1

[使用 Python3] 我正在使用 pandas 读取 csv 文件,对数据帧进行分组,将函数应用于分组数据并将这些结果添加回原始数据帧。

我的输入如下所示:

email                   cc  timebucket  total_value
john@john.com           us  1           110.50
example@example.com     uk  3           208.84
...                     ... ...         ...

基本上我正在尝试分组cc并计算该组中每个值的百分位数total_value。其次,我想对这些结果应用一个流程语句。我需要将这些结果添加回原始/父 DataFrame。这样它看起来像这样:

email                   cc  timebucket  total_value     percentrank rankbucket
john@john.com           us  1           110.50          48.59       mid50
example@example.com     uk  3           208.84          99.24       top25
...                     ... ...         ...             ...         ...

下面的代码给了我一个AssertionError,我不知道为什么。我对 Python 和 pandas 很陌生,所以这可能会解释一个又一个。

代码:

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

def percentilerank(frame, groupkey='cc', rankkey='total_value'):
    from pandas.compat.scipy import percentileofscore

    # Technically the below percentileofscore function should do the trick but I cannot
    # get that to work, hence the alternative below. It would be great if the answer would
    # include both so that I can understand why one works and the other doesnt.
    # func = lambda x, score: percentileofscore(x[rankkey], score, kind='mean')

    func = lambda x: (rankdata(x.total_value)-1)/(len(x.total_value)-1)*100
    frame['percentrank'] = frame.groupby(groupkey).transform(func)


def calc_and_write(filename):
    """
    Function reads the file (must be tab-separated) and stores in a pandas DataFrame.
    Next, the percentile rank score based is calculated based on total_value and is done so within a country.
    Secondly, based on the percentile rank score (prs) a row is assigned to one of three buckets:
        rankbucket = 'top25' if prs > 75
        rankbucket = 'mid50' if 25 > prs < 75
        rankbucket = 'bottom25' if prs < 25
    """

    # Define headers for pandas to read in DataFrame, stored in a list
    headers = [
        'email',            # 0
        'cc',               # 1
        'last_trans_date',  # 3
        'timebucket',       # 4
        'total_value',      # 5
    ]

    # Reading csv file in chunks and creating an iterator (is supposed to be much faster than reading at once)
    tp = pd.read_csv(filename, delimiter='\t', names=headers, iterator=True, chunksize=50000)
    # Concatenating the chunks and sorting total DataFrame by booker_cc and total_nett_spend
    df = pd.concat(tp, ignore_index=True).sort(['cc', 'total_value'], ascending=False)

    percentilerank(df)

编辑:根据要求,这是回溯日志:

Traceback (most recent call last):
  File "C:\Users\m\Documents\Python\filter_n_split_3.py", line 85, in <module>
    print(calc_and_write('tsv/test.tsv'))
  File "C:\Users\m\Documents\Python\filter_n_split_3.py", line 74, in calc_and_write
    percentilerank(df)
  File "C:\Users\m\Documents\Python\filter_n_split_3.py", line 33, in percentilerank
    frame['percentrank'] = frame.groupby(groupkey).transform(func)
  File "C:\Python33\lib\site-packages\pandas\core\groupby.py", line 1844, in transform
    axis=self.axis, verify_integrity=False)
  File "C:\Python33\lib\site-packages\pandas\tools\merge.py", line 894, in concat
    verify_integrity=verify_integrity)
  File "C:\Python33\lib\site-packages\pandas\tools\merge.py", line 964, in __init__
    self.new_axes = self._get_new_axes()
  File "C:\Python33\lib\site-packages\pandas\tools\merge.py", line 1124, in _get_new_axes
    assert(len(self.join_axes) == ndim - 1)
AssertionError
4

1 回答 1

3

尝试这个。您的示例从转换函数返回一个系列,但应该返回一个值。(这使用熊猫等级功能仅供参考)

In [33]: df
Out[33]: 
                 email  cc  timebucket  total_value
0        john@john.com  us           1       110.50
1  example@example.com  uk           3       208.84
2          foo@foo.com  us           2        50.00

In [34]: df.groupby('cc')['total_value'].apply(lambda x: 100*x.rank()/len(x))
Out[34]: 
0    100
1    100
2     50
dtype: float64

In [35]: df['prank'] = df.groupby('cc')['total_value'].apply(lambda x: 100*x.rank()/len(x))

In [36]: df
Out[36]: 
                 email  cc  timebucket  total_value  prank
0        john@john.com  us           1       110.50    100
1  example@example.com  uk           3       208.84    100
2          foo@foo.com  us           2        50.00     50
于 2013-07-09T15:15:13.197 回答