piRSquared piRSquared - 2 months ago 15
Python Question

convenient way to reindex one level of a multiindex

consider the

pd.Series
s
and
pd.MultiIndex
idx


idx = pd.MultiIndex.from_product([list('AB'), [1, 3], list('XY')],
names=['one', 'two', 'three'])
s = pd.Series(np.arange(8), idx)
s

one two three
A 1 X 0
Y 1
3 X 2
Y 3
B 1 X 4
Y 5
3 X 6
Y 7
dtype: int32


I want to
reindex
on
level='two'
with
np.arange(4)


I can do it with:

s.unstack([0, 2]).reindex(np.arange(4), fill_value=0).stack().unstack([0, 1])

one two three
A 0 X 0
Y 0
1 X 0
Y 1
2 X 0
Y 0
3 X 2
Y 3
B 0 X 0
Y 0
1 X 4
Y 5
2 X 0
Y 0
3 X 6
Y 7
dtype: int32


But I'm looking for something more direct if it exists. Any ideas?

Answer

Unfortunately if need reindex with MultiIndex, need all levels:

mux = pd.MultiIndex.from_product([list('AB'), np.arange(4), list('XY')],
                                 names=['one', 'two', 'three'])

print (s.reindex(mux, fill_value=0))
one  two  three
A    0    X        0
          Y        0
     1    X        0
          Y        1
     2    X        0
          Y        0
     3    X        2
          Y        3
B    0    X        0
          Y        0
     1    X        4
          Y        5
     2    X        0
          Y        0
     3    X        6
          Y        7
dtype: int32

EDIT by comment:

idx = pd.MultiIndex.from_tuples([('A', 1, 'X'), ('B', 3, 'Y')],
                                 names=['one', 'two', 'three'])
s = pd.Series([5,6], idx)
print (s)
one  two  three
A    1    X        5
B    3    Y        6
dtype: int64

mux = pd.MultiIndex.from_tuples([('A', 0, 'X'), ('A', 1, 'X'), 
                                 ('A', 2, 'X'), ('A', 3, 'X'), 
                                 ('B', 0, 'Y'), ('B', 1, 'Y'), 
                                 ('B', 2, 'Y'), ('B', 3, 'Y')],
                                 names=['one', 'two', 'three'])

print (s.reindex(mux, fill_value=0))
one  two  three
A    0    X        0
     1    X        5
     2    X        0
     3    X        0
B    0    Y        0
     1    Y        0
     2    Y        0
     3    Y        6
dtype: int64

Direct Solution

new_lvl = np.arange(4)
mux = [(a, b, c) for b in new_lvl for a, c in s.reset_index('two').index.unique()]
s.reindex(mux, fill_value=0).sort_index()

one  two  three
A    0    X        0
          Y        0
     1    X        0
          Y        1
     2    X        0
          Y        0
     3    X        2
          Y        3
B    0    X        0
          Y        0
     1    X        4
          Y        5
     2    X        0
          Y        0
     3    X        6
          Y        7
dtype: int64
Comments