how to apply a function to each row in a dataframe and get a series of dicts?

Question

I want to apply a function to each row in a dataframe and get a series. It works if the function returns a number, but does not work when the function returns a dict.

In [31]: d
Out[31]: 
                a         b
bar one -0.185677 -0.554356
    two -0.457943 -1.094836
baz one -0.731338 -0.027821
    two -1.061098  0.258291
foo one -1.392160  2.287989
    two  2.010208 -1.350581
qux one -0.792229 -0.323397
    two -1.063265  0.048641

In [32]: d.apply(lambda x: x.a+x.b, axis=1)
Out[32]: 
bar  one   -0.740034
     two   -1.552779
baz  one   -0.759159
     two   -0.802806
foo  one    0.895829
     two    0.659626
qux  one   -1.115627
     two   -1.014624
dtype: float64

In [33]: d.apply(lambda x: {"boo": x.a}, axis=1)// I want a series of dict
Out[33]: 
          a   b
bar one NaN NaN
    two NaN NaN
baz one NaN NaN
    two NaN NaN
foo one NaN NaN
    two NaN NaN
qux one NaN NaN
    two NaN NaN

It does not seem to matter if the reduce argument of apply is None/True/False. Pandas seems to be too smart by accessing value of keys "a" and "b" in the dict {"boo":xa}.

Answer 1

An answer is to not try to perform the calculation and data type translation in one step. Calculate a series of values, then restructure into the dictionary.

import pandas
import random

mi1 = ['bar','baz']
mi2 = ['one', 'two']
data = [[s,t, random.random(), random.random()] for s in mi1 for t in mi2]
df = pandas.DataFrame(data, columns=['i1', 'i2', 'a', 'b'])
df.set_index(['i1', 'i2'], inplace=True)
print(df)

Out

            a         b
i1  i2                     
bar one  0.438596  0.734058
    two  0.183522  0.272922
baz one  0.581694  0.522173
    two  0.776081  0.941120

In

# calculate the series
# the data type is still floats
sr = df.apply(lambda row: row.a + row.b, axis=1)

# construct the series of objects
result = sr.apply(lambda x: {"boo":x})
print(result)

Out

i1   i2 
bar  one     {'boo': 1.17265404527}
     two    {'boo': 0.456443829892}
baz  one     {'boo': 1.10386719117}
     two     {'boo': 1.71720149706}
dtype: object