pd.core.groupby.Groupby.cumsum
¶
pandas.core.groupby.Groupby.cumsum(axis=0)
Note
cumsum
is only supported on numeric columns and is not supported on boolean columns
Example Usage¶
>>> @bodo.jit
... def f(df):
... return df.groupby("B").cumsum()
>>> df = pd.DataFrame(
... {
... "A": [1, 2, 24, None] * 5,
... "B": ["421", "f31"] * 10,
... "C": [1.51, 2.421, 233232, 12.21] * 5
... }
... )
>>> f(df)
A C
0 1.0 1.510
1 2.0 2.421
2 25.0 233233.510
3 NaN 14.631
4 26.0 233235.020
5 4.0 17.052
6 50.0 466467.020
7 NaN 29.262
8 51.0 466468.530
9 6.0 31.683
10 75.0 699700.530
11 NaN 43.893
12 76.0 699702.040
13 8.0 46.314
14 100.0 932934.040
15 NaN 58.524
16 101.0 932935.550
17 10.0 60.945
18 125.0 1166167.550
19 NaN 73.155