'mode' not recognized by df.groupby().agg(), but pd.Series.mode works

[patricksurry]

This works:

df = pd.DataFrame({'A': [1, 2, 1, 2, 1, 2, 3], 'B': [1, 1, 1, 2, 2, 2, 2]})
df.groupby('B').agg(pd.Series.mode)

but this doesn't:

df.groupby('B').agg('mode')

...
AttributeError: Cannot access callable attribute 'mode' of 'DataFrameGroupBy' objects, try using the 'apply' method

I thought all the series aggregate methods propagated automatically to groupby, but I've probably misunderstood?

[patricksurry]

Hmm, I guess this might be because pd.Series.mode() returns a series, not a scalar. So maybe I need my own mode that decides how to handle the multi-modal case, e.g. pd.Series.mode().mean() or whatever?

[TomAugspurger]

might be because pd.Series.mode() returns a series, not a scalar

Correct. IIRC there's an older issue about this, where we decided to keep our behavior of always returning a series, and not adding a flag to reduce if possible. I could be misremembering though.

In these cases I'll usually just use scipy's

df.groupby('B').agg(lambda x: scipy.stats.mode(x)[0])

scipy.stats.mode returns a tuple of (mode, count) and we just want the mode.

[jreback]

Here's a mini-example; could be like .value_counts()

In [6]: df = DataFrame({'A' : [1,2,1,2], 'B' : [1,1,1,1]})

In [7]: df
Out[7]: 
   A  B
0  1  1
1  2  1
2  1  1
3  2  1

In [8]: df.groupby('A').B.value_counts()
Out[8]: 
A  B
1  1    2
2  1    2
Name: B, dtype: int64

In [9]: df.groupby('A').B.apply(lambda x: x.mode())
Out[9]: 
A   
1  0    1
2  0    1
Name: B, dtype: int64

[kernc]

What about when grouping Series?

I have no issue with .agg('mode') returning the first mode, if any, while issuing a warning if the modes were multuple.

[gosuto-inzasheru]

I encountered this problem and ended up settling for this:
.agg({'column': lambda x: pd.Series.mode(x)[0][0]})

But yes, I agree with @kernc, I would not mind .agg('mode') returning the first mode if multiple modes are returned.