如何使用多级索引pandas数据帧中的列的总和值作为新列中值的条件

Question

我有一个多级索引pandas数据帧。我想创建一个新列，其中此列中的值基于条件。此条件基于对该索引的另一列求和，然后将其减半。如果这小于存储在单独列表中的最后一个值，则新列中的值将采用与数据帧中另一列相同的值。如果不满足此条件，则新列中的所有值都应为0。

使用这个问题试图实现这个Sum columns by level in a Multi-Index DataFrame我使用了np.where和df.sum(level=0, axis=1)的组合，但这会导致以下错误：

ValueError: operands could not be broadcast together with shapes (2,8) (21,) ()

这是我的数据帧和我迄今使用的代码的示例：

import pandas as pd
import numpy as np

balance = [1400]

data = {'EVENT_ID': [112335580,112335580,112335580,112335580,112335580,112335580,112335580,112335580, 112335582,
                     112335582,112335582,112335582,112335582,112335582,112335582,112335582,112335582,112335582,
                     112335582,112335582,112335582],

 'SELECTION_ID': [6356576,2554439,2503211,6297034,4233251,2522967,5284417,7660920,8112876,7546023,8175276,8145908,
                  8175274,7300754,8065540,8175275,8106158,8086265,2291406,8065533,8125015],

 'Pot_Bet': [3.236731,2.416966,2.278365,2.264023,2.225353,2.174407, 2.141420,2.122386,2.832997,2.411094,
         2.167218,2.138972,2.132137,2.128341,2.116338,2.115239,2.115123,2.114284362,2.113420,
         2.113186,2.112729],

  'Liability':[3.236731, 2.416966, 12.245492, 12.795112, 15.079176, 23.336171, 50.741182, 571.003118, 2.832997, 6.691736, 15.808607, 27.935834, 35.954927, 43.275250, 147.165537, 193.017915, 199.622454, 265.809019, 405.808678, 473.926781, 706.332594]}

df = pd.DataFrame(data, columns=['EVENT_ID', 'SELECTION_ID', 'Pot_Bet','WIN_LOSE'])

df.set_index(['EVENT_ID', 'SELECTION_ID'], inplace=True) #Selecting columns for indexing

df['Bet'] = np.where(df.sum(level = 0) > 0.5*balance[-1], df['Pot_Bet'], 0)

这导致前面所述的错误。

对于索引112335580，新列应具有与'Pot_Bet'相同的值。而对于索引112335582，新列应具有0的值。

干杯，桑迪

Answer 1

问题是如果使用df.sum(level=0)它像df.groupby(level = 0).sum()一样 - 聚集在第一级MultiIndex。

解决方案是使用GroupBy.transform为Series与原始DataFrame相同的大小：

df['Bet'] = np.where(df.groupby(level = 0)['Pot_Bet'].transform('sum') > 0.5*balance[-1], 
                     df['Pot_Bet'], 0)

详情：

print (df.groupby(level = 0)['Pot_Bet'].transform('sum'))
EVENT_ID   SELECTION_ID
112335580  6356576         18.859651
           2554439         18.859651
           2503211         18.859651
           6297034         18.859651
           4233251         18.859651
           2522967         18.859651
           5284417         18.859651
           7660920         18.859651
112335582  8112876         28.611078
           7546023         28.611078
           8175276         28.611078
           8145908         28.611078
           8175274         28.611078
           7300754         28.611078
           8065540         28.611078
           8175275         28.611078
           8106158         28.611078
           8086265         28.611078
           2291406         28.611078
           8065533         28.611078
           8125015         28.611078
Name: Pot_Bet, dtype: float64

如果只需要工作机智列，可以按列名称选择Series：

print (df['Pot_Bet'].sum(level=0))
EVENT_ID
112335580    18.859651
112335582    28.611078
Name: Pot_Bet, dtype: float64

print (df.groupby(level = 0)['Pot_Bet'].sum())
EVENT_ID
112335580    18.859651
112335582    28.611078
Name: Pot_Bet, dtype: float64