new row not created upon assignment to nonexistent row label in DataFrame with MultiIndex

[lebedov]

According to the documentation for pandas 0.13+, assigning to a non-existent key via the .ix or .loc operations should create a new row in the DataFrame. Using pandas 0.13.1 with Python 2.7.5, I noticed that this doesn't seem to be the case when the DataFrame has a MultiIndex. To illustrate, consider a DataFrame created as follows:

import itertools
import numpy as np
import pandas

rows = 2 
cols = 3 
a = np.random.randint(0, 2, rows*cols).reshape((rows, cols))
b = a*np.random.rand(rows, cols)
idx = pandas.MultiIndex.from_tuples([(i, j) for i, j in \
             itertools.product(xrange(rows), xrange(cols))],
                              names=['rows','cols'])
df = pandas.DataFrame({'a': a.flatten(),
                       'b': b.flatten()}, index=idx)

If df is

           a         b
rows cols             
0    0     0  0.000000
     1     0  0.000000
     2     1  0.315458
1    0     0  0.000000
     1     0  0.000000
     2     1  0.061142

then running df.loc[(0, 1), :] = (10, 20.0) modifies df as follows:

            a          b
rows cols               
0    0      0   0.000000
     1     10  20.000000
     2      1   0.315458
1    0      0   0.000000
     1      0   0.000000
     2      1   0.061142

However, if one attempts to run df.loc[(0,3), :] = (10, 20.0) on the original df, one obtains

            a          b
rows cols               
0    0     10  20.000000
     1      0   0.000000
     2      1   0.315458
1    0     10  20.000000
     1      0   0.000000
     2      1   0.061142

Is this behavior expected?

[jreback]

Its not implemented with a multi-index, should prob just raise an error for now.

want to do a PR to raise NotImplementedError?

[toobaz]

This seems to be fixed (at least, the proposed example behaves fine in git):

In [2]: df.loc[(0,3), :] = (10, 20.0)

In [3]: df
Out[3]: 
              a          b
rows cols                 
0    0      1.0   0.274632
     1      0.0   0.000000
     2      1.0   0.056103
1    0      1.0   0.941639
     1      1.0   0.546224
     2      1.0   0.866284
0    3     10.0  20.000000

[lebedov]

I concur - looks good in pandas 0.19.2. Closing.