PDF下载 下载

机器学习(股票)

阅读 8415
  1. # coding=utf-8
  2. from __future__ import print_function, absolute_import, unicode_literals
  3. from datetime import datetime
  4. import numpy as np
  5. from gm.api import *
  6. import sys
  7. try:
  8. from sklearn import svm
  9. except:
  10. print('请安装scikit-learn库和带mkl的numpy')
  11. sys.exit(-1)
  12. '''
  13. 本策略选取了七个特征变量组成了滑动窗口长度为15天的训练集,随后训练了一个二分类(上涨/下跌)的支持向量机模型.
  14. 若没有仓位则在每个星期一的时候输入标的股票近15个交易日的特征变量进行预测,并在预测结果为上涨的时候购买标的.
  15. 若已经持有仓位则在盈利大于10%的时候止盈,在星期五损失大于2%的时候止损.
  16. 特征变量为:1.收盘价/均值2.现量/均量3.最高价/均价4.最低价/均价5.现量6.区间收益率7.区间标准差
  17. 训练数据为:SHSE.600000浦发银行,时间从2016-03-01到2017-06-30
  18. 回测时间为:2017-07-01 09:00:00到2017-10-01 09:00:00
  19. '''
  20. def init(context):
  21. # 订阅浦发银行的分钟bar行情
  22. context.symbol = 'SHSE.600000'
  23. subscribe(symbols=context.symbol, frequency='60s')
  24. start_date = '2016-03-01' # SVM训练起始时间
  25. end_date = '2017-06-30' # SVM训练终止时间
  26. # 用于记录工作日
  27. # 获取目标股票的daily历史行情
  28. recent_data = history(context.symbol, frequency='1d', start_time=start_date, end_time=end_date, fill_missing='last',
  29. df=True)
  30. days_value = recent_data['bob'].values
  31. days_close = recent_data['close'].values
  32. days = []
  33. # 获取行情日期列表
  34. print('准备数据训练SVM')
  35. for i in range(len(days_value)):
  36. days.append(str(days_value[i])[0:10])
  37. x_all = []
  38. y_all = []
  39. for index in range(15, (len(days) - 5)):
  40. # 计算三星期共15个交易日相关数据
  41. start_day = days[index - 15]
  42. end_day = days[index]
  43. data = history(context.symbol, frequency='1d', start_time=start_day, end_time=end_day, fill_missing='last',
  44. df=True)
  45. close = data['close'].values
  46. max_x = data['high'].values
  47. min_n = data['low'].values
  48. amount = data['amount'].values
  49. volume = []
  50. for i in range(len(close)):
  51. volume_temp = amount[i] / close[i]
  52. volume.append(volume_temp)
  53. close_mean = close[-1] / np.mean(close) # 收盘价/均值
  54. volume_mean = volume[-1] / np.mean(volume) # 现量/均量
  55. max_mean = max_x[-1] / np.mean(max_x) # 最高价/均价
  56. min_mean = min_n[-1] / np.mean(min_n) # 最低价/均价
  57. vol = volume[-1] # 现量
  58. return_now = close[-1] / close[0] # 区间收益率
  59. std = np.std(np.array(close), axis=0) # 区间标准差
  60. # 将计算出的指标添加到训练集X
  61. # features用于存放因子
  62. features = [close_mean, volume_mean, max_mean, min_mean, vol, return_now, std]
  63. x_all.append(features)
  64. # 准备算法需要用到的数据
  65. for i in range(len(days_close) - 20):
  66. if days_close[i + 20] > days_close[i + 15]:
  67. label = 1
  68. else:
  69. label = 0
  70. y_all.append(label)
  71. x_train = x_all[: -1]
  72. y_train = y_all[: -1]
  73. # 训练SVM
  74. context.clf = svm.SVC(C=1.0, kernel='rbf', degree=3, gamma='auto', coef0=0.0, shrinking=True, probability=False,
  75. tol=0.001, cache_size=200, verbose=False, max_iter=-1,
  76. decision_function_shape='ovr', random_state=None)
