python - Pandas groupby with dict

Question

Is it possible to use a dict to group on elements of a column?

For example:

In [3]: df = pd.DataFrame({'A' : ['one', 'one', 'two', 'three','two', 'two', 'one', 'three'],
   ...:          'B' : np.random.randn(8)})
In [4]: df
Out[4]: 
       A         B
0    one  0.751612
1    one  0.333008
2    two  0.395667
3  three  1.636125
4    two  0.916435
5    two  1.076679
6    one -0.992324
7  three -0.593476

In [5]: d = {'one':'Start', 'two':'Start', 'three':'End'}
In [6]: grouped = df[['A','B']].groupby(d)

This (and other variations) returns an empty groupby object. And my variations on using .apply all fail too.

I'd like to match the values of column A to the keys of the dictionary and put rows into the groups defined by the values. The output would look something like this:

 Start:
           A         B
    0    one  0.751612
    1    one  0.333008
    2    two  0.395667
    4    two  0.916435
    5    two  1.076679
    6    one -0.992324
End:
           A         B
    3  three  1.636125
    7  three -0.593476

Answer 1

From the docs, the dict has to map from labels to group names, so this will work if you put 'A' into the index:

grouped2 = df.set_index('A').groupby(d)
for group_name, data in grouped2:
    print group_name
    print '---------'
    print data

# Output:
End
---------
              B
A              
three -1.234795
three  0.239209

Start
---------
            B
A            
one -1.924156
one  0.506046
two -1.681980
two  0.605248
two -0.861364
one  0.800431

Column names and row indices are both labels, whereas before you put 'A' into the index, the elements of 'A' are values.

If you have other info in the index that makes doing a set_index() tricky, you can just create a grouping column with map():

df['group'] = df['A'].map(d)
grouped3 = df.groupby('group')