groupby transform valueerror length of passed values-CodePudding

Hello can you help me understand what is the issue here and how to solve it?

dft = pd.DataFrame({'B': [0, 1, 2, 5, 4, 7, 2, 2, 2, 5, 6, 7]})
dft['user'] = ['a','b','c','b','a','c','a','b','b','c','a', 'c']

dft.groupby('user')['B'].transform(lambda row: row.ewm(span=2)).mean()

gives ValueError: Length of passed values is 1, index implies 4.

CodePudding user response：

I suppose that invocation of mean() should be a part of your lambda function.

So maybe your code should be:

dft.groupby('user')['B'].transform(lambda row: row.ewm(span=2).mean())

For your sample data I got:

0     0.000000
1     1.000000
2     2.000000
3     4.000000
4     3.000000
5     5.750000
6     2.307692
7     2.615385
8     2.200000
9     5.230769
10    4.800000
11    6.425000
Name: B, dtype: float64

Another hint is that transform invokes the passed function to each column, not row. So argument named row is in this case misleading.

CodePudding user response：

Use ewm on the groupby directly, support has been added for a while now:

dft = pd.DataFrame({'B': [0, 1, 2, 5, 4, 7, 2, 2, 2, 5, 6, 7]})
dft['user'] = ['a','b','c','b','a','c','a','b','b','c','a', 'c']

dft.groupby('user')['B'].ewm(span=2).mean()

output:

user    
a     0     0.000000
      4     3.000000
      6     2.307692
      10    4.800000
b     1     1.000000
      3     4.000000
      7     2.615385
      8     2.200000
c     2     2.000000
      5     5.750000
      9     5.230769
      11    6.425000
Name: B, dtype: float64