Repository navigation
Expand file tree
/
Copy pathtrain.py
More file actions
190 lines (154 loc) · 6.96 KB
/
Copy pathtrain.py
File metadata and controls
190 lines (154 loc) · 6.96 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
from QlearningAgent import QLearningAgent, QLearningTrainer
from ApproximateQAgent import ApproximateQAgent, ApproximateQTrainer
from DQN import DQNAgent, DQNTrainer
from MCTS import MCTSAgent
# from new_sim import RandomPlayer, Game
import new_sim as HM
import AgentFight as AG
from Agent_Evaluator import compare_agents_performance, compare_multiple_agents
import matplotlib.pyplot as plt
data_path = r"model_data/"
def train_rl_agent():
rl_agent = QLearningAgent(player_id=0, learning_rate=0.1, epsilon=0.1)
random_opponent = AG.RandomPlayer(player_id=1)
minimax_opponent = AG.MinimaxPlayer(player_id=1, max_depth=2) # 使用Minimax对手进行训练
trainer = QLearningTrainer(rl_agent, random_opponent)
trainer.train(episodes=100, save_interval=100)
trainer.train(episodes=100, opponent_agent=minimax_opponent, save_interval=100)
return rl_agent
def train_approximate_q_agent():
approx_agent = ApproximateQAgent(
player_id=0,
learning_rate=0.001, # 降低学习率
epsilon=0.9, # 初始探索率高一些
epsilon_min=0.01, # 最小探索率
epsilon_decay=0.995, # 基础衰减率
discount_factor=0.99 # 提高未来奖励权重
)
random_opponent = AG.RandomPlayer(player_id=1)
minimax_opponent = AG.MinimaxPlayer(player_id=1, max_depth=2)
trainer = ApproximateQTrainer(approx_agent, random_opponent)
trainer.train(episodes=100, save_interval=100)
trainer.train(episodes=100, opponent_agent=minimax_opponent, save_interval=100)
return approx_agent
def train_dqn_agent(): # 这里用 id=1 训练了，记得保持一致
dqn_agent = DQNAgent(
player_id=0,
learning_rate=1e-4,
epsilon=0.9, # 初始探索率
epsilon_decay=0.997, # 更缓慢的衰减
epsilon_min=0.05, # 更高的最小探索率
batch_size=128,
memory_size=50000
)
# try:
# dqn_agent.load_model(data_path + "final_dqn_model.pth")
# print("成功加载DQN模型")
# except:
# print("DQN模型加载失败, 使用未训练版本")
# dqn_agent = DQNAgent(
# player_id=0,
# learning_rate=1e-4,
# epsilon=0.9, # 初始探索率
# epsilon_decay=0.997, # 更缓慢的衰减
# epsilon_min=0.05, # 更高的最小探索率
# batch_size=128,
# memory_size=50000
# )
random_opponent = AG.RandomPlayer(player_id=1)
minimax_opponent = AG.MinimaxPlayer(player_id=1, max_depth=2) # 使用Minimax对手进行训练
trainer = DQNTrainer(dqn_agent, random_opponent)
trainer.train(episodes=1000, opponent_agent=random_opponent, save_interval=100)
trainer.train(episodes=200, opponent_agent=minimax_opponent, save_interval=100)
trainer.train(episodes=100, opponent_agent=random_opponent, save_interval=100)
trainer.train(episodes=200, opponent_agent=minimax_opponent, save_interval=100)
return dqn_agent
def play_with_trained_agent():
rl_agent = QLearningAgent(player_id=1, epsilon=0.0) # 测试时不探索
rl_agent.load_q_table("model_data/final_q_table.pkl")
game = HM.Game(rl_agent)
game.run()
def play_with_approximate_agent():
approx_agent = ApproximateQAgent(player_id=1, epsilon=0.0) # 测试时不探索
approx_agent.load_model("model_data/final_aq_model.pth")
# 修改游戏设置
game = HM.Game(approx_agent)
game.run()
def play_with_dqn_agent():
"""与训练好的DQN智能体对战"""
print("加载训练好的DQN智能体...")
