非重叠滚动 windows 在 pandas groupby

Question

I want to create non-overlapping rolling or sliding window in pandas groupby

import pandas as pd
df1 = pd.DataFrame( {'a1':['A','A','B','B','B','B','B','B'],'a2':[1,1,1,2,2,2,2,2], 'b':[1,2,5,5,5,4,6,2]})

For overlapping rolling window, I can do this

df1.groupby(['a1','a2']).rolling(2).mean()

But is there any way to make it non-overlapping?

The output should be like this

pd.DataFrame('a1':['A','B','B','B','B'],'a2':[1,1,2,2,2],'b':[1.5,NaN,5,5,NaN])

Explanation

When a1 is A and a2 is 1 , the value of b is 1 and 2 . Adding both results in 1.5 .
When a1 is B and a2 is 1 , the value of b is 5 . As the value of b is less than the length of the sliding window, we got NaN .
When a1 is B and a2 is 2 , the value of b is 5,5,4,6,2 . As sliding window is 2 , so adding (5+5)/2=5 , (4+6)/2=5 . And last value is NaN as length is less than sliding window.

Answer 1

Well, one approach (not very elegant), is to do:

def non_overlapping_mean(x, window=2):
    return x.groupby(np.arange(len(x)) // window).apply(lambda x: np.nan if len(x) < 2 else x.mean())


res = df1.groupby(['a1', 'a2'])['b'].apply(non_overlapping_mean).droplevel(-1).reset_index()
print(res)

Output

  a1  a2    b
0  A   1  1.5
1  B   1  NaN
2  B   2  5.0
3  B   2  5.0
4  B   2  NaN

The main idea is to groupby into consecutive chunks, and is done here:

x.groupby(np.arange(len(x)) // window)