Assign new values to slice from MultiIndex DataFrame

Question

I would like to modify some values from a column in my DataFrame. At the moment I have a view from select via the multi index of my original df (and modifying does change df).

Here's an example:

In [1]: arrays = [np.array(['bar', 'bar', 'baz', 'qux', 'qux', 'bar']),
                  np.array(['one', 'two', 'one', 'one', 'two', 'one']),
                  np.arange(0, 6, 1)]
In [2]: df = pd.DataFrame(randn(6, 3), index=arrays, columns=['A', 'B', 'C'])

In [3]: df
                  A         B         C
bar one 0 -0.088671  1.902021 -0.540959
    two 1  0.782919 -0.733581 -0.824522
baz one 2 -0.827128 -0.849712  0.072431
qux one 3 -0.328493  1.456945  0.587793
    two 4 -1.466625  0.720638  0.976438
bar one 5 -0.456558  1.163404  0.464295

I try to modify a slice of df to a scalar value:

In [4]: df.ix['bar', 'two', :]['A']
Out[4]:
1    0.782919
Name: A, dtype: float64

In [5]: df.ix['bar', 'two', :]['A'] = 9999
# df is unchanged

I really want to modify several values in the column (and since indexing returns a vector, not a scalar value, I think this would make more sense):

In [6]: df.ix['bar', 'one', :]['A'] = [999, 888]
# again df remains unchanged

I'm using pandas 0.11. Is there is a simple way to do this?

The current solution is to recreate df from a new one and modify values I want to. But it's not elegant and can be very heavy on complex dataframe. In my opinion the problem should come from .ix and .loc not returning a view but a copy.

Answer 1

Sort the frame, then select/set using a tuple for the multi-index

In [12]: df = pd.DataFrame(randn(6, 3), index=arrays, columns=['A', 'B', 'C'])

In [13]: df
Out[13]: 
                  A         B         C
bar one 0 -0.694240  0.725163  0.131891
    two 1 -0.729186  0.244860  0.530870
baz one 2  0.757816  1.129989  0.893080
qux one 3 -2.275694  0.680023 -1.054816
    two 4  0.291889 -0.409024 -0.307302
bar one 5  1.697974 -1.828872 -1.004187

In [14]: df = df.sortlevel(0)

In [15]: df
Out[15]: 
                  A         B         C
bar one 0 -0.694240  0.725163  0.131891
        5  1.697974 -1.828872 -1.004187
    two 1 -0.729186  0.244860  0.530870
baz one 2  0.757816  1.129989  0.893080
qux one 3 -2.275694  0.680023 -1.054816
    two 4  0.291889 -0.409024 -0.307302

In [16]: df.loc[('bar','two'),'A'] = 9999

In [17]: df
Out[17]: 
                     A         B         C
bar one 0    -0.694240  0.725163  0.131891
        5     1.697974 -1.828872 -1.004187
    two 1  9999.000000  0.244860  0.530870
baz one 2     0.757816  1.129989  0.893080
qux one 3    -2.275694  0.680023 -1.054816
    two 4     0.291889 -0.409024 -0.307302

You can also do it with out sorting if you specify the complete index, e.g.

In [23]: df.loc[('bar','two',1),'A'] = 999

In [24]: df
Out[24]: 
                    A         B         C
bar one 0   -0.113216  0.878715 -0.183941
    two 1  999.000000 -1.405693  0.253388
baz one 2    0.441543  0.470768  1.155103
qux one 3   -0.008763  0.917800 -0.699279
    two 4    0.061586  0.537913  0.380175
bar one 5    0.857231  1.144246 -2.369694

To check the sort depth

In [27]: df.index.lexsort_depth
Out[27]: 0

In [28]: df.sortlevel(0).index.lexsort_depth
Out[28]: 3

The last part of your question, assigning with a list (note that you must have the same number of elements as you are trying to replace), and this MUST be sorted for this to work

In [12]: df.loc[('bar','one'),'A'] = [999,888]

In [13]: df
Out[13]: 
                    A         B         C
bar one 0  999.000000 -0.645641  0.369443
        5  888.000000 -0.990632 -0.577401
    two 1   -1.071410  2.308711  2.018476
baz one 2    1.211887  1.516925  0.064023
qux one 3   -0.862670 -0.770585 -0.843773
    two 4   -0.644855 -1.431962  0.232528