dqn_agent = DQNAgent(player_id=1, epsilon=0.0) # 测试时不探索
dqn_agent.load_model("final_dqn_model.pth")
# 修改游戏设置
game = HM.Game(dqn_agent)
game.run()
def play_with_mcts_agent():
"""与MCTS智能体对战"""
print("加载MCTS智能体...")
mcts_agent = MCTSAgent(player_id=1, simulation_limit=10, exploration_constant=1.414)
# 修改游戏设置
game = HM.Game(mcts_agent)
game.run()
def test_rl_agent():
rl_agent = QLearningAgent(player_id=1, epsilon=0.0) # 测试时不探索
rl_agent.load_q_table("model_data/final_q_table.pkl")
game = AG.Game(rl_agent, display=False)
game.run()
def test_approximate_agent():
approx_agent = ApproximateQAgent(player_id=1, epsilon=0.0) # 测试时不探索
approx_agent.load_model("model_data/final_aq_model.pth")
game = AG.Game(approx_agent, display=False)
game.run()
def test_dqn_agent():
dqn_agent = DQNAgent(player_id=1, epsilon=0.0) # 测试时不探索
dqn_agent.load_model("final_dqn_model.pth")
game = AG.Game(dqn_agent, display=False)
game.run()
def evaluate_trained_agents():
"""评估训练好的智能体"""
# 加载训练好的Q-Learning智能体
# rl_agent = QLearningAgent(player_id=0, epsilon=0.0)
# try:
# rl_agent.load_q_table(data_path + "final_q_table.pkl")
# print("成功加载Q-Learning模型")
# except:
# print("Q-Learning模型加载失败, 使用未训练版本")
# 加载训练好的Approximate Q智能体
# approx_agent = ApproximateQAgent(player_id=1, epsilon=0.0) # 训练的时候就记得用 id=1 训练
# try:
# approx_agent.load_model(data_path + "final_aq_model.pkl")
# print("成功加载Approximate Q模型")
# except:
# print("Approximate Q模型加载失败, 使用未训练版本")
dqn_agent = DQNAgent(player_id=0, epsilon=0.0) # 测试时不探索
try:
dqn_agent.load_model(data_path + "final_dqn_model.pth")
print("成功加载DQN模型")
except:
print("DQN模型加载失败, 使用未训练版本")
minimax_agent_0 = AG.MinimaxPlayer(player_id=0, max_depth=2)
minimax_agent_1 = AG.MinimaxPlayer(player_id=1, max_depth=2)
# 创建随机对手
random_agent_0 = AG.RandomPlayer(player_id=0)
random_agent_1 = AG.RandomPlayer(player_id=1)
# 1. Q-Learning vs Random
# print("\n1. Q-Learning vs Random:")
# compare_agents_performance(rl_agent, random_agent_1, 1000, "Q-Learning", "Random")
# 2. Approximate Q vs Random
# print("\n2. Approximate Q vs Random:")
# compare_agents_performance(random_agent_0, approx_agent, 1000, "Approximate Q", "Random")
# 3. Q-Learning vs Approximate Q
# print("\n3. Q-Learning vs Approximate Q:")
# compare_agents_performance(rl_agent, approx_agent, 1000, "Q-Learning", "Approximate Q")
compare_agents_performance(dqn_agent, random_agent_1, 100, "DQN", "Random")
compare_agents_performance(dqn_agent, minimax_agent_1, 100, "DQN", "Minimax")
if __name__ == "__main__":
# train_rl_agent()
# train_approximate_q_agent()
train_dqn_agent()
# play_with_trained_agent()
# play_with_mcts_agent()
# test_rl_agent()
# test_approximate_agent()
# test_dqn_agent()
# evaluate_trained_agents()