324 lines
15 KiB
Python
324 lines
15 KiB
Python
# -*- encoding:utf-8 -*-
|
||
"""
|
||
择时具体工作者，整合金融时间序列，买入因子，卖出因子，资金类进行
|
||
择时操作，以时间驱动择时事件的发生
|
||
"""
|
||
|
||
from __future__ import absolute_import
|
||
from __future__ import division
|
||
from __future__ import print_function
|
||
|
||
import copy
|
||
|
||
import numpy as np
|
||
|
||
from ..MarketBu import ABuSymbolPd
|
||
from ..FactorBuyBu.ABuFactorBuyBase import AbuFactorBuyBase
|
||
from ..FactorSellBu.ABuFactorSellBase import AbuFactorSellBase
|
||
from .ABuPickBase import AbuPickTimeWorkBase
|
||
# noinspection PyUnresolvedReferences
|
||
from ..CoreBu.ABuFixes import filter
|
||
from ..UtilBu.ABuProgress import AbuMulPidProgress
|
||
|
||
__author__ = '阿布'
|
||
__weixin__ = 'abu_quant'
|
||
|
||
"""
|
||
是否使用自然周，自然月，默认开启，如需关闭使用下面代码：
|
||
abupy.alpha.pick_time_worker.g_natural_long_task = False
|
||
"""
|
||
g_natural_long_task = True
|
||
|
||
|
||
# noinspection PyAttributeOutsideInit
|
||
class AbuPickTimeWorker(AbuPickTimeWorkBase):
|
||
"""择时类"""
|
||
|
||
def __init__(self, cap, kl_pd, benchmark, buy_factors, sell_factors):
|
||
"""
|
||
:param cap: 资金类AbuCapital实例化对象
|
||
:param kl_pd: 择时时间段交易数据
|
||
:param benchmark: 交易基准对象，AbuBenchmark实例对象
|
||
:param buy_factors: 买入因子序列，序列中的对象为dict，每一个dict针对一个具体因子
|
||
:param sell_factors: 卖出因子序列，序列中的对象为dict，每一个dict针对一个具体因子
|
||
"""
|
||
self.capital = cap
|
||
# 回测阶段kl
|
||
self.kl_pd = kl_pd
|
||
# 合并加上回测之前1年的数据，为了生成特征数据
|
||
self.combine_kl_pd = ABuSymbolPd.combine_pre_kl_pd(self.kl_pd, n_folds=1)
|
||
# 如特别在乎效率性能，打开下面注释的方式，只在g_enable_ml_feature模式下开启, 注释上一行
|
||
# self.combine_kl_pd = ABuSymbolPd.combine_pre_kl_pd(self.kl_pd,
|
||
# n_folds=1) if ABuEnv.g_enable_ml_feature else None
|
||
# 传递给因子系列，因子内部可有选择性使用
|
||
self.benchmark = benchmark
|
||
# 初始化买入因子列表
|
||
self.init_buy_factors(buy_factors)
|
||
# 初始化卖出因子列表
|
||
self.init_sell_factors(sell_factors)
|
||
# 根据因子是否支持周，月任务属性，筛选周月任务因子对象列表，在初始化时做，提高时间驱动效率
|
||
self.filter_long_task_factors()
|
||
# 择时最终买入卖出行为列表，列表中每一个对象都为AbuOrder对象
|
||
self.orders = list()
|
||
# 择时进度条，默认空, 即不打开，不显示择时进度
|
||
self.task_pg = None
|
||
|
||
def __str__(self):
|
||
"""打印对象显示：买入因子列表＋卖出因子列表"""
|
||
return 'buy_factors:{}\nsell_factors:{}'.format(self.buy_factors, self.sell_factors)
|
||
|
||
__repr__ = __str__
|
||
|
||
def enable_task_pg(self):
|
||
"""启动择时内部任务进度条"""
|
||
if self.kl_pd is not None and hasattr(self.kl_pd, 'name') and len(self.kl_pd) > 120:
|
||
self.task_pg = AbuMulPidProgress(len(self.kl_pd), 'pick {} times'.format(self.kl_pd.name))
|
||
self.task_pg.init_ui_progress()
|
||
self.task_pg.display_step = 42
|
||
|
||
def _week_task(self, today):
|
||
"""
|
||
周任务：使用self.week_buy_factors，self.week_sell_factors进行迭代
|
||
不需再使用hasattr进行是否支持判断fit_week
|
||
"""
|
||
# 优先执行买入择时因子专属卖出择时因子，而且即使买入因子被锁但附属于买入因子的卖出不能锁
|
||
