在Pandas DataFrame的组内上采样int系列

Question

我的问题是如何对我的数据帧中的多个“分组”中的每一个进行上采样。 （就我而言，对于每个'团队'和'LeadWeek'分组）。

我看到内置函数和大量示例用于对时间序列进行上采样，但不是用于上采样整数。由于各种原因，我现在不会进入，我想用整数代替时间序列。

在我的情况下，我有'团队'和'LeadWeeks'，我想为每个'团队'和'LeadWeek'组合上传'转换周'为[0,1,2,3,4]。

我认为有一种方法可以用multi-index / groupby + resample()来做到这一点，但是我不够聪明，经过几个小时的修修补补。在这里向明智的人寻求帮助......

所以这是示例数据框：

df = pd.DataFrame([
['Team A', pd.datetime(2017, 12, 1), 0, 2]
,['Team A', pd.datetime(2017, 12, 1), 2, 1]
,['Team A', pd.datetime(2017, 12, 1), 4, 1]
,['Team A', pd.datetime(2017, 12, 8), 3, 2]
,['Team B', pd.datetime(2017, 12, 1), 0, 1]
,['Team B', pd.datetime(2017, 12, 1), 2, 3]
,['Team B', pd.datetime(2017, 12, 8), 1, 3]
,['Team B', pd.datetime(2017, 12, 8), 3, 2]
]
, columns=['Team', 'LeadWeek', 'ConversionWeek', 'Conversions']
)

我想要的输出如下，每个Team / LeadWeek分组有5个“ConversionWeek”行，编号为0到4：

       Team     LeadWeek     ConversionWeek     Conversions
0      Team A     2017-12-01     0     2.0
1      Team A     2017-12-01     1     0.0
2      Team A     2017-12-01     2     1.0
3      Team A     2017-12-01     3     0.0
4      Team A     2017-12-01     4     1.0
5      Team A     2017-12-08     0     0.0
6      Team A     2017-12-08     1     0.0
7      Team A     2017-12-08     2     0.0
8      Team A     2017-12-08     3     2.0
9      Team A     2017-12-08     4     0.0
10     Team B     2017-12-01     0     1.0
11     Team B     2017-12-01     1     0.0
12     Team B     2017-12-01     2     3.0
13     Team B     2017-12-01     3     0.0
14     Team B     2017-12-01     4     0.0
15     Team B     2017-12-08     0     0.0
16     Team B     2017-12-08     1     3.0
17     Team B     2017-12-08     2     0.0
18     Team B     2017-12-08     3     2.0
19     Team B     2017-12-08     4     0.0

我确实有一个解决方案，但它不是非常pythonic。它与我在SQL中如何解决它是一样的，即使用所有不同元素的笛卡尔积来创建一个“脚手架”，然后将我的实际转换加入到它中。在Python中，此方法使用itertools.product()

我的解决方案是：

import pandas as pd
import numpy as np
import itertools as it

df = pd.DataFrame([
['Team A', pd.datetime(2017, 12, 1), 0, 2]
,['Team A', pd.datetime(2017, 12, 1), 2, 1]
,['Team A', pd.datetime(2017, 12, 1), 4, 1]
,['Team A', pd.datetime(2017, 12, 8), 3, 2]
,['Team B', pd.datetime(2017, 12, 1), 0, 1]
,['Team B', pd.datetime(2017, 12, 1), 2, 3]
,['Team B', pd.datetime(2017, 12, 8), 1, 3]
,['Team B', pd.datetime(2017, 12, 8), 3, 2]
]
, columns=['Team', 'LeadWeek', 'ConversionWeek', 'Conversions']
)

ConversionWeek = np.linspace(0, 4, 5, dtype=int)

Team = df['Team'].unique()

LeadWeek = df['LeadWeek'].unique()

scaffold_raw = []

for i in it.product(Team, LeadWeek, ConversionWeek):
    scaffold_raw.append(i)

scaffold = pd.DataFrame(scaffold_raw, columns=['Team', 'LeadWeek', 'ConversionWeek'])

new_frame = scaffold.merge(df, how='left')

new_frame = new_frame.sort_values(by=['Team', 'LeadWeek', 'ConversionWeek']).reset_index(drop=True)

new_frame['Conversions'].fillna(0, inplace=True)

在更优雅的解决方案上获得任何帮助。

Answer 1

通过reindex使用pd.MultiIndex -

idx = pd.MultiIndex.from_product(
      [df.Team.unique(), df.LeadWeek.unique(), np.arange(5)]
)   

v = df.set_index(['Team', 'LeadWeek', 'ConversionWeek'])\
      .reindex(idx)\
      .fillna(0)\
      .reset_index()

v.columns = df.columns    
v

      Team   LeadWeek  ConversionWeek  Conversions
0   Team A 2017-12-01               0          2.0
1   Team A 2017-12-01               1          0.0
2   Team A 2017-12-01               2          1.0
3   Team A 2017-12-01               3          0.0
4   Team A 2017-12-01               4          1.0
5   Team A 2017-12-08               0          0.0
6   Team A 2017-12-08               1          0.0
7   Team A 2017-12-08               2          0.0
8   Team A 2017-12-08               3          2.0
9   Team A 2017-12-08               4          0.0
10  Team B 2017-12-01               0          1.0
11  Team B 2017-12-01               1          0.0
12  Team B 2017-12-01               2          3.0
13  Team B 2017-12-01               3          0.0
14  Team B 2017-12-01               4          0.0
15  Team B 2017-12-08               0          0.0
16  Team B 2017-12-08               1          3.0
17  Team B 2017-12-08               2          0.0
18  Team B 2017-12-08               3          2.0
19  Team B 2017-12-08               4          0.0