Indexing and Slicing

Fancy Index

Four elements gathered from a 6-element list at non-contiguous positions [0, 2, 4, 5]. The replay shows p taking each index in turn and out growing with the selected value.

By hand

With NumPy

a[idx] accepts a list (or array) of integer positions and returns a new array of the gathered elements in one call — no explicit loop needed. The snapshot shows the result's shape, dtype, and values.

naive.py 
values = [10, 20, 30, 40, 50, 60]
idx = [0, 2, 4, 5]
out = []
for p in idx:
    out.append(values[p])
print('RESULT:', out)
library.py 
import numpy as np

values = [10, 20, 30, 40, 50, 60]
idx = [0, 2, 4, 5]
a = np.array(values)
out = a[idx]
print('shape:', out.shape)
print('dtype:', out.dtype)
print('values:', out.tolist())
print('RESULT:', out.tolist())

shape: (4,)
dtype: int64
values: [10, 30, 50, 60]
RESULT: [10, 30, 50, 60]

Implementation notes

  • Fancy indexing always returns a copy, not a view. Modifying the output array does not affect the original — unlike basic slicing, which returns a view.
  • The index list may contain duplicates or out-of-order positions: a[[2, 0, 2]] returns [30, 10, 30]. This flexibility is the key difference from slice syntax, which can only express contiguous or strided ranges.
  • Shape, dtype, and values are shown explicitly here because ndarray.__repr__ output varies with NumPy version and print options.