Fancy index

Fancy indexing is a NumPy term, which refers to the use of integer arrays for indexing. Suppose we have an 8 × 4 array:

In [117]: arr = np.empty((8, 4))

In [118]: for i in range(8):
   .....:     arr[i] = i

In [119]: arr
Out[119]: 
array([[ 0.,  0.,  0.,  0.],
       [ 1.,  1.,  1.,  1.],
       [ 2.,  2.,  2.,  2.],
       [ 3.,  3.,  3.,  3.],
       [ 4.,  4.,  4.,  4.],
       [ 5.,  5.,  5.,  5.],
       [ 6.,  6.,  6.,  6.],
       [ 7.,  7.,  7.,  7.]])

To select a subset of rows in a specific order, simply pass in a list of integers or ndarray s for the specified order:

In [120]: arr[[4, 3, 0, 6]]
Out[120]: 
array([[ 4.,  4.,  4.,  4.],
       [ 3.,  3.,  3.,  3.],
       [ 0.,  0.,  0.,  0.],
       [ 6.,  6.,  6.,  6.]])

This code really meets our requirements! Using a negative index will pick rows from the end:

In [121]: arr[[-3, -5, -7]]
Out[121]: 
array([[ 5.,  5.,  5.,  5.],
       [ 3.,  3.,  3.,  3.],
       [ 1.,  1.,  1.,  1.]])

Passing in more than one index array at a time is a little special. It returns a one-dimensional array in which the elements correspond to the index tuples:

In [122]: arr = np.arange(32).reshape((8, 4))

In [123]: arr
Out[123]: 
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11],
       [12, 13, 14, 15],
       [16, 17, 18, 19],
       [20, 21, 22, 23],
       [24, 25, 26, 27],
       [28, 29, 30, 31]])

In [124]: arr[[1, 5, 7, 2], [0, 3, 1, 2]]
Out[124]: array([ 4, 23, 29, 10])

The reshape method is described in detail in Appendix A.

Finally, elements (1,0), (5,3), (7,1) and (2,2) were selected. No matter how many dimensions an array is, a fancy index is always one-dimensional.

The behavior of this fancy index may be different from what some users expect (including me). The row column subset of the selection matrix should be in the form of a rectangular region. Here is a way to get this result:

In [125]: arr[[1, 5, 7, 2]][:, [0, 3, 1, 2]]
Out[125]: 
array([[ 4,  7,  5,  6],
       [20, 23, 21, 22],
       [28, 31, 29, 30],
       [ 8, 11,  9, 10]])

Remember, unlike slicing, a fancy index always copies data into a new array.