Pandas 将多列分组聚合作为用户定义函数的输入

Question

I am still trying to learn pandas. I have a custom user defined function which will require two columns as input. It is an aggregation function so it needs to be done by group.

This is my question: How can I get a grouped aggregation with multiple columns as inputs to a user defined function?

Here is my reproducible example along with a couple things I tried.

import pandas as pd
import numpy as np


def first_b_over_avg_c(b,c):
    first_b = b.first()
    avg_c = np.mean(c)
    return first_b / avg_c


np.random.seed(42)
df = pd.DataFrame(
        {
            "a": ["one", "one", "one", "one", "two", "two", "two", "two"],
            "b": np.random.uniform(0,1,8),
            "c": np.random.uniform(0,1,8)
            }
        )
print(df)

df.groupby(['a'],as_index = False).assign(d = lambda df: first_b_over_avg_c(df['b'],df['c']))
df.groupby(['a'],as_index = False).apply(first_b_over_avg_c, b=('b'), c=('c'))

Here is the output:

     a         b         c
0  one  0.374540  0.601115
1  one  0.950714  0.708073
2  one  0.731994  0.020584
3  one  0.598658  0.969910
4  two  0.156019  0.832443
5  two  0.155995  0.212339
6  two  0.058084  0.181825
7  two  0.866176  0.183405

And the error

Traceback (most recent call last): File "reprex.py", line 21, in df.groupby(['a'],as_index = False).assign(d = lambda df: first_b_over_avg_c(df['b'],df['c'])) File "/home/ubuntu/.local/lib/python3.8/site-packages/pandas/core/groupby/groupby.py", line 703, in getattr raise AttributeError( AttributeError: 'DataFrameGroupBy' object has no attribute 'assign'

Answer 1

I think this does what you are looking for

import pandas as pd
import numpy as np


def first_b_over_avg_c(group):
    first_b = group['b'].iloc[0]
    avg_c = np.mean(group['c'])
    return first_b / avg_c


np.random.seed(42)
df = pd.DataFrame(
        {
            "a": ["one", "one", "one", "one", "two", "two", "two", "two"],
            "b": np.random.uniform(0,1,8),
            "c": np.random.uniform(0,1,8)
            }
        )

df.groupby(['a'],as_index = False).apply(first_b_over_avg_c)

If I am reading correctly all you want to do is be able to access multiple columns from a user defined function.

In this example the entire row/group is being passed into the function. If you print out group in the function:

     a         b         c
0  one  0.374540  0.601115
1  one  0.950714  0.708073
2  one  0.731994  0.020584
3  one  0.598658  0.969910

you can see that the groups 'one' and 'two' get passed separately.

Had the object not been grouped. Each row would get passed separately.

I do not see a point in assigning and passing the columns in separately unless there is a specific reason for it.