get first and last values in a groupby

Question

I have a dataframe df

df = pd.DataFrame(np.arange(20).reshape(10, -1),
                  [[\'a\', \'a\', \'a\', \'a\', \'b\', \'b\', \'b\', \'c\         


        

Answer 1

Option 1

def first_last(df):
    return df.ix[[0, -1]]

df.groupby(level=0, group_keys=False).apply(first_last)

---

Option 2 - only works if index is unique

idx = df.index.to_series().groupby(level=0).agg(['first', 'last']).stack()
df.loc[idx]

---

Option 3 - per notes below, this only makes sense when there are no NAs

I also abused the agg function. The code below works, but is far uglier.

df.reset_index(1).groupby(level=0).agg(['first', 'last']).stack() \
    .set_index('level_1', append=True).reset_index(1, drop=True) \
    .rename_axis([None, None])

---

Note

per @unutbu: agg(['first', 'last']) take the firs non-na values.

I interpreted this as, it must then be necessary to run this column by column. Further, forcing index level=1 to align may not even make sense.

Let's include another test

df = pd.DataFrame(np.arange(20).reshape(10, -1),
                  [list('aaaabbbccd'),
                   list('abcdefghij')],
                  list('XY'))

df.loc[tuple('aa'), 'X'] = np.nan

---

def first_last(df):
    return df.ix[[0, -1]]

df.groupby(level=0, group_keys=False).apply(first_last)

df.reset_index(1).groupby(level=0).agg(['first', 'last']).stack() \
    .set_index('level_1', append=True).reset_index(1, drop=True) \
    .rename_axis([None, None])

Sure enough! This second solution is taking the first valid value in column X. It is now nonsensical to have forced that value to align with the index a.