Navigation

    Backtrader Community
    • Register
    • Login
    • Search
    • Categories
    • Recent
    • Tags
    • Popular
    • Users
    • Groups
    • Search
    For code/output blocks: Use ``` (aka backtick or grave accent) in a single line before and after the block. See: http://commonmark.org/help/

    Why optstrategy not run the stratege in the 2nd circle

    General Discussion
    1
    4
    156
    Loading More Posts
    • Oldest to Newest
    • Newest to Oldest
    • Most Votes
    Reply
    • Reply as topic
    Log in to reply
    This topic has been deleted. Only users with topic management privileges can see it.
    • M
      michael_1589 last edited by

      Here is my code. I can see "FactorStrategy next" log in the first run but none in the 2nd run. I tried other demos and found that "Next" method should be called in the 2nd run. Is there anything wrong with my code?

      #!/usr/bin/env python3
# -*- coding: UTF-8 -*-

from __future__ import (absolute_import, division, print_function,
                        unicode_literals)

from datetime import datetime
from datetime import timedelta
import backtrader as bt
import warnings
import math
import numpy as np
import pandas as pd
from feed import tdx_feed
from factor.factor import get_factors
from enum import IntEnum, unique
import alphalens
import logging

warnings.filterwarnings('ignore')

logging.basicConfig(format="[%(asctime)s %(levelname)s %(thread)d %(filename)s:%(lineno)d] %(message)s",
                    level=logging.DEBUG,
                    )

@unique
class Universe(IntEnum):
    HS300 = 0
    ZZ500 = 1
    ZZ800 = 2
    ZZ1000 = 3

@unique
class Period(IntEnum):
    LAST_WEEK = 1
    LAST_MONTH = 2
    LAST_HALF_YEAR = 3

@unique
class Portfolio(IntEnum):
    PURE_LONG_1 = 1
    PURE_LONG_5 = 5
    LONG_SHORT_I = 3
    LONG_SHORT_II = 4


@unique
class FeeSlippage(IntEnum):
    NONE = 1
    FEEONLY = 2
    FEESLIPPAGE = 3
@unique
class FilterStop(IntEnum):
    NO = 1
    YES = 2

@unique
class RebalancePeriod(IntEnum):
    ONE_WEEK = 1

class input_param:
    universe = 0
    period = 0
    portfolio = 0
    feeSlippage = 1
    filterStop = 0
    rebancePeriod = 0
    factors = []
    customize_universe = []
    modname = ""
    # fromdate = 0
    # todate = 0

    fromdate = None
    todate = None
    start_cash = 100000000.0
    data_length = 365
    code = '000300'
    setcode = 1
    # factorCls = MyIndicator
    reblance_period = 1
    group_size = 5
    debug = True

    def set(self, js):
        self.universe = js["universe"]
        self.period = js["period"]
        self.portfolio = js["portfolio"]
        self.feeSlippage = js["feeSlippage"]
        self.filterStop = js["filterStop"]
        self.rebancePeriod = js["rebancePeriod"]

    def parse(self):
        # Get the dates from the args
        if self.period:
            fromdate = (datetime.today() - timedelta(days=180)).date()
            todate = datetime.today().date()
        elif self.fromdate:
            fromdate = datetime.strptime(self.fromdate, '%Y-%m-%d')
            todate = datetime.strptime(self.todate, '%Y-%m-%d')
        else:
            fromdate = (datetime.today() - timedelta(days=30)).date()
            todate = datetime.today().date()
        self.todate = todate.year * 10000 + todate.month * 100 + todate.day


class FactorStrategy(bt.Strategy):
    '''
    group factors
    '''
    params = dict(period=5,
                  # factorCls=MyIndicator,
                  reblance_period=1,
                  group_size=5,
                  portfolio=Portfolio.LONG_SHORT_I,
                  )

    def __init__(self):
        self.headers = [d._name for d in self.datas]
        self.factor_data = pd.DataFrame()
        self.turnover = pd.DataFrame()
        logging.debug('FactorStrategy __init__')


