Indexing, Masks, Views and Copies

Slices, boolean masks, fancy indexing

Slicing (a[1:4], a[::3]) returns a view that shares memory with the original, so writing to it changes the source. Boolean masks (a[a > 50]) and fancy indexing with lists of positions (a[[0, 2, 4]]) return copies. Multi-dimensional arrays take one index per axis (m[1, 2], m[:, 0]). Use .copy() explicitly when you need independent data.

Selecting and the view versus copy rule, run

I ran this with Python 3.12.3 and NumPy 2.5.3. Writing -1 through a slice changes the original array; writing 999 into the result of fancy indexing does not.

import numpy as np

a = np.arange(10, 100, 10)
print(a)
print(a[2], a[-1], a[1:4], a[::3])
mask = a > 50
print(mask)
print(a[mask])
print(a[[0, 2, 4]])                        # fancy indexing

m = np.arange(12).reshape(3, 4)
print(m[1, 2], m[:, 0], m[m % 2 == 0])

view = a[1:4]
view[0] = -1                               # slices are views
print(a[:4])
copy = a[[1, 2, 3]]
copy[0] = 999                              # fancy indexing returns a copy
print(a[:4])

Output:

[10 20 30 40 50 60 70 80 90]
30 90 [20 30 40] [10 40 70]
[False False False False False  True  True  True  True]
[60 70 80 90]
[10 30 50]
6 [0 4 8] [ 0  2  4  6  8 10]
[10 -1 30 40]
[10 -1 30 40]

Masks replace if statements

Filtering with a boolean mask is the vectorised version of a loop with an if statement, and is much faster.

त्वरित जाँच: Which operation returns a view sharing memory with the original?

• np.copy(a)
• A boolean mask
• Fancy indexing with a list
• A basic slice such as a[1:4]