self._task_attached_sell(today, how='week')
|
||
|
||
# 周任务中不建议生成买单，执行卖单，全部在日任务完成，如需判断通过today.exec_week，today.exec_month
|
||
for sell_factor in self.week_sell_factors:
|
||
sell_factor.fit_week(today, self.orders)
|
||
|
||
# 执行买入择时因子专属选股因子，决策是否封锁择时买入因子，注意需要从self.buy_factors遍历不是self.week_buy_factors
|
||
self._task_attached_ps(today, is_week=True)
|
||
|
||
for buy_factor in self.week_buy_factors:
|
||
if not buy_factor.lock_factor:
|
||
# 如果买入因子没有被封锁执行任务
|
||
buy_factor.fit_week(today)
|
||
|
||
def _month_task(self, today):
|
||
"""
|
||
月任务：使用self.month_buy_factors，self.month_sell_factors进行迭代
|
||
不需再使用hasattr进行是否支持判断fit_month
|
||
"""
|
||
# 优先执行买入择时因子专属卖出择时因子，而且即使买入因子被锁但附属于买入因子的卖出不能锁
|
||
self._task_attached_sell(today, how='month')
|
||
|
||
# 月任务中不建议生成买单，执行卖单，全部在日任务完成，如需判断通过today.exec_week，today.exec_month
|
||
for sell_factor in self.month_sell_factors:
|
||
sell_factor.fit_month(today, self.orders)
|
||
|
||
# 执行择时因子专属选股因子，决策是否封锁择时买入因子，注意需要从self.buy_factors遍历不是self.month_buy_factors
|
||
self._task_attached_ps(today, is_week=False)
|
||
|
||
# 执行带有fit_month的择时买入因子
|
||
for buy_factor in self.month_buy_factors:
|
||
if not buy_factor.lock_factor:
|
||
# 如果买入因子没有被封锁执行任务
|
||
buy_factor.fit_month(today)
|
||
|
||
def _day_task(self, today):
|
||
"""
|
||
日任务：迭代买入卖出因子序列进行择时
|
||
:param today: 今日的交易数据
|
||
:return:
|
||
"""
|
||
# 优先执行买入择时因子专属卖出择时因子，不受买入因子是否被锁的影响
|
||
self._task_attached_sell(today, how='day')
|
||
|
||
# 注意回测模式下始终非高频，非当日买卖，不区分美股，A股市场，卖出因子要先于买入因子的执行
|
||
for sell_factor in self.sell_factors:
|
||
# 迭代卖出因子，每个卖出因子针对今日交易数据，已经所以交易单进行择时
|
||
sell_factor.read_fit_day(today, self.orders)
|
||
|
||
# 买入因子行为要在卖出因子下面，否则为高频日交易模式
|
||
for buy_factor in self.buy_factors:
|
||
# 如果择时买入因子没有被封锁执行任务
|
||
if not buy_factor.lock_factor:
|
||
# 迭代买入因子，每个因子都对今天进行择时，如果生成order加入self.orders
|
||
order = buy_factor.read_fit_day(today)
|
||
if order and order.order_deal:
|
||
self.orders.append(order)
|
||
|
||
def _task_attached_sell(self, today, how):
|
||
"""专属择时买入因子的择时卖出因子任务：日任务择时卖出因子， 周任务择时卖出因子，月任务择时卖出因子"""
|
||
|
||
for buy_factor in self.buy_factors:
|
||
# 筛选出当前买入因子所对应的所有单子, 注意这里使用buy_factor_class不是buy_factor，buy_factor带参数做为唯一标示
|
||
factor_orders = list(filter(lambda order: order.buy_factor_class == buy_factor.__class__.__name__,
|
||
self.orders))
|
||
if len(factor_orders) == 0:
|
||
# 当前因子没有对应单子
|
||
continue
|
||
|
||
# TODO 不要使用字符串进行eq比对
|
||
for sell_factor in buy_factor.sell_factors:
|
||
if how == 'day':
|
||
# 所有日任务都要用read_fit_day，且一定存在
|
||
sell_factor.read_fit_day(today, factor_orders)
|
||
elif how == 'week' and hasattr(sell_factor, 'fit_week'):
|
||
# 周任务，可选择
|
||
sell_factor.fit_week(today, factor_orders)
|
||
elif how == 'month'and hasattr(sell_factor, 'fit_month'):
|
||
# 月任务，可选择
|
||
sell_factor.fit_month(today, factor_orders)
|
||
|
||
def _task_attached_ps(self, today, is_week):
