将函数应用于多索引 Pandas DataFrame

Question

这是我正在处理的 DataFrame 示例：

import pandas as pd
import numpy as np
from scipy.stats import zscore

df = pd.DataFrame(
    index=pd.MultiIndex.from_tuples(
    [('Monday', '2019-11-04'),('Monday', '2019-11-11'), ('Monday', '2019-11-18'),
    ('Tuesday', '2019-11-05'), ('Tuesday', '2019-11-12'), ('Tuesday', '2019-11-19'),
    ('Wednesday', '2019-11-06'), ('Wednesday', '2019-11-13'), ('Wednesday', '2019-11-20'),
    ( 'Thursday', '2019-11-07'), ('Thursday', '2019-11-14'), ('Thursday', '2019-11-21'),
    ('Friday', '2019-11-01'), ('Friday', '2019-11-08'), ('Friday', '2019-11-15'),
    ('Saturday', '2019-11-02'), ('Saturday', '2019-11-09'), ('Saturday', '2019-11-16'),
    ('Sunday', '2019-11-03'), ('Sunday', '2019-11-10'), ('Sunday', '2019-11-17')]),

    data={'A': [363287, 348759, 295711, 346276, 350785, 292794, 328048, 315418, 
                303901, 324330, 302850, 308500, 415665, 324196, 289739, 444184,
                361214, 359573, 436543, 375668, 379184],
          'B': [263641, 293827, 272811, 267064, 307886, 269061, 266336, 292442,
                273714, 268377, 278113, 270378, 268556, 274989, 268869, 312046,
                321059, 322694, 323546, 332234, 333341],
          'C': [263678, 293870, 272855, 267092, 307931, 269114, 266378, 292488,
                273769, 268426, 278156, 270422, 268602, 275021, 268906, 312084,
                321116, 322741, 323602, 332298, 333405]})

现在，我通过使用 for 循环将scipy.stats.zscore应用于每一列来获取每列中每个值的scipy.stats.zscore ：

for col in df.columns:
    df[col] = zscore(df[col])

在应用 zscore 函数时，不是考虑每列中的所有数字，而是如何在应用该函数之前按索引的第一级（星期几）进行分组？ 例如，我想先将该函数应用于df.loc[('Monday'), 'A'] ，然后是df.loc[('Tuesday'), 'A']等等。

此外，是否有一种方法可以不涉及将 DataFrame 的子集附加到列表中，然后在处理它们后将它们连接起来。

谢谢！

Answer 1

使用Groupby.transform ：

df.groupby(level=0)['A','B','C'].transform(zscore)

#                             A         B         C
#weekdays  dates                                   
#Monday    2019-11-04  0.942314 -1.038220 -1.038401
#          2019-11-11  0.442097  1.350720  1.350641
#          2019-11-18 -1.384411 -0.312500 -0.312240
#Tuesday   2019-11-05  0.619782 -0.759579 -0.760220
#          2019-11-12  0.790974  1.412882  1.412849
#          2019-11-19 -1.410756 -0.653303 -0.652628
#Wednesday 2019-11-06  1.243122 -1.015742 -1.016228
#          2019-11-13 -0.037621  1.360045  1.359854
#          2019-11-20 -1.205501 -0.344304 -0.343626
#Thursday  2019-11-07  1.367941 -0.931907 -0.931481
#          2019-11-14 -0.994700  1.387182  1.387292
#          2019-11-21 -0.373242 -0.455275 -0.455811
#Friday    2019-11-01  1.363756 -0.759293 -0.757889
#          2019-11-08 -0.357646  1.412897  1.412967
#          2019-11-15 -1.006110 -0.653604 -0.655078
#Saturday  2019-11-02  1.414010 -1.399768 -1.399981
#          2019-11-09 -0.686236  0.525278  0.526673
#          2019-11-16 -0.727775  0.874490  0.873309
#Sunday    2019-11-03  1.412341 -1.406665 -1.406678
#          2019-11-10 -0.769170  0.576959  0.577073
#          2019-11-17 -0.643171  0.829706  0.829605

这组按级别 = 0 的索引（星期一、星期二...）

或者如果你想重命名索引

df = df.rename_axis(index = ['weekdays','dates'])
df.groupby('weekdays').transform(zscore)