    def log(self, txt):
        dt = self.datas[0].datetime.datetime(0)
        logging.debug(f'{dt.isoformat()}, {txt}')

    def get_quantile_index(self, portfolio):
        p_index, short_index = None, None
        if portfolio == Portfolio.PURE_LONG_1:
            p_index, short_index = 0, None
        elif portfolio == Portfolio.PURE_LONG_5:
            p_index, short_index = 4, None
        elif portfolio == Portfolio.LONG_SHORT_I:
            p_index, short_index = [0, 4], 0
        elif portfolio == Portfolio.LONG_SHORT_II:
            p_index, short_index = [0, 4], 4
        return p_index, short_index

    def get_factor_quantile(self, dt, p_index, column='factor_quantile'):
        if isinstance(p_index, int):
            p_index = [p_index]
        for index, row in dt.iterrows():
            if row[column] not in p_index:
                dt.loc[index, column] = math.nan
        return dt.dropna(subset=[column])

    def next(self):
        logging.debug('FactorStrategy next')
        self.get_pd_data()
        if len(self.data0) % self.p.reblance_period == 0:
            p_index, short_index = None, None
            if self.p.portfolio == Portfolio.PURE_LONG_1:
                p_index, short_index = 0, None
            elif self.p.portfolio == Portfolio.PURE_LONG_5:
                p_index, short_index = 4, None
            elif self.p.portfolio == Portfolio.LONG_SHORT_I:
                p_index, short_index = [0,4], 0
            elif self.p.portfolio == Portfolio.LONG_SHORT_II:
                p_index, short_index = [0,4], 4

            self.trade_portfolio(p_index, short_index)

    def get_pd_data(self):
        factor_row = []
        price_row = []
        tradeable_row = []
        dt0 = bt.num2date(self.data0.datetime[0])
        for d in self.datas:
            factor = d.l.factor[0]
            factor_row.append(factor)

            price = d.l.open[0]
            price_row.append(price)
            trade = d.l.tradeable[0]
            tradeable_row.append(trade)

        dates = list(np.array([dt0] * len(self.datas)))
        df_factor = pd.DataFrame(factor_row, index=[dates, self.headers],
                          columns=["factor"])
        df_factor.index.names = ["date", "asset"]

        df_quantile = pd.DataFrame(pd.qcut(factor_row, self.p.group_size, labels=False, duplicates='drop'),
                                   index=[dates, self.headers], columns=["factor_quantile"])

        df_price = pd.DataFrame(price_row, index=[dates, self.headers],
                                 columns=["price"])

        df_tradeable = pd.DataFrame(tradeable_row, index=[dates, self.headers],
                                columns=["tradeable"])
        df = pd.concat([df_factor, df_quantile, df_price, df_tradeable], axis=1)
        self.factor_data = self.factor_data.append(df)

    def trade_portfolio(self, p_index, short_index):
        """
        trade portfolio every period
        :param p_index:
        :param short_index:
        :return:
        """
        try:
            trade_count = 0
            group_quantile_last = pd.DataFrame()
            dt0 = bt.num2date(self.data0.datetime[0])
            dt_last = bt.num2date(self.data.datetime[-self.p.reblance_period])
            dt = self.factor_data.loc[dt_last, :]
            group_quantile_last = self.get_factor_quantile(dt, p_index)
        except Exception as e:
            logging.error(str(e))

        dt = self.factor_data.loc[dt0, :]
        group_quantile = self.get_factor_quantile(dt, p_index)
        # sell these holdings
        if group_quantile_last.size > 0:
            sell_hold = set(group_quantile_last.index) - set(group_quantile.index)
            for s in sell_hold:
                t = group_quantile_last.loc[s, :]
                if t.tradeable:
                    self.close(s)
                    trade_count = trade_count + 1
            buy_target = list(set(group_quantile.index) - set(group_quantile_last.index))
            logging.debug('close prev: %s', str(sell_hold))
        else:
            buy_target = list(set(group_quantile.index))