|
||
"""专属择时买入因子的选股因子任务：周任务选股因子，月任务选股因子"""
|
||
for buy_factor in self.buy_factors:
|
||
# 不能使用today.exec_week或者today.exec_month来判定，因为有可能都时true
|
||
buy_factor.fit_ps_week(today) if is_week else buy_factor.fit_ps_month(today)
|
||
|
||
def _task_loop(self, today):
|
||
"""
|
||
开始时间驱动，进行日任务，周任务，月任务，
|
||
如果使用自然周，就会在每个周五进行择时操作
|
||
自然月在每个月末最后一天进行择时，否则就以
|
||
天数作为触发条件，这个时候定性任务本身的性质
|
||
只是以时间跨度作为阀值，触发条件
|
||
:param today: 对self.kl_pd apply操作，且axis＝1结果为一天的交易数据
|
||
:return:
|
||
"""
|
||
if self.task_pg is not None:
|
||
self.task_pg.show()
|
||
|
||
day_cnt = today.key
|
||
# 判断是否执行周任务, 返回结果赋予today对象
|
||
today.exec_week = today.week_task == 1 if g_natural_long_task else day_cnt % 5 == 0
|
||
# 判断是否执行月任务, 返回结果赋予today对象
|
||
today.exec_month = today.month_task == 1 if g_natural_long_task else day_cnt % 20 == 0
|
||
|
||
if day_cnt == 0 and not today.exec_week:
|
||
# 如果是择时第一天，且没有执行周任务，需要初始化买入因子专属周任务选股池子
|
||
self._task_attached_ps(today, is_week=True)
|
||
if day_cnt == 0 and not today.exec_month:
|
||
# 如果是择时第一天，且没有执行月任务，需要初始化买入因子专属月任务选股池子
|
||
self._task_attached_ps(today, is_week=False)
|
||
|
||
if today.exec_month:
|
||
# 执行因子月任务
|
||
self._month_task(today)
|
||
if today.exec_week:
|
||
# 执行因子周任务
|
||
self._week_task(today)
|
||
# 执行择时因子日任务
|
||
self._day_task(today)
|
||
|
||
# noinspection PyTypeChecker
|
||
def fit(self, *args, **kwargs):
|
||
"""
|
||
根据交易数据，因子等输入数据，拟合择时
|
||
"""
|
||
if g_natural_long_task:
|
||
"""如果要进行自然周，自然月择时任务，需要在kl_pd中添加自然周，自然月标记"""
|
||
# 自然周: 每个周五进行标记
|
||
self.kl_pd['week_task'] = np.where(self.kl_pd.date_week == 4, 1, 0)
|
||
"""
|
||
自然月: 即前后两个日期，相互减，得到的数 > 60 必然为月末，20140801 - 20140731
|
||
没有使用时间api，因为这样做运行效率快
|
||
self.kl_pd.shift(-1)['date'] - self.kl_pd['date']
|
||
->
|
||
>>>>
|
||
2014-07-28 1.0
|
||
2014-07-29 1.0
|
||
2014-07-30 1.0
|
||
2014-07-31 70.0
|
||
2014-08-01 3.0
|
||
2014-08-04 1.0
|
||
2014-08-05 1.0
|
||
>>>
|
||
2014-08-22 3.0
|
||
2014-08-25 1.0
|
||
2014-08-26 1.0
|
||
2014-08-27 1.0
|
||
2014-08-28 1.0
|
||
2014-08-29 73.0
|
||
2014-09-02 1.0
|
||
2014-09-03 1.0
|
||
>>>>
|
||
"""
|
||
self.kl_pd['month_task'] = np.where(self.kl_pd.shift(-1)['date'] - self.kl_pd['date'] > 60, 1, 0)
|
||
# 通过pandas apply进行交易日递进择时
|
||
self.kl_pd.apply(self._task_loop, axis=1)
|
||
|
||
if self.task_pg is not None:
|
||
self.task_pg.close_ui_progress()
|
||
|
||
def init_sell_factors(self, sell_factors):
|
||
"""
|
||
通过sell_factors实例化各个卖出因子
|
||
:param sell_factors: list中元素为dict，每个dict为因子的构造元素，如class，构造参数等
|
||
:return:
|
||
"""
|
||
self.sell_factors = list()
|
||
|
||
if sell_factors is None:
|
||
return
|
||
|
||
for factor_class in sell_factors:
|
||
if factor_class is None:
|
||
continue
|
||
if 'class' not in factor_class:
|
||
# 必须要有需要实例化的类信息
|
||
raise ValueError('factor class key must name class !!!')
