Pandas: get string value with most occurrence in group

Question

I have the following DataFrame:

item    response
1       A       
1       A       
1       B       
2       A       
2       A   

I want to add a column with the most given response for an item. which should result in:

item    response  mostGivenResponse
1       A          A
1       A          A      
1       B          A       
2       C          C
2       C          C

I tried something like this:

df["responseCount"] = df.groupby(["ItemCode", "Response"])["Response"].transform("count")

df["mostGivenResponse"] = df.groupby(['ItemCode'])['responseCount'].transform(max)

But mostGivenResponse is now the count of the response in stead of the response itself.

Answer 1

There is pd.Series.mode:

df.groupby('item').response.transform(pd.Series.mode)
Out[28]: 
0    A
1    A
2    A
3    C
4    C
Name: response, dtype: object

Answer 2

Use value_counts and return first index value:

df["responseCount"] = (df.groupby("item")["response"]
                        .transform(lambda x: x.value_counts().index[0]))

print (df)
   item response responseCount
0     1        A             A
1     1        A             A
2     1        B             A
3     2        C             C
4     2        C             C

Or collections.Counter.most_common:

from collections import Counter

df["responseCount"] = (df.groupby("item")["response"]
                         .transform(lambda x: Counter(x).most_common(1)[0][0]))

print (df)
   item response responseCount
0     1        A             A
1     1        A             A
2     1        B             A
3     2        C             C
4     2        C             C

EDIT:

Problem is with one or multiple NaNs only groups, solution is filter with if-else:

print (df)
   item response
0     1        A
1     1        A
2     2      NaN
3     2      NaN
4     3      NaN

def f(x):
    s = x.value_counts()
    print (s)

    A    2
    Name: 1, dtype: int64
    Series([], Name: 2, dtype: int64)
    Series([], Name: 3, dtype: int64)

    #return np.nan if s.empty else s.index[0]
    return np.nan if len(s) == 0 else s.index[0]

df["responseCount"] = df.groupby("item")["response"].transform(f)
print (df)
   item response responseCount
0     1        A             A
1     1        A             A
2     2      NaN           NaN
3     2      NaN           NaN
4     3      NaN           NaN