def train(self, total_games=500):
fixed_epsilon = None
alpha = 0.5
gamma = 0.9
epsilon = 1
agent = QLearningAgent(epsilon=epsilon, fixed_epsilon=fixed_epsilon, alpha=alpha, gamma=gamma, total_games=total_games)
for game_number in range(total_games):
last_action = None
while True:
game_result = game.game_results()
if game_result is not None:
pygame.display.flip()
game.reset()
game.initialize()
break
last_action = game.play(agent)
pygame.display.flip()
agent.save_policy()
return agent
from tqdm import tqdm
def q(text=''):
print(f'>{text}<')
sys.exit()
from environment import TicTacToe
from agent import QLearningAgent, Hoooman
import config as cfg
from config import display_board
# initializing the TicTacToe environment and a QLearningAgent (the master Tic-Toc-Toc player, your opponent!)
env = TicTacToe()
player1 = QLearningAgent(name=cfg.playerX_QLearningAgent_name)
player1.loadQtable() # load the learnt Q-Table
player1.epsilon = 0.0 # greedy actions only, 0 exploration
# initializing the agent class that let's you, the human user take the actions in the game
player2 = Hoooman()
# replay decides whether to rematch or not, at the end of a game
replay = True
while replay:
done = False # the episode goes on as long as done is False
# deciding which player makes a move first
playerID = random.choice([True, False]) # True means player1
if __name__ == '__main__':
episode = 30 # 训练多少回合
epsilon = 0.8 # 使用历史经验的概率, 若值为0.9,则有 90% 的情况下,会根据历史经验选择 action, 10% 的情况下,随机选择 action
learning_rate = 0.01 # 根据公式可知,该值越大,则旧训练数据被保留的就越少
discount_factor = 0.9 #
from maze_game.game import startGame
env = startGame()
key = input('Do you want to see the training process? [y] ')
key = key == 'y' or key == ''
agent = QLearningAgent(
epsilon=epsilon,
learning_rate=learning_rate,
discount_factor=discount_factor,
actions=Game.DIRECTION.ACTIONS
)
successful_step_counter_arr = []
failed_step_counter_arr = []
if key:
env.display()
for eps in range(1, episode + 1):
cur_state = env.reset()
step_counter = 0
while True:
step_counter += 1
import numpy as np
import pandas as pd
from env import TicTacToeEnv
from agent import QLearningAgent
env = TicTacToeEnv()
agent = QLearningAgent(env)
for game_nr in range(1000000):
if game_nr % 10000 == 0:
print(game_nr)
done = False
s = env.reset().copy()
# print('Init', s)
while not done:
a = agent.take_action(s)
r, s_, done, _ = env.step(a)
agent.learn(s, a, r, s_, done)
# print(s, a, r, s_, done)
s = s_.copy()
V = pd.DataFrame.from_dict(agent._V,
orient='index',
dtype=np.float32,
columns=['V'])
N = pd.DataFrame.from_dict(agent._N,
orient='index',
dtype=np.uint32,
columns=['N'])
df = V.merge(N, how='left', left_index=True, right_index=True)
states = pd.DataFrame(df.index.values.tolist(), index=df.index)