Groupby Apply Custom Function Pandas

Question

I'm trying to apply a custom function in pandas similar to the groupby and mutate functionality in dplyr.

What I'm trying to do is say given a pandas dataframe like this:

df = pd.DataFrame({'category1':['a','a','a', 'b', 'b','b'],
  'category2':['a', 'b', 'a', 'b', 'a', 'b'],
  'var1':np.random.randint(0,100,6),
  'var2':np.random.randint(0,100,6)}
)

df
  category1 category2  var1  var2
0         a         a    23    59
1         a         b    54    20
2         a         a    48    62
3         b         b    45    76
4         b         a    60    26
5         b         b    13    70

apply some function that returns the same number of elements as the number of elements in the group by:

def myfunc(s):
  return [np.mean(s)] * len(s)

to get this result

df
  category1 category2  var1  var2   var3
0         a         a    23    59   35.5
1         a         b    54    20   54
2         a         a    48    62   35.5
3         b         b    45    76   29
4         b         a    60    26   60
5         b         b    13    70   29

I was thinking of something along the lines of:

df['var3'] = df.groupby(['category1', 'category2'], group_keys=False).apply(lambda x: myfunc(x.var1))

but haven't been able to get the index to match.

In R with dplyr this would be

df <- df %>%
  group_by(category1, category2) %>%
  mutate(
    var3 = myfunc(var1)
  )

So I was able to solve it by using a custom function like:

def myfunc_data(data):

  data['var3'] = myfunc(data.var1)
  return data

and

df = df.groupby(['category1', 'category2']).apply(myfunc_data)

but I guess I was still wondering if there's a way to do it without defining this custom function.

Answer 1

Use GroupBy.transform for return Series with same size like original DataFrame, so possible assign to new column:

np.random.seed(123)

df = pd.DataFrame({'category1':['a','a','a', 'b', 'b','b'],
  'category2':['a', 'b', 'a', 'b', 'a', 'b'],
  'var1':np.random.randint(0,100,6),
  'var2':np.random.randint(0,100,6)}
)

df['var3'] = df.groupby(['category1', 'category2'])['var1'].transform(myfunc)
print (df)
  category1 category2  var1  var2  var3
0         a         a    66    86    82
1         a         b    92    97    92
2         a         a    98    96    82
3         b         b    17    47    37
4         b         a    83    73    83
5         b         b    57    32    37

Alternative with lambda function:

df['var3'] = (df.groupby(['category1', 'category2'])['var1']
                .transform(lambda s: [np.mean(s)] * len(s)))