|
||
|
||
factor_class_cp = copy.deepcopy(factor_class)
|
||
# pop出类信息后剩下的都为类需要的参数
|
||
class_fac = factor_class_cp.pop('class')
|
||
# 整合capital，kl_pd等实例化因子对象
|
||
factor = class_fac(self.capital, self.kl_pd, self.combine_kl_pd, self.benchmark, **factor_class_cp)
|
||
|
||
if not isinstance(factor, AbuFactorSellBase):
|
||
# 因子对象类型检测
|
||
raise TypeError('factor must base AbuFactorSellBase')
|
||
# 添加到卖出因子序列
|
||
self.sell_factors.append(factor)
|
||
|
||
def init_buy_factors(self, buy_factors):
|
||
"""
|
||
通过buy_factors实例化各个买入因子
|
||
:param buy_factors: list中元素为dict，每个dict为因子的构造元素，如class，构造参数等
|
||
:return:
|
||
"""
|
||
self.buy_factors = list()
|
||
|
||
if buy_factors is None:
|
||
return
|
||
|
||
for factor_class in buy_factors:
|
||
if factor_class is None:
|
||
continue
|
||
|
||
if 'class' not in factor_class:
|
||
# 必须要有需要实例化的类信息
|
||
raise ValueError('factor class key must name class !!!')
|
||
|
||
factor_class_cp = copy.deepcopy(factor_class)
|
||
# pop出类信息后剩下的都为类需要的参数
|
||
class_fac = factor_class_cp.pop('class')
|
||
# 整合capital，kl_pd等实例化因子对象
|
||
factor = class_fac(self.capital, self.kl_pd, self.combine_kl_pd, self.benchmark, **factor_class_cp)
|
||
|
||
if not isinstance(factor, AbuFactorBuyBase):
|
||
# 因子对象类型检测
|
||
raise TypeError('factor must base AbuFactorBuyBase')
|
||
# 添加到买入因子序列
|
||
self.buy_factors.append(factor)
|
||
|
||
def filter_long_task_factors(self):
|
||
"""
|
||
根据每一个因子是否有fit_week筛选周任务因子
|
||
根据每一个因子是否有fit_month筛选月任务因子
|
||
在初始化时完成筛选工作，避免在时间序列中迭代
|
||
不断的进行hasattr判断是否支持
|
||
"""
|
||
self.week_buy_factors = list(filter(lambda buy_factor: hasattr(buy_factor, 'fit_week'),
|
||
self.buy_factors))
|
||
self.month_buy_factors = list(filter(lambda buy_factor: hasattr(buy_factor, 'fit_month'),
|
||
self.buy_factors))
|
||
|
||
self.week_sell_factors = list(filter(lambda sell_factor: hasattr(sell_factor, 'fit_week'),
|
||
self.sell_factors))
|
||
self.month_sell_factors = list(filter(lambda sell_factor: hasattr(sell_factor, 'fit_month'),
|
||
self.sell_factors))
|