import akshare as ak import numpy as np import pandas as pd import random import matplotlib.pyplot as plt class StockTradingEnv: def __init__(self): self.df = ak.stock_zh_a_daily(symbol='sh000001', adjust="qfq").iloc[::-1] self.observation_space = self.df.shape[1] self.action_space = 3 self.reset() def reset(self): self.current_step = 0 self.total_profit = 0 self.done = False self.state = self.df.iloc[self.current_step].values return self.state def step(self, action): assert self.action_space.contains(action) if action == 0: # 买入 self.buy_stock() elif action == 1: # 卖出 self.sell_stock() else: # 保持不变 pass self.current_step += 1 if self.current_step >= len(self.df) - 1: self.done = True else: self.state = self.df.iloc[self.current_step].values reward = self.get_reward() self.total_profit += reward return self.state, reward, self.done, {} def buy_stock(self): pass def sell_stock(self): pass def get_reward(self): pass class QLearningAgent: def __init__(self, state_size, action_size): self.state_size = state_size self.action_size = action_size self.epsilon = 1.0 self.epsilon_min = 0.01 self.epsilon_decay = 0.995 self.learning_rate = 0.1 self.discount_factor = 0.99 self.q_table = np.zeros((self.state_size, self.action_size)) def act(self, state): if np.random.rand() <= self.epsilon: return random.randrange(self.action_size) else: return np.argmax(self.q_table[state, :]) def learn(self, state, action, reward, next_state, done): target = reward + self.discount_factor * np.max(self.q_table[next_state, :]) self.q_table[state, action] = (1 - self.learning_rate) * self.q_table[state, action] + self.learning_rate * target if self.epsilon > self.epsilon_min: self.epsilon *= self.epsilon_decay env = StockTradingEnv() agent = QLearningAgent(env.observation_space, env.action_space) for episode in range(1000): state = env.reset() done = False while not done: action = agent.act(state) next_state, reward, done, _ = env.step(action) agent.learn(state, action, reward, next_state, done) state = next_state if episode % 10 == 0: print("Episode: %d, Total Profit: %f" % (episode, env.total_profit)) agent.save_model("model-%d.h5" % episode) def plot_profit(env, title): plt.figure(figsize=(12, 6)) plt.plot(env.df.index, env.df.close, label="Price") plt.plot(env.df.index, env.profits, label="Profits") plt.legend() plt.title(title) plt.show() env = StockTradingEnv() agent = QLearningAgent(env.observation_space, env.action_space) agent.load_model("model-100.h5") state = env.reset() done = False while not done: action = agent.act(state) next_state, reward, done, _ = env.step(action) state = next_state plot_profit(env, "QLearning Trading Strategy")优化代码

1. 对于环境类 `StockTradingEnv`，可以考虑将 `buy_stock` 和 `sell_stock` 方法的具体实现写入 `step` 方法中，避免方法数量过多。
2. 可以将 `get_reward` 方法中的具体实现改为直接计算当前持仓的收益。
3. 在循环训练过程中，可以记录每个 episode 的总收益，并将这些数据保存下来，在训练完成后进行可视化分析。
4. 可以添加更多的参数来控制训练过程，比如学习率、衰减系数等。
5. 可以将 QLearningAgent 类中的方法进行整理和封装，提高代码的可读性和可维护性。同时，也可以添加一些对模型进行保存和加载的方法，便于模型的重用和共享。