        # buy these portfolio
        can_trade = sum(group_quantile['tradeable'].loc[buy_target])
        if can_trade > 0.5:
            value = self.broker.getcash() / can_trade
            for s in buy_target:
                t = group_quantile.loc[s, :]
                if t.tradeable and int(t.factor_quantile) != short_index:
                    ss = int((value / 100) / t.price) * 100
                    self.buy(size=ss)
                    trade_count = trade_count + 1
                elif t.tradeable:
                    ss = int((value / 100) / t.price) * 100
                    self.buy(size=-ss)
                    trade_count = trade_count + 1
            logging.debug('buy today: %s', str(buy_target))
        else:
            self.log("can't trade this day")
            pass

        # compute turnover
        if group_quantile_last.size > 0:
            turnover = trade_count / group_quantile_last.size
        else:
            turnover = math.nan
        turnover_dt = pd.DataFrame(turnover, index=[dt0], columns=['turnover'])
        self.turnover = self.turnover.append(turnover_dt)

    def stop(self):
        self.value = self.broker.get_value()
        logging.debug("finnal value: %f", self.value)
        logging.debug("final factors:")
        logging.debug(self.factor_data)

def get_code_result(results):
    '''
    fetch all results from mysql one time.
    :param results:
    :return: map, key=code, value=list(record)
    '''
    dict_code_result = {}
    c0 = ''
    start = end = 0
    for i, r in enumerate(results):
        if c0 != r['GpCode']:
            if c0 != '':
                dict_code_result[c0] = results[start:end + 1]

            c0 = r['GpCode']
            start = i
            dict_code_result[r['GpCode']] = []

        else:
            end = i
    if c0 != '':
        dict_code_result[c0] = results[start:]

    # todo...make sure every code in result has the same length.
    return dict_code_result

def filter_codes(dict_code_result):
    '''
    :param dict_code_result:
    :return:
    '''
    for key in list(dict_code_result.keys()):
        if len(dict_code_result[key]) < 3*250:
            del dict_code_result[key]
    return dict_code_result

def get_benchmark_returns(bench):
    dt0 = bench[0]['close']
    dt1 = bench[-1]['close']
    return (dt1 - dt0) / dt0

    # ignore
    ret = [i['close'] for i in bench]
    dt = [datetime.strptime(str(i['datetime']), "%Y%m%d") for i in bench]
    a = pd.DataFrame(ret, index=dt, columns=['returns'])
    return a

def get_statistics(factor_data, turnover, net_profit_rate, bench_returns_rate, data_length, point_number=2):
    """
    get statistics
    :param factor_data:
    :param turnover:
    :param net_profit_rate:
    :param bench_returns_rate:
    :param data_length:
    :param point_number:
    :return:
    """

    annual_net_profit_rate = round(net_profit_rate / data_length * 365, point_number)

    alpha_returns_rate = round(net_profit_rate - bench_returns_rate, point_number)

    annual_alpha_returns_rate = round(alpha_returns_rate * 365 / data_length, point_number)

    ic = alphalens.performance.factor_information_coefficient(factor_data)
    IC = round(ic.mean()[0], point_number)
    IR = round((ic.mean() / ic.std())[0], point_number)

    return dict({
        "annulReturn":annual_net_profit_rate,
        "alpha": annual_alpha_returns_rate,
        "turnover":round(turnover.mean(), point_number),
        "IC":IC,
        "IR":IR,
    })

def get_analysis(analyzer):
    if analyzer:
        pss = analyzer.get_analysis()
        # ps = [[k] + v[-2:] for k, v in iteritems(pss)]
        ps = [[k] + v for k, v in iter(pss.items())]
        cols = ps.pop(0)  # headers are in the first entry
        positions = pd.DataFrame.from_records(ps[1:], columns=cols)
        positions.index = pd.to_datetime(positions['Datetime'])
        del positions['Datetime']
        # positions.index = positions.index.tz_localize('UTC')
        positions = positions.stack()
        positions.index.set_names(['date', 'asset'], inplace=True)
        positions.columns = ['returns']
        return positions
    return None

def cb(strategies):
    for strategy in strategies:
        logging.debug(strategy.p.portfolio)
        logging.debug(strategy.value)
        logging.debug(strategy.factor_data)

