我有以下 df.当满足 2 个条件时，需要将“数值”的值替换为“类型_1 的 4%”的计算:类型 = 类型_2 且数值 > 0
df_in = pd.DataFrame([['Jan-2021','P_1','Type_1','5'],['Jan-2021','P_1','Type_2','10'],['Jan-2021','P_1','Type_3','15'],
['Feb-2021','P_1','Type_1','7'],['Feb-2021','P_1','Type_2','0'],['Feb-2021','P_1','Type_3','21'],
['Mar-2021','P_2','Type_1','3'],['Mar-2021','P_2','Type_2','6'],['Mar-2021','P_2','Type_3','9']],
columns = ['Month-Yr','Product','Type','Numerical'])
Month-Yr Product Type Numerical
0 Jan-2021 P_1 Type_1 5
1 **Jan-2021 P_1 Type_2 10**
2 Jan-2021 P_1 Type_3 15
3 Feb-2021 P_1 Type_1 7
4 **Feb-2021 P_1 Type_2 0**
5 Feb-2021 P_1 Type_3 21
6 Mar-2021 P_2 Type_1 3
7 **Mar-2021 P_2 Type_2 6**
8 Mar-2021 P_2 Type_3 9
预期结果:
Month-Yr Product Type Numerical
0 Jan-2021 P_1 Type_1 5
1 **Jan-2021 P_1 Type_2 0.2**
2 Jan-2021 P_1 Type_3 15
3 Feb-2021 P_1 Type_1 7
4 Feb-2021 P_1 Type_2 0
5 Feb-2021 P_1 Type_3 21
6 Mar-2021 P_2 Type_1 3
7 **Mar-2021 P_2 Type_2 0.12**
8 Mar-2021 P_2 Type_3 9
最佳答案
一种选择是将数据翻转为水平形式，然后再计算条件，最后翻转回垂直形式；这应该比 for 循环更快，因为它是矢量化的并且依赖于 Pandas 方法:
# you can ignore this,
# if you are not fuzzy on the order
from pandas.api.types import CategoricalDtype
categories = CategoricalDtype(categories = df_in['Month-Yr'].unique(),
ordered = True)
temp = (df_in
# need the column as numbers
.astype({'Numerical': int, 'Month-Yr':categories})
.pivot(index=['Month-Yr', 'Product'],
columns='Type',
values='Numerical')
# conditional statement here,
# similar to python's if-else statement
.assign(Type_2 = lambda df: np.where(df.Type_2 > 0,
df.Type_1 * 0.04,
df.Type_2))
.stack()
.reset_index(name='NUmerical')
)
Month-Yr Product Type Numerical
0 Jan-2021 P_1 Type_1 5.00
1 Jan-2021 P_1 Type_2 0.20
2 Jan-2021 P_1 Type_3 15.00
3 Feb-2021 P_1 Type_1 7.00
4 Feb-2021 P_1 Type_2 0.00
5 Feb-2021 P_1 Type_3 21.00
6 Mar-2021 P_2 Type_1 3.00
7 Mar-2021 P_2 Type_2 0.12
8 Mar-2021 P_2 Type_3 9.00
关于具有多个条件和计算的 Pandas 数据框，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/70614339/