Question

我正在尝试使用apply评估Herfindahl指数。我通过将数据帧转换为numpy矩阵来完成此操作。实际上，函数evalHerfindahlIndex运行良好，并且它为每行评估Herfindahl索引的正确值。但是，当我尝试使用相同的函数（evalHerfindahlIndexForDF）来应用apply时，会出现一个非常奇怪的错误：

ValueError: ("No axis named 1 for object type <class 'pandas.core.series.Series'>", 'occurred at index A')

整个代码是这样的：

import pandas as pd
import numpy as np
import datetime


def evalHerfindahlIndex(x):
    soma=np.sum(x,axis=1)
    y=np.empty(np.shape(x))
    for line in range(len(soma)):
        y[line,:]=np.power(x[line,:]/soma[line],2.0)
    hhi=np.sum(y,axis=1)    
    return hhi

def evalHerfindahlIndexForDF(x):
    soma=x.sum(axis=1)

def creatingDataFrame():

    dateList=[]
    dateList.append(datetime.date(2002,1,1))
    dateList.append(datetime.date(2002,2,1))
    dateList.append(datetime.date(2002,1,1))
    dateList.append(datetime.date(2002,1,1))
    dateList.append(datetime.date(2002,2,1))
    raw_data = {'Date': dateList,            
                'Company': ['A', 'B', 'B', 'C' , 'C'],                
                'var1': [10, 20, 30, 40 , 50]}

    df = pd.DataFrame(raw_data, columns = ['Date','Company', 'var1'])
    df.loc[1, 'var1'] = np.nan
    return df


if __name__=="__main__":
    df=creatingDataFrame()
    print(df)
    dfPivot=df.pivot(index='Date', columns='Company', values='var1')
    #print(dfPivot)
    dfPivot=dfPivot.fillna(0)
    dfPivot['Date']=dfPivot.index

    listOfCompanies=list(set(df['Company']))
    Pivot=dfPivot.as_matrix(columns=listOfCompanies)
    print(evalHerfindahlIndex(Pivot))
    print(dfPivot)

    print(dfPivot[listOfCompanies].apply(evalHerfindahlIndexForDF))

我正在使用的数据框是dfPivot：

Company        A     B     C        Date
Date                                    
2002-01-01  10.0  30.0  40.0  2002-01-01
2002-02-01   0.0   0.0  50.0  2002-02-01

使用evalHerfindahlIndex评估的赫芬达尔指数的正确值为：

[0.40625 1.     ]

我想将其作为数据帧dfPivot的额外列返回。

Answer 1

考虑更新您的方法，然后通过专门转换数组的方式返回到pandas Series

def evalHerfindahlIndex(df):
    x = df.as_matrix(columns = listOfCompanies)          # MOVE MATRIX OPERATION WITHIN FCT        
    soma = np.sum(x,axis = 1)
    y = np.empty(np.shape(x))
    for line in range(len(soma)):
        y[line,:] = np.power(x[line,:]/soma[line],2.0)
    hhi = pd.Series(np.sum(y,axis = 1))                  # CONVERT TO SERIES

    return hhi

...

if __name__=="__main__":
    df = creatingDataFrame()
    print(df)
    dfPivot = df.pivot(index = 'Date', columns = 'Company', values = 'var1')
    #print(dfPivot)
    dfPivot = dfPivot.fillna(0)
    dfPivot['Date'] = dfPivot.index

    # ASSIGN SERIES VALUES (.values to IGNORE INDEX)
    dfPivot['HE_Result'] = evalHerfindahlIndex(dfPivot).values   

    # OUTPUT  
    print(evalHerfindahlIndex(dfPivot))
    # 0    0.40625
    # 1    1.00000
    # dtype: float64

    print(dfPivot)
    # Company        A     B     C        Date  HE_Result
    # Date                                               
    # 2002-01-01  10.0  30.0  40.0  2002-01-01    0.40625
    # 2002-02-01   0.0   0.0  50.0  2002-02-01    1.00000

如何使用Apply进行此功能

1 个答案: