Question

我有一个包含Groups，两个日期和值的数据框。

我想要一个数据帧的子集，它为每个GRP保留所有行的唯一B_DATE值。如果每个组中存在重复的B_DATE值，我希望保留具有最大A_DATE值的行。

所以，如果我的初始数据框是：

GRP  A_DATE        B_DATE   VALUE
A   12/31/2012  2/19/2014   546.2
A   12/31/2013  2/19/2014   543.7
A   3/31/2013   4/30/2014   473.3
A   3/31/2014   4/30/2014   472.5
A   6/30/2013   7/30/2014   528.7
A   6/30/2014   7/30/2014   531.5
A   9/30/2013   10/30/2014  529
A   9/30/2014   10/30/2014  546.7
A   12/31/2014  2/18/2015   573.5
A   3/31/2015   4/30/2015   458.7
A   6/30/2015   7/30/2015   519.5
B   3/31/2014   7/7/2015    1329
B   12/31/2014  7/7/2015    1683
B   3/31/2015   7/7/2015    1361
B   6/30/2014   8/13/2015   1452
B   6/30/2015   8/13/2015   1429
B   9/30/2014   10/29/2015  1488
B   9/30/2015   10/29/2015  1595
B   12/31/2015  2/16/2016   1763
B   3/31/2016   4/28/2016   1548

我希望结果看起来像这样：

GRP  A_DATE        B_DATE   VALUE
A   12/31/2013  2/19/2014   543.7
A   3/31/2014   4/30/2014   472.5
A   6/30/2014   7/30/2014   531.5
A   9/30/2014   10/30/2014  546.7
A   12/31/2014  2/18/2015   573.5
A   3/31/2015   4/30/2015   458.7
A   6/30/2015   7/30/2015   519.5
B   3/31/2015   7/7/2015    1361
B   6/30/2015   8/13/2015   1429
B   9/30/2015   10/29/2015  1595
B   12/31/2015  2/16/2016   1763
B   3/31/2016   4/28/2016   1548

我知道如何通过繁琐的循环和使用argmax（）来做到这一点。然而，想知道是否有一种“干净”，高效，Pythonic的方式来接近。

提前致谢。

Answer 1

让我们使用sort_values和drop_duplicates：

df.sort_values(['GRP','A_DATE'], ascending=[True,False])\
  .drop_duplicates(subset=['GRP','B_DATE'])

输出：

   GRP      A_DATE      B_DATE   VALUE
7    A   9/30/2014  10/30/2014   546.7
10   A   6/30/2015   7/30/2015   519.5
5    A   6/30/2014   7/30/2014   531.5
9    A   3/31/2015   4/30/2015   458.7
3    A   3/31/2014   4/30/2014   472.5
8    A  12/31/2014   2/18/2015   573.5
1    A  12/31/2013   2/19/2014   543.7
17   B   9/30/2015  10/29/2015  1595.0
15   B   6/30/2015   8/13/2015  1429.0
19   B   3/31/2016   4/28/2016  1548.0
13   B   3/31/2015    7/7/2015  1361.0
18   B  12/31/2015   2/16/2016  1763.0

并添加sort_index以取回原始订单：

df.sort_values(['GRP','A_DATE'], ascending=[True,False])\
  .drop_duplicates(subset=['GRP','B_DATE']).sort_index()

   GRP      A_DATE      B_DATE   VALUE
1    A  12/31/2013   2/19/2014   543.7
3    A   3/31/2014   4/30/2014   472.5
5    A   6/30/2014   7/30/2014   531.5
7    A   9/30/2014  10/30/2014   546.7
8    A  12/31/2014   2/18/2015   573.5
9    A   3/31/2015   4/30/2015   458.7
10   A   6/30/2015   7/30/2015   519.5
13   B   3/31/2015    7/7/2015  1361.0
15   B   6/30/2015   8/13/2015  1429.0
17   B   9/30/2015  10/29/2015  1595.0
18   B  12/31/2015   2/16/2016  1763.0
19   B   3/31/2016   4/28/2016  1548.0

Answer 2

我认为你想分组B_DATE和'GRP'汇总最后一个值，即

df['A_DATE'] = pd.to_datetime(df['A_DATE'])
df['B_DATE'] = pd.to_datetime(df['B_DATE'])

ndf = df.groupby(['GRP',df['B_DATE']]).agg('last').reset_index()

     GRP     B_DATE     A_DATE   VALUE
0    A 2014-02-19 2013-12-31   543.7
1    A 2014-04-30 2014-03-31   472.5
2    A 2014-07-30 2014-06-30   531.5
3    A 2014-10-30 2014-09-30   546.7
4    A 2015-02-18 2014-12-31   573.5
5    A 2015-04-30 2015-03-31   458.7
6    A 2015-07-30 2015-06-30   519.5
7    B 2015-07-07 2015-03-31  1361.0
8    B 2015-08-13 2015-06-30  1429.0
9    B 2015-10-29 2015-09-30  1595.0
10   B 2016-02-16 2015-12-31  1763.0
11   B 2016-04-28 2016-03-31  1548.0

Python Pandas：来自Dataframe的复杂子集

2 个答案: