【原】WorldQuant101因子集整合Quantlab表达式引擎

AI量化实验室 2024-04-22 发布于北京

展开全文

原创文章第518篇，专注“AI量化投资、世界运行的规律、个人成长与财富自由"。

本周计划（源代码预计仍然是周五在星球发布）：

1、复现world quant 101因子集。

2、这些函数集合与gplearn和Deep alpha复用。

3、行业数据集用于因子挖掘测试。

其他：数据梳理：dagster。

回测平台：streamlit。

在Quantlab里复现WorldQuant101因子表达式引擎（代码+数据）

函数集在如下位置，因子集在worldquant101.py下：

from datafeed.expr_functions import *


class AlphaBase:
    pass


# https://www./data/dict/alpha101
class WorldQuant101(AlphaBase):
    def get_names_features(self):
        names = []
        features = []

        # names.append('alpha001')
        # features.append('(rank(ts_argmax(signed_power((stddev(returns, 20) if (returns < 0) else close), 2.), '
        #                '5)) - 0.5)')

        names.append('alpha002')
        features.append('(-1 * correlation(rank(delta(log(volume), 2)), rank(((close - open) / open)), 6))')

        names.append('alpha003')
        features.append('(-1 * correlation(rank(open), rank(volume), 10))')

        names.append('alpha004')
        features.append('(-1 * ts_rank(rank(low), 9))')

        '''
  Alpha#5: (rank((open - (sum(vwap, 10) / 10))) * (-1 * abs(rank((close - vwap))))) 
        Alpha#6: (-1 * correlation(open, volume, 10))         '''

        names.append('alpha006')
        features.append('(-1 * correlation(open, volume, 10))')

        '''
        Alpha#7: ((adv20 < volume) ? ((-1 * ts_rank(abs(delta(close, 7)), 60)) * sign(delta(close, 7))) : (-1 * 1)) 
Alpha#8: (-1 * rank(((sum(open, 5) * sum(returns, 5)) - delay((sum(open, 5) * sum(returns, 5)), 10)))) 
        '''

        # todo: 这里把sum替换成ts_sum,sum是python内置函数
        names.append('alpha008')
        features.append(
            '(-1 * rank(((ts_sum(open, 5) * ts_sum(returns, 5)) - delay((ts_sum(open, 5) * ts_sum(returns, 5)), 10))))')

        return names, features


if __name__ == '__main__':
    from config import DATA_DIR
    from datafeed.expr import calc_expr

    names, features = WorldQuant101().get_names_features()
    name = names[-1]
    feature = features[-1]

    dfs = []
    for s in ['159915.SZ', '159920.SZ', '159934.SZ', '159967.SZ', '510050.SH', '510300.SH', '510500.SH', '510880.SH',
              '512100.SH', '513100.SH']:
        df = pd.read_csv(DATA_DIR.joinpath('universe').joinpath('20240415').joinpath('{}.csv'.format(s)).resolve())
        dfs.append(df)
    df_all = pd.concat(dfs)
    df_all.set_index(['date', 'symbol'], inplace=True)
    df_all.dropna(inplace=True)
    df_all.sort_index(level=0, inplace=True)
    df_all['returns'] = calc_expr(df_all, 'close/delay(close,1)-1')

    print(df_all)
    df_all[name] = calc_expr(df_all, feature)

    print(df_all)

直接算了因子值：

我们规范了worldquant函数的命名，统一使用python的变量命名，比如ts_rank(x,d)。

基础函数就那么几十个，本周我们基本都会准备好：

import numpy as np
import pandas as pd
from .expr_utils import calc_by_symbol, calc_by_date


@calc_by_symbol
def delay(se: pd.Series, periods=5):
    return se.shift(periods=periods)


@calc_by_symbol
def delta(se: pd.Series, periods=20):
    se_result = se - se.shift(periods=periods)
    return se_result


@calc_by_symbol
def ts_min(se: pd.Series, periods=5):
    return se.rolling(periods=periods).min()


@calc_by_symbol
def ts_max(se: pd.Series, periods=5):
    return se.rolling(periods=periods).max()


@calc_by_symbol
def ts_argmin(se: pd.Series, periods=5):
    return se.rolling(periods=periods).argmin()


@calc_by_symbol
def ts_argmax(se: pd.Series, periods=5):
    return se.rolling(periods=periods).argmax()


@calc_by_symbol
def stddev(se, periods=5):
    return se.rolling(periods=periods).std()


@calc_by_date
def rank(se: pd.Series):
    ret = se.rank(pct=True)
    return ret


@calc_by_symbol
def ts_rank(se: pd.Series, periods=9):
    ret = se.rolling(window=periods).rank(pct=True)
    return ret


@calc_by_symbol
def sum(se: pd.Series, N):
    return se.rolling(N).sum()


@calc_by_symbol
def shift(se: pd.Series, N):
    return se.shift(N)


@calc_by_symbol
def roc(se: pd.Series, N):
    return se / shift(se, N) - 1