+LastEditTime: 2021-04-29 17:01:43
+Discription: 
+Environment: 
+'''
+import sys,os
+curr_path = os.path.dirname(__file__)
+parent_path=os.path.dirname(curr_path) 
+sys.path.append(parent_path) # add current terminal path to sys.path
+
+import gym
+import datetime
+
+from envs.gridworld_env import CliffWalkingWapper
+from QLearning.agent import QLearning
+from common.plot import plot_rewards
+from common.utils import save_results
+
+curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S") # obtain current time
+
+class QlearningConfig:
+    '''训练相关参数'''
+    def __init__(self):
+        self.algo = 'Qlearning'
+        self.env = 'CliffWalking-v0' # 0 up, 1 right, 2 down, 3 left
+        self.result_path = curr_path+"/outputs/" +self.env+'/'+curr_time+'/results/'  # path to save results
+        self.model_path = curr_path+"/outputs/" +self.env+'/'+curr_time+'/models/'  # path to save models
+        self.train_eps = 300 # 训练的episode数目
+        self.eval_eps = 30
+        self.gamma = 0.9 # reward的衰减率
+        self.epsilon_start = 0.95 # e-greedy策略中初始epsilon
+        self.epsilon_end = 0.01 # e-greedy策略中的终止epsilon
+        self.epsilon_decay = 200 # e-greedy策略中epsilon的衰减率
+        self.lr = 0.1 # learning rate
+
+def env_agent_config(cfg,seed=1):
+    env = gym.make(cfg.env)  
+    env = CliffWalkingWapper(env)
+    env.seed(seed)
+    state_dim = env.observation_space.n
+    action_dim = env.action_space.n
+    agent = QLearning(state_dim,action_dim,cfg)
+    return env,agent
+    
+def eval(cfg,env,agent):
+    # env = gym.make("FrozenLake-v0", is_slippery=False)  # 0 left, 1 down, 2 right, 3 up
+    # env = FrozenLakeWapper(env)
+    rewards = []  # 记录所有episode的reward
+    ma_rewards = [] # 滑动平均的reward
+    for i_ep in range(cfg.eval_eps):
+        ep_reward = 0  # 记录每个episode的reward
+        state = env.reset()  # 重置环境, 重新开一局（即开始新的一个episode）
+        while True:
+            action = agent.predict(state)  # 根据算法选择一个动作
+            next_state, reward, done, _ = env.step(action)  # 与环境进行一个交互
+            state = next_state  # 存储上一个观察值
+            ep_reward += reward
+            if done:
+                break
+        rewards.append(ep_reward)
+        if ma_rewards:
+            ma_rewards.append(ma_rewards[-1]*0.9+ep_reward*0.1)
+        else:
+            ma_rewards.append(ep_reward)
+        print(f"Episode:{i_ep+1}/{cfg.eval_eps}, reward:{ep_reward:.1f}")
+    return rewards,ma_rewards
+
+if __name__ == "__main__":
+    cfg = QlearningConfig()
+    env,agent = env_agent_config(cfg,seed=15)
+    cfg.model_path = './'+'QLearning/outputs/CliffWalking-v0/20210429-165825/models'+'/'
+    cfg.result_path = './'+'QLearning/outputs/CliffWalking-v0/20210429-165825/results'+'/'
+    agent.load(path=cfg.model_path)
+    rewards,ma_rewards = eval(cfg,env,agent)
+    save_results(rewards,ma_rewards,tag='eval',path=cfg.result_path)
+    plot_rewards(rewards,ma_rewards,tag="eval",env=cfg.env,algo = cfg.algo,path=cfg.result_path)
+    
+    

codes/QLearning/task0_train.py

@@ -5,11 +5,10 @@ Author: John
 Email: johnjim0816@gmail.com
 Date: 2020-09-11 23:03:00
 LastEditor: John
-LastEditTime: 2021-03-31 18:14:59
+LastEditTime: 2021-04-29 17:01:08
 Discription: 
 Environment: 
 '''
-
 import sys,os
 curr_path = os.path.dirname(__file__)
 parent_path=os.path.dirname(curr_path) 
@@ -18,40 +17,41 @@ sys.path.append(parent_path) # add current terminal path to sys.path
 import gym
 import datetime
 
-from envs.gridworld_env import CliffWalkingWapper, FrozenLakeWapper
+from envs.gridworld_env import CliffWalkingWapper
 from QLearning.agent import QLearning
 from common.plot import plot_rewards
-from common.utils import save_results
-
-SEQUENCE = datetime.datetime.now().strftime("%Y%m%d-%H%M%S") # obtain current time
-SAVED_MODEL_PATH = curr_path+"/saved_model/"+SEQUENCE+'/' # path to save model
-if not os.path.exists(curr_path+"/saved_model/"): 
-    os.mkdir(curr_path+"/saved_model/")
-if not os.path.exists(SAVED_MODEL_PATH): 
-    os.mkdir(SAVED_MODEL_PATH)
-RESULT_PATH = curr_path+"/results/"+SEQUENCE+'/' # path to save rewards
-if not os.path.exists(curr_path+"/results/"): 
-    os.mkdir(curr_path+"/results/")
-if not os.path.exists(RESULT_PATH): 
-    os.mkdir(RESULT_PATH)
+from common.utils import save_results,make_dir
+curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S") # obtain current time
 