def run_once(args, factors, map_data, bench_returns_rate):
    logging.debug('run_once run_once run_once run_once')
    groups = factors.groupby("asset")
    cerebro = bt.Cerebro(optreturn=False,  stdstats=False)

    for k, v in map_data.items():
        f = groups.get_group(k)
        data = tdx_feed.TdxFeed(
            result=v,
            factors= f,
            plot=False,
        )
        cerebro.adddata(data, name=k)

    if args.portfolio == 0:
        cerebro.optstrategy(FactorStrategy,
                            group_size=args.group_size,
                            reblance_period=args.reblance_period,
                            portfolio=(Portfolio.PURE_LONG_1, Portfolio.PURE_LONG_5),
                            )
    elif args.portfolio == 1:
        cerebro.optstrategy(FactorStrategy,
                            group_size=args.group_size,
                            reblance_period=args.reblance_period,
                            portfolio=(Portfolio.LONG_SHORT_I,),
                            )
    elif args.portfolio == 2:
        cerebro.optstrategy(FactorStrategy,
                            group_size=args.group_size,
                            reblance_period=args.reblance_period,
                            portfolio=(Portfolio.LONG_SHORT_II,),
                            )
    else:
        logging.error("can't be here")
    cerebro.broker.setcash(args.start_cash)
    cerebro.broker.set_slippage_fixed(0.001)
    result_dict = dict()
    cerebro.optcallback(cb)
    # Execute
    cerebro.run(maxcpus=1)

    return result_dict


def runstrat(input):
    args = input_param()
    args.set(input)
    args.parse()

    logging.debug('loading from pickle...')
    # for debug use only:
    import pickle
    with open("E:/code/src/tdx_factor/last_year.pickle", 'rb') as ff:
        map_data = pickle.load(ff)
    with open("E:/code/src/tdx_factor/000300_last_year.pickle", 'rb') as ff:
        bench = pickle.load(ff)


    bench_returns_rate = get_benchmark_returns(bench)
    logging.debug('bench_returns_rate %f', bench_returns_rate)
    factor_all = get_factors(map_data, bench)
    logging.debug(len(factor_all))
    result_statistics = list()
    for k, v in factor_all.items():
        statis = run_once(args, v, map_data, bench_returns_rate)
        statis["factorid"] = k[0]
        statis["factorName"] = k[1]
        result_statistics.append(statis)
    return result_statistics


if __name__ == '__main__':
    input = {"universe": 0,"period": 1,"portfolio": 0,"feeSlippage": 1,"filterStop": 0,"rebancePeriod": 0,"factors": [],"customize_universe": [],"modname": "mod_pypf64.dll"}
    runstrat(input)
    logging.debug('===========================')

      Here is the log:

