如何将 pandas 数据帧的值除以每一组的第一行？

Question

一个pandas数据框：

>>> df
                  sales  net_pft
STK_ID RPT_Date                 
002138 20140930   3.325    0.607
       20150930   3.619    0.738
       20160930   4.779    0.948
600004 20140930  13.986    2.205
       20150930  14.226    3.080
       20160930  15.499    3.619
600660 20140930  31.773    5.286
       20150930  31.040    6.333
       20160930  40.062    7.186

只想知道如何获得输出，因为每行的值除以每组的第一行，如下所示：

                  sales  net_pft
STK_ID RPT_Date                 
002138 20140930   1.000    1.000
       20150930   1.088    1.216
       20160930   1.437    1.562
600004 20140930   1.000    1.000
       20150930   1.017    1.397
       20160930   1.108    1.641
600660 20140930   1.000    1.000
       20150930   0.977    1.198
       20160930   1.261    1.359

谢谢，

Answer 1

import pandas as pd

df = pd.DataFrame({'RPT_Date': ['20140930', '20150930', '20160930', '20140930', '20150930', '20160930', '20140930', '20150930', '20160930'], 'STK_ID': ['002138', '002138', '002138', '600004', '600004', '600004', '600660', '600660', '600660'], 'net_pft': [0.607, 0.738, 0.948, 2.205, 3.080, 3.619, 5.286, 6.333, 7.186], 'sales': [3.325, 3.619, 4.779, 13.986, 14.226, 15.499, 31.773, 31.040, 40.062]})
df = df.set_index(['STK_ID','RPT_Date'])

firsts = (df.groupby(level=['STK_ID']).transform('first'))
result = df / firsts

产量

                  net_pft     sales
STK_ID RPT_Date                    
002138 20140930  1.000000  1.000000
       20150930  1.215815  1.088421
       20160930  1.561779  1.437293
600004 20140930  1.000000  1.000000
       20150930  1.396825  1.017160
       20160930  1.641270  1.108180
600660 20140930  1.000000  1.000000
       20150930  1.198070  0.976930
       20160930  1.359440  1.260882

上面的主要技巧是使用groupby/transform('first')创建一个DataFrame 与 df 形状相同，但其值来自每组的第一行：

firsts = df.groupby(level=['STK_ID']).transform('first')
#                  net_pft   sales
# STK_ID RPT_Date                 
# 002138 20140930    0.607   3.325
#        20150930    0.607   3.325
#        20160930    0.607   3.325
# 600004 20140930    2.205  13.986
#        20150930    2.205  13.986
#        20160930    2.205  13.986
# 600660 20140930    5.286  31.773
#        20150930    5.286  31.773
#        20160930    5.286  31.773

虽然这是对内存的浪费，但这可能是获得所需结果的最快方法，因为它避免了遍历 Python 中的组。

如果上述代码在 Pandas 版本 0.13 中引发 TypeError: Transform function invalid for data types，您可以尝试使用此解决方法：

result = list()
for key, grp in df.groupby(level=['STK_ID']):
    result.append(grp/grp.iloc[0])
result = pd.concat(result)
print(result)

如何将 pandas 数据帧的值除以每一组的第一行？

How to divide pandas dataframe's value by its first row by each group?

python

normalization

pandas