Copy and View

A copy is a new array that owns its values. A view is another way of looking at the same values. Change the original, and a view changes with it. A copy does not.

copy owns its data

Changes to the original do not reach the copy, and changes to the copy do not reach the original.

import numpy as np

arr = np.array([1, 2, 3, 4, 5])
x = arr.copy()
arr[0] = 42
print(arr)
print(x)

Output:

[42  2  3  4  5]
[1 2 3 4 5]

view shares the data

A view reads the same memory. A change on either side shows on both.

import numpy as np

arr = np.array([1, 2, 3, 4, 5])
y = arr.view()
arr[0] = 42
print(y)

y[1] = 31
print(arr)

Output:

[42  2  3  4  5]
[42 31  3  4  5]

Check who owns the data

base is None when an array owns its data. For a view, base is the array it came from. A basic slice is a view too. np.shares_memory(a, b