  77. context.clf.fit(x_train, y_train)
  78. print('训练完成!')
  79. def on_bar(context, bars):
  80. bar = bars[0]
  81. # 获取当前年月日
  82. today = bar.bob.strftime('%Y-%m-%d')
  83. # 获取数据并计算相应的因子
  84. # 于星期一的09:31:00进行操作
  85. # 当前bar的工作日
  86. weekday = datetime.strptime(today, '%Y-%m-%d').isoweekday()
  87. # 获取模型相关的数据
  88. # 获取持仓
  89. position = context.account().position(symbol=context.symbol, side=PositionSide_Long)
  90. # 如果bar是新的星期一且没有仓位则开始预测
  91. if not position and weekday == 1:
  92. # 获取预测用的历史数据
  93. data = history_n(symbol=context.symbol, frequency='1d', end_time=today, count=15,
  94. fill_missing='last', df=True)
  95. close = data['close'].values
  96. train_max_x = data['high'].values
  97. train_min_n = data['low'].values
  98. train_amount = data['amount'].values
  99. volume = []
  100. for i in range(len(close)):
  101. volume_temp = train_amount[i] / close[i]
  102. volume.append(volume_temp)
  103. close_mean = close[-1] / np.mean(close)
  104. volume_mean = volume[-1] / np.mean(volume)
  105. max_mean = train_max_x[-1] / np.mean(train_max_x)
  106. min_mean = train_min_n[-1] / np.mean(train_min_n)
  107. vol = volume[-1]
  108. return_now = close[-1] / close[0]
  109. std = np.std(np.array(close), axis=0)
  110. # 得到本次输入模型的因子
  111. features = [close_mean, volume_mean, max_mean, min_mean, vol, return_now, std]
  112. features = np.array(features).reshape(1, -1)
  113. prediction = context.clf.predict(features)[0]
  114. # 若预测值为上涨则开仓
  115. if prediction == 1:
  116. # 获取昨收盘价
  117. context.price = close[-1]
  118. # 把浦发银行的仓位调至95%
  119. order_target_percent(symbol=context.symbol, percent=0.95, order_type=OrderType_Market,
  120. position_side=PositionSide_Long)
  121. print('SHSE.600000以市价单开多仓到仓位0.95')
  122. # 当涨幅大于10%,平掉所有仓位止盈
  123. elif position and bar.close / context.price >= 1.10:
  124. order_close_all()
  125. print('SHSE.600000以市价单全平多仓止盈')
  126. # 当时间为周五并且跌幅大于2%时,平掉所有仓位止损
  127. elif position and bar.close / context.price < 1.02 and weekday == 5:
  128. order_close_all()
  129. print('SHSE.600000以市价单全平多仓止损')
  130. if __name__ == '__main__':
  131. '''
  132. strategy_id策略ID,由系统生成
  133. filename文件名,请与本文件名保持一致
  134. mode实时模式:MODE_LIVE回测模式:MODE_BACKTEST
  135. token绑定计算机的ID,可在系统设置-密钥管理中生成
  136. backtest_start_time回测开始时间
  137. backtest_end_time回测结束时间
  138. backtest_adjust股票复权方式不复权:ADJUST_NONE前复权:ADJUST_PREV后复权:ADJUST_POST
  139. backtest_initial_cash回测初始资金
  140. backtest_commission_ratio回测佣金比例
  141. backtest_slippage_ratio回测滑点比例
  142. '''
  143. run(strategy_id='strategy_id',
  144. filename='main.py',
  145. mode=MODE_BACKTEST,
  146. token='token_id',
  147. backtest_start_time='2017-07-01 09:00:00',
  148. backtest_end_time='2017-10-01 09:00:00',
  149. backtest_adjust=ADJUST_PREV,
  150. backtest_initial_cash=10000000,
  151. backtest_commission_ratio=0.0001,
  152. backtest_slippage_ratio=0.0001)

机器学习(股票)

0 篇笔记