# base.py
def compose_signal(data):
    # 整合交易信号,连续出现两次买入信号只保留第一次
    data['buy_signal'] = np.where((data['buy_signal'] == 1) & (data['buy_signal'].shift(1) == 1), 0, data['buy_signal'])
    data['sell_signal'] = np.where((data['sell_signal'] == -1) & (data['sell_signal'].shift(1) == -1), 0,
                                   data['sell_signal'])

    # 合并信号，1买入，-1卖出，后面按照这个来计算收益
    data['signal'] = data['buy_signal'] + data['sell_signal']
    return data

def calculate_prof_pct(data):
    # 计算单次收益率：开仓、平仓（开仓的全部股数）
    data.loc[data['signal'] != 0, 'profit_pct'] = data['close'].pct_change()
    data = data[data['signal'] == -1]   # 筛选平仓后的数据：单次收益
    return data

def calculate_cum_prof(data):
    df = data.copy()
    df['cum_profit'] = pd.DataFrame(1+ df['profit_pct']).cumprod()-1
    return df

# ma.py
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import data.stock as st
import strategy.base as stra

def ma_strategy(data, short_windows=5, long_window=20):
    # 计算技术指标，ma短期，ma长期
    data['short_ma'] = data['close'].rolling(window=short_windows).mean()
    data['long_ma'] = data['close'].rolling(window=long_window).mean()

    # 生成信号：金叉买入，死叉卖出
    data['buy_signal'] = np.where(data['short_ma'] > data['long_ma'], 1, 0)
    data['sell_signal'] = np.where(data['short_ma'] < data['long_ma'], -1, 0)

    # 过滤信号
    data = stra.compose_signal(data)
    
    # 计算单次收益
    data = stra.calculate_prof_pct(data)

    # 计算累计收益
    data = stra.calculate_cum_prof(data)

    return data

if __name__ == '__main__':

    # # 计算多个股的累计收益率
    stocks = ['000001.XSHE', '000858.XSHE', '002594.XSHE']
    cum_profits = pd.DataFrame()
    
    for code in stocks:
        df = st.get_single_price(code=code, time_freq='daily', start_date='2016-01-01', end_date='2021-01-01')
        df = ma_strategy(df)
        cum_profits[code] = df['cum_profit'].reset_index(drop=True)
        df['cum_profit'].plot(label=code)
        
    print(cum_profits)
    plt.legend()
    plt.show()

说明：为了避免 SettingwithCopyWarning 的问题，我改写了 calculate_cum_prof 函数，即便是按照视频 calculate_cum_prof 函数写法，我测试结果依旧一样，和视频结果大相径庭。不知道哪里有问题，我核对了所有函数，没发现哪里有写错。

另：4-11 寻找最优参数章节计算的数字和视频吻合，说明 ma_strategy 函数无误，我裂了…