如何让你的pandas循环更快

本文转载至How To Make Your Pandas Loop 71803 Times Faster

结论 - numpy向量化操作最快

标准循环

def soc_loop(leaguedf,TEAM,):
    leaguedf['Draws'] = 99999
    for row in range(0, len(leaguedf)):
        if ((leaguedf['HomeTeam'].iloc[row] == TEAM) & (leaguedf['FTR'].iloc[row] == 'D')) | \
            ((leaguedf['AwayTeam'].iloc[row] == TEAM) & (leaguedf['FTR'].iloc[row] == 'D')):
            leaguedf['Draws'].iloc[row] = 'Draw'
        elif ((leaguedf['HomeTeam'].iloc[row] == TEAM) & (leaguedf['FTR'].iloc[row] != 'D')) | \
            ((leaguedf['AwayTeam'].iloc[row] == TEAM) & (leaguedf['FTR'].iloc[row] != 'D')):
            leaguedf['Draws'].iloc[row] = 'No_Draw'
        else:
            leaguedf['Draws'].iloc[row] = 'No_Game'

pandas内置函数 - 300倍

def soc_iter(TEAM,home,away,ftr):
    #team, row['HomeTeam'], row['AwayTeam'], row['FTR']
    if [((home == TEAM) & (ftr == 'D')) | ((away == TEAM) & (ftr == 'D'))]:
        result = 'Draw'
    elif [((home == TEAM) & (ftr != 'D')) | ((away == TEAM) & (ftr != 'D'))]:
        result = 'No_Draw'
    else:
        result = 'No_Game'
    return result

apply()方法 - 800倍

pandas向量化操作 - 9,000倍

def soc_iter(TEAM,home,away,ftr):
    df['Draws'] = 'No_Game'
    df.loc[((home == TEAM) & (ftr == 'D')) | ((away == TEAM) & (ftr == 'D')), 'Draws'] = 'Draw'
    df.loc[((home == TEAM) & (ftr != 'D')) | ((away == TEAM) & (ftr != 'D')), 'Draws'] = 'No_Draw'

numpy向量化操作 - 70,000倍