def run_training(maze_size=(6, 6),
trap_number=1,
epoch=20,
epsilon0=0.3,
alpha=0.3,
gamma=0.9):
# # 可选的参数：
# epoch = 20
# # 随机探索的初始概率
# epsilon0 = 0.3
# # 松弛变量
# alpha = 0.3
# # 折扣因子
# gamma = 0.9
# # 地图大小
# maze_size = (6, 6)
# # 陷阱数量
# trap_number = 1
g = Maze(maze_size=maze_size, trap_number=trap_number)
r = Robot(g, alpha=alpha, epsilon0=epsilon0, gamma=gamma)
r.set_status(learning=True)
runner = Runner(r, g)
runner.run_training(epoch, display_direction=True)
# runner.generate_movie(filename = "final1.mp4") # 你可以注释该行代码，加快运行速度，不过你就无法观察到视频了。
# runner.plot_results()
return runner
def test_different_parameter(alpha_test, gamma_test, epsilon_test, epoch_test):
g = Maze(maze_size=maze_size, trap_number=trap_number)
r = Robot(g, alpha=alpha_test, epsilon0=epsilon_test, gamma=gamma_test)
r.set_status(learning=True)
runner = Runner(r, g)
runner.run_training(epoch_test, display_direction=True)
print("alpha: {}, gamma: {}, epsilon: {}, epoch: {}".format(
alpha_test, gamma_test, epsilon_test, epoch_test))
runner.plot_results()
def train_by_dqn_robot(times, maze_size=5):
print("start times:", times)
maze = Maze(maze_size=maze_size)
"""choose Keras or Torch version"""
robot = KerasRobot(maze=maze)
# robot = TorchRobot(maze=maze)
robot.memory.build_full_view(maze=maze)
"""training by runner"""
runner = Runner(robot=robot)
runner.run_training(15, 75)
"""Test Robot"""
robot.reset()
for _ in range(25):
a, r = robot.test_update()
if r < -20:
print(
"SUCCESSFUL!",
"| TIMES:",
times,
)
break
#!/usr/bin/env python3 # -*- coding: utf-8 -*- """ Created on Fri Oct 5 14:12:30 2018 @author: htaiwan """ from Maze import Maze from Robot import Robot from Runner import Runner epoch = 20 epsilon0 = 0.7 alpha = 0.5 gamma = 0.9 maze_size = (6,6) trap_number = 1 maze = Maze(maze_size=maze_size, trap_number=trap_number) robot = Robot(maze, alpha=alpha, epsilon0=epsilon0, gamma=gamma) robot.set_status(learning=True) runner = Runner(robot, maze) runner.run_training(epoch, display_direction=True) #runner.generate_movie(filename = "final.avi") # 你可以注释该行代码，加快运行速度，不过你就无法观察到视频了。 runner.plot_results()
print("the returned reward: ", action)
'''
""" Qlearning 算法相关参数： """
epoch = 20 # 训练轮数
epsilon0 = 1 # 初始探索概率
alpha = 0.5 # 公式中的 ⍺
gamma = 0.94 # 公式中的 γ
maze_size = 11 # 迷宫size
""" 使用 QLearning 算法训练过程 """
g = Maze(maze_size=maze_size)
r = QRobot(g, alpha=alpha, epsilon0=epsilon0, gamma=gamma)
runner = Runner(r)
runner.run_training(epoch, training_per_epoch=int(maze_size * maze_size * 1.5))
# 生成训练过程的gif图, 建议下载到本地查看；也可以注释该行代码，加快运行速度。
# runner.generate_gif(filename="results/size5.gif")
runner.plot_results() # 输出训练结果，可根据该结果对您的机器人进行分析。
'''
test_memory = ReplayDataSet(max_size=1e3) # 初始化并设定最大容量
actions = ['u', 'r', 'd', 'l']
test_memory.add((0,1), actions.index("r"), -10, (0,1), 1) # 添加一条数据（state, action_index, reward, next_state）
print(test_memory.random_sample(1)) # 从中随机抽取一条（因为只有一条数据）
'''
'''
os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE" # 允许重复载入lib文件
maze = Maze(maze_size=5)
from torch_py.MinDQNRobot import MinDQNRobot as Robot # PyTorch版本
""" Deep Qlearning 算法相关参数： """
epoch = 20 # 训练轮数
maze_size = 5 # 迷宫size
training_per_epoch = int(maze_size * maze_size * 2)
""" 使用 DQN 算法训练 """
maze = Maze(maze_size=maze_size)
robot = Robot(maze)
print(robot.maze.reward) # 输出最小值选择策略的reward值
"""开启金手指，获取全图视野"""
robot.memory.build_full_view(maze=maze)
runner = Runner(robot)
runner.run_training(epoch, training_per_epoch)
runner.plot_results()
# """Test Robot"""
# robot.reset()
# for _ in range(25):
# a, r = robot.test_update()
# print("action:", a, "reward:", r)
# if r == maze.reward["destination"]:
# print("success")
# break
# 生成训练过程的gif图, 建议下载到本地查看；也可以注释该行代码，加快运行速度。
# runner.generate_gif(filename="results/dqn_size10.gif")
# %%