Assignment with MultiIndex replaces dataframe contents with NaNs

[floux]

Bug:
df.loc['bar'] = df['bar'] * 2; assignment with rhs having a multi-index

Related to http://stackoverflow.com/questions/16833842/assign-new-values-to-slice-from-multiindex-dataframe

In [246]: arrays = [np.array(['bar', 'bar', 'baz', 'qux', 'qux', 'bar']),
                  np.array(['one', 'two', 'one', 'one', 'two', 'one']),
                  np.arange(0, 6, 1)]

In [247]: df = pd.DataFrame(randn(6, 3),
   .....: index=arrays,
   .....: columns=['A', 'B', 'C']).sort_index()

In [248]: df
Out[248]:
                  A         B         C
bar one 0  2.186362 -1.749287 -0.776363
        5  0.953196 -0.717784 -0.528991
    two 1  0.692802 -0.166100  1.403452
baz one 2 -0.065485  0.642380 -1.154436
qux one 3  1.515987  1.138678 -0.690564
    two 4  0.261254 -0.066171 -1.352841

In [249]: df.loc['bar']
Out[249]:
              A         B         C
one 0  2.186362 -1.749287 -0.776363
    5  0.953196 -0.717784 -0.528991
two 1  0.692802 -0.166100  1.403452

In [250]: df.loc['bar'] = df.loc['bar'] * 2

In [251]: df
Out[251]:
                  A         B         C
bar one 0       NaN       NaN       NaN
        5       NaN       NaN       NaN
    two 1       NaN       NaN       NaN
baz one 2 -0.065485  0.642380 -1.154436
qux one 3  1.515987  1.138678 -0.690564
    two 4  0.261254 -0.066171 -1.352841

This however works:

In [270]: df
Out[270]:
                  A         B         C
bar one 0  1.506060 -2.921602  1.057659
        5 -1.264088  6.483417  3.259962
    two 1  0.580368 -2.877875  2.916771
baz one 2 -0.003460 -1.779877  0.197661
qux one 3  0.463015  0.193559 -0.705382
    two 4  0.074625 -0.258451  0.389310

In [271]: idx = [x for x in df.index if x[0] == 'bar']

In [272]: df.loc[idx] = df.loc[idx] * 2

In [273]: df
Out[273]:
                  A          B         C
bar one 0  3.012121  -5.843204  2.115318
        5 -2.528176  12.966833  6.519925
    two 1  1.160736  -5.755750  5.833541
baz one 2 -0.003460  -1.779877  0.197661
qux one 3  0.463015   0.193559 -0.705382
    two 4  0.074625  -0.258451  0.389310

I was also wondering: if I want to assign to a selection via MultiIndex, what is the best way to achieve that? Examples:

df.xs('blah', level=2) = df.xs('blah', level=2) * 2
crit = [x for x in df.index where x[2] == 'blah'] 
df.ix[crit] = df.ix[crit] * 2
df.loc[crit] = df.loc[crit] * 2
df.reset_index(level=2); df[df.level_2 == 'blah'] = df[df.level_2 == 'blah'] * 2

[cpcloud]

it looks like the first example you give is a bug. as for your last question

• df.xs('blah', level=2) = df.xs('blah', level=2) * 2 is not syntactically valid python.
• all of the other methods look valid except that in place assignment might be slightly more efficient here.
• ix tries to be smart about integer vs labelled indexing
• loc only allows labelled indexing
• iloc only allows integer indexing

here is the documentation for the indexers

[jreback]

@floux 1st case is a bug (in fact we really just started supporting arbitrary rhs in 0.11), marking will get looked at for 0.12