      [2021-03-23 18:14:02,057 DEBUG 5040 factor_trade.py:576] loading from pickle...
[2021-03-23 18:14:02,195 DEBUG 5040 factor_trade.py:590] bench_returns_rate 0.338562
[2021-03-23 18:14:11,536 INFO 5040 utils.py:129] Note: NumExpr detected 12 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 8.
[2021-03-23 18:14:11,536 INFO 5040 utils.py:141] NumExpr defaulting to 8 threads.
[2021-03-23 18:14:14,232 DEBUG 5040 factor_trade.py:592] 1
[2021-03-23 18:14:14,232 DEBUG 5040 factor_trade.py:452] run_once run_once run_once run_once
[2021-03-23 18:14:16,579 DEBUG 5040 factor_trade.py:185] FactorStrategy __init__
[2021-03-23 18:14:16,583 DEBUG 5040 factor_trade.py:216] FactorStrategy next
[2021-03-23 18:14:16,591 ERROR 5040 factor_trade.py:277] datetime.datetime(2020, 12, 31, 0, 0)
[2021-03-23 18:14:16,597 DEBUG 5040 factor_trade.py:193] 2019-07-05T00:00:00, can't trade this day
.
.
.
[2021-03-23 18:14:24,715 DEBUG 5040 factor_trade.py:343] finnal value: 118570513.235212
[2021-03-23 18:14:24,715 DEBUG 5040 factor_trade.py:344] final factors:
[2021-03-23 18:14:24,715 DEBUG 5040 factor_trade.py:345]                      factor  factor_quantile       price  tradeable
date       asset                                                   
2019-07-05 000001       NaN              NaN   13.730000        1.0
           000002       NaN              NaN   27.389999        1.0
           000063       NaN              NaN   33.090000        1.0
           000066       NaN              NaN    9.860000        1.0
           000069       NaN              NaN    7.170000        1.0
                     ...              ...         ...        ...
2020-12-31 000768  1.514446              4.0   34.790001        1.0
           000776 -0.080605              1.0   15.850000        1.0
           000783  0.642726              2.0    8.210000        1.0
           000786  0.607996              2.0   37.900002        1.0
           000858  1.026616              3.0  288.010010        1.0
[10950 rows x 4 columns]
[2021-03-23 18:14:24,725 DEBUG 5040 factor_trade.py:520] ****************************************************************
[2021-03-23 18:14:24,725 DEBUG 5040 factor_trade.py:521] Portfolio.PURE_LONG_1
Dropped 16.4% entries from factor data: 16.4% in forward returns computation and 0.0% in binning phase (set max_loss=0 to see potentially suppressed Exceptions).
max_loss is 35.0%, not exceeded: OK!
[2021-03-23 18:14:26,075 DEBUG 5040 factor_trade.py:531] {'annulReturn': 0.27, 'alpha': -0.22, 'turnover': 0.07, 'IC': 0.19, 'IR': 0.77}
[2021-03-23 18:14:26,078 DEBUG 5040 factor_trade.py:185] FactorStrategy __init__
[2021-03-23 18:14:26,079 DEBUG 5040 factor_trade.py:343] finnal value: 100000000.000000
[2021-03-23 18:14:26,079 DEBUG 5040 factor_trade.py:344] final factors:
[2021-03-23 18:14:26,079 DEBUG 5040 factor_trade.py:345] Empty DataFrame
Columns: []
Index: []
[2021-03-23 18:14:26,080 DEBUG 5040 factor_trade.py:520] ****************************************************************
[2021-03-23 18:14:26,080 DEBUG 5040 factor_trade.py:521] Portfolio.PURE_LONG_5
      M 1 Reply Last reply Reply Quote 0
      • M
        michael_1589 @michael_1589 last edited by

        I found self.buflen() returns 0 in the 2nd run when call advance_peek method. buflen returns 202 in the first run(which exact contains 202 days). So why buflen returns 0 in the 2nd run? I will check...

        M 1 Reply Last reply Reply Quote 0
        • M
          michael_1589 @michael_1589 last edited by

          d.advance_peek() returns inf in the function _runonce in cerebro.py. Strange! It returns a value in the first run but inf in the 2nd run.

              def _runonce(self, runstrats):
        '''
        Actual implementation of run in vector mode.

        Strategies are still invoked on a pseudo-event mode in which ``next``
        is called for each data arrival
        '''
        for strat in runstrats:
            strat._once()
            strat.reset()  # strat called next by next - reset lines

        # The default once for strategies does nothing and therefore
        # has not moved forward all datas/indicators/observers that
        # were homed before calling once, Hence no "need" to do it
        # here again, because pointers are at 0
        datas = sorted(self.datas,
                       key=lambda x: (x._timeframe, x._compression))

        while True:
            # Check next incoming date in the datas
            dts = [d.advance_peek() for d in datas]
            dt0 = min(dts)
            if dt0 == float('inf'):
                break  # no data delivers anything
          M 1 Reply Last reply Reply Quote 0
          • M
            michael_1589 @michael_1589 last edited by

            data feed issue. Change the feed can fix.

            1 Reply Last reply Reply Quote 0
            • 1 / 1
            • First post
              Last post
            Copyright © 2016, 2017, 2018, 2019, 2020, 2021 NodeBB Forums | Contributors