有没有办法在满足特定条件之前使用 Python 的 cumsum()？

编程问答 2022-05-03

问题描述

我想使用 cumsum() 进行累计，直到满足特定条件。就我而言，我想累积值直到看到负值，然后重置为 0。这是我的数据集；

Name    Date    Amount
ABC 4/30/2020   0
ABC 5/31/2020   2500
ABC 6/30/2020   0
ABC 7/31/2020   0
ABC 8/31/2020   0
ABC 9/30/2020   0
ABC 10/31/2020  0
ABC 11/30/2020  0
ABC 12/31/2020  -1925
ABC 1/31/2021   0
ABC 2/28/2021   0
ABC 3/31/2021   0

我尝试使用 df['Rolling_Amount'] = df.groupby(['Name'])['Amount'].cumsum()。但它在看到负数后不会重置。这是我得到的；

Name    Date    Amount  Rolling_Amount
ABC 4/30/2020   0       0
ABC 5/31/2020   2500    2500
ABC 6/30/2020   0       2500
ABC 7/31/2020   0       2500
ABC 8/31/2020   0       2500
ABC 9/30/2020   0       2500
ABC 10/31/2020  0       2500
ABC 11/30/2020  0       2500
ABC 12/31/2020  -1925   575
ABC 1/31/2021   0       575
ABC 2/28/2021   0       575
ABC 3/31/2021   0       575

但是，我想在 Amount -1925 年之后将我的计数器重置为 0。预期的输出应该是这样的

ABC 4/30/2020   0       0
ABC 5/31/2020   2500    2500
ABC 6/30/2020   0       2500
ABC 7/31/2020   0       2500
ABC 8/31/2020   0       2500
ABC 9/30/2020   0       2500
ABC 10/31/2020  0       2500
ABC 11/30/2020  0       2500
ABC 12/31/2020  -1925   0
ABC 1/31/2021   0       0
ABC 2/28/2021   0       0
ABC 3/31/2021   0       0

解决方法

让我们尝试创建一个布尔索引来检测数字何时从正数变为负数，或何时从负数变为正数以分隔组，然后获取每个组的总和：

m = df.Amount.replace(to_replace=0,method='ffill')
df['Rolling_Amount'] = df.Amount \
    .groupby((m < 0).eq((m > 0).shift()).cumsum()) \
    .cumsum() \
    .mask(lambda s: s < 0,0)

样本输入：

Name    Date    Amount
ABC 4/30/2020   0
ABC 5/31/2020   2500
ABC 6/30/2020   0
ABC 7/31/2020   50
ABC 8/31/2020   0
ABC 9/30/2020   -1925
ABC 10/31/2020  0
ABC 11/30/2020  100
ABC 12/31/2020  0
ABC 1/31/2021   200
ABC 2/28/2021   0
ABC 3/31/2021   0

输出：

   Name        Date  Amount  Rolling_Amount
0   ABC   4/30/2020       0               0
1   ABC   5/31/2020    2500            2500
2   ABC   6/30/2020       0            2500
3   ABC   7/31/2020      50            2550
4   ABC   8/31/2020       0            2550
5   ABC   9/30/2020   -1925               0
6   ABC  10/31/2020       0               0
7   ABC  11/30/2020     100             100
8   ABC  12/31/2020       0             100
9   ABC   1/31/2021     200             300
10  ABC   2/28/2021       0             300
11  ABC   3/31/2021       0             300
cumsumpandaspandaspython