 class QlearningConfig:
     '''训练相关参数'''
     def __init__(self):
-        self.train_eps = 200 # 训练的episode数目
+        self.algo = 'Qlearning'
+        self.env = 'CliffWalking-v0' # 0 up, 1 right, 2 down, 3 left
+        self.result_path = curr_path+"/outputs/" +self.env+'/'+curr_time+'/results/'  # path to save results
+        self.model_path = curr_path+"/outputs/" +self.env+'/'+curr_time+'/models/'  # path to save models
+        self.train_eps = 300 # 训练的episode数目
+        self.eval_eps = 30
         self.gamma = 0.9 # reward的衰减率
-        self.epsilon_start = 0.99 # e-greedy策略中初始epsilon
+        self.epsilon_start = 0.95 # e-greedy策略中初始epsilon
         self.epsilon_end = 0.01 # e-greedy策略中的终止epsilon
         self.epsilon_decay = 200 # e-greedy策略中epsilon的衰减率
         self.lr = 0.1 # learning rate
+        
+def env_agent_config(cfg,seed=1):
+    env = gym.make(cfg.env)  
+    env = CliffWalkingWapper(env)
+    env.seed(seed)
+    state_dim = env.observation_space.n
+    action_dim = env.action_space.n
+    agent = QLearning(state_dim,action_dim,cfg)
+    return env,agent
 
 def train(cfg,env,agent):
     rewards = []  
     ma_rewards = [] # moving average reward
-    steps = []  # 记录所有episode的steps
-    for i_episode in range(cfg.train_eps):
+    for i_ep in range(cfg.train_eps):
         ep_reward = 0  # 记录每个episode的reward
-        ep_steps = 0  # 记录每个episode走了多少step
         state = env.reset()  # 重置环境, 重新开一局（即开始新的一个episode）
         while True:
             action = agent.choose_action(state)  # 根据算法选择一个动作
@@ -59,55 +59,52 @@ def train(cfg,env,agent):
             agent.update(state, action, reward, next_state, done)  # Q-learning算法更新
             state = next_state  # 存储上一个观察值
             ep_reward += reward
-            ep_steps += 1  # 计算step数
             if done:
                 break
-        steps.append(ep_steps)
         rewards.append(ep_reward)
         if ma_rewards:
             ma_rewards.append(ma_rewards[-1]*0.9+ep_reward*0.1)
         else:
             ma_rewards.append(ep_reward)
-        print("Episode:{}/{}: reward:{:.1f}".format(i_episode+1, cfg.train_eps,ep_reward))
+        print("Episode:{}/{}: reward:{:.1f}".format(i_ep+1, cfg.train_eps,ep_reward))
     return rewards,ma_rewards
-
+    
 def eval(cfg,env,agent):
     # env = gym.make("FrozenLake-v0", is_slippery=False)  # 0 left, 1 down, 2 right, 3 up
     # env = FrozenLakeWapper(env)
     rewards = []  # 记录所有episode的reward
     ma_rewards = [] # 滑动平均的reward
-    steps = []  # 记录所有episode的steps
-    for i_episode in range(cfg.train_eps):
+    for i_ep in range(cfg.eval_eps):
         ep_reward = 0  # 记录每个episode的reward
-        ep_steps = 0  # 记录每个episode走了多少step
         state = env.reset()  # 重置环境, 重新开一局（即开始新的一个episode）
         while True:
-            action = agent.choose_action(state)  # 根据算法选择一个动作
+            action = agent.predict(state)  # 根据算法选择一个动作
             next_state, reward, done, _ = env.step(action)  # 与环境进行一个交互
             state = next_state  # 存储上一个观察值
             ep_reward += reward
-            ep_steps += 1  # 计算step数
             if done:
                 break
-        steps.append(ep_steps)
         rewards.append(ep_reward)
-        # 计算滑动平均的reward
         if ma_rewards:
-            ma_rewards.append(rewards[-1]*0.9+ep_reward*0.1)
+            ma_rewards.append(ma_rewards[-1]*0.9+ep_reward*0.1)
         else:
             ma_rewards.append(ep_reward)
-        print("Episode:{}/{}: reward:{:.1f}".format(i_episode+1, cfg.train_eps,ep_reward))
+        print(f"Episode:{i_ep+1}/{cfg.eval_eps}, reward:{ep_reward:.1f}")
     return rewards,ma_rewards
     
 if __name__ == "__main__":
     cfg = QlearningConfig()
-    env = gym.make("CliffWalking-v0")  # 0 up, 1 right, 2 down, 3 left
-    env = CliffWalkingWapper(env)
-    action_dim = env.action_space.n
-    agent = QLearning(action_dim,cfg)
+    env,agent = env_agent_config(cfg,seed=1)
     rewards,ma_rewards = train(cfg,env,agent)
-    agent.save(path=SAVED_MODEL_PATH)
-    save_results(rewards,ma_rewards,tag='train',path=RESULT_PATH)
-    plot_rewards(rewards,ma_rewards,tag="train",algo = "On-Policy First-Visit MC Control",path=RESULT_PATH)
+    make_dir(cfg.result_path,cfg.model_path)
+    agent.save(path=cfg.model_path)
+    save_results(rewards,ma_rewards,tag='train',path=cfg.result_path)
+    plot_rewards(rewards,ma_rewards,tag="train",env=cfg.env,algo = cfg.algo,path=cfg.result_path)
+
+    env,agent = env_agent_config(cfg,seed=10)
+    agent.load(path=cfg.model_path)
+    rewards,ma_rewards = eval(cfg,env,agent)
+    save_results(rewards,ma_rewards,tag='eval',path=cfg.result_path)
+    plot_rewards(rewards,ma_rewards,tag="eval",env=cfg.env,algo = cfg.algo,path=cfg.result_path)