熊猫组过滤问题

Question

I cannot for the life of me figure out why the filter method refuses to work on my dataframes in pandas.

Here is an example showing my issue:

In [99]: dff4
Out[99]: <pandas.core.groupby.DataFrameGroupBy object at 0x1143cbf90>

In [100]: dff3
Out[100]: <pandas.core.groupby.DataFrameGroupBy object at 0x11439a810>

In [101]: dff3.groups
Out[101]: 
{'iphone': [85373, 85374],
 'remote_api_created': [85363,
  85364,
  85365,
  85412]}

In [102]: dff4.groups
Out[102]: {'bye': [3], 'bye bye': [4], 'hello': [0, 1, 2]}

In [103]: dff4.filter(lambda x: len(x) >2)
Out[103]: 
   A      B
0  0  hello
1  1  hello
2  2  hello

In [104]: dff3.filter(lambda x: len(x) >2)
Out[104]: 
Empty DataFrame
Columns: [source]
Index: []

Notice how filter refuses to work on dff3.

Any help appreciated.

Answer 1

If you group by column name, you move it to index, so your dataframe becomes empty, if no other columns is present, see:

>>> def report(x):
...     print x
...     return True
>>> df
                   source
85363  remote_api_created
85364  remote_api_created
85365  remote_api_created
85373              iphone
85374              iphone
85412  remote_api_created

>>> df.groupby('source').filter(report)
Series([], dtype: float64)
Empty DataFrame
Columns: []
Index: [85373, 85374]
Series([], dtype: float64)
Empty DataFrame
Columns: [source]
Index: []

You can group by column values:

>>> df.groupby(df['source']).filter(lambda x: len(x)>2)
                   source
85363  remote_api_created
85364  remote_api_created
85365  remote_api_created
85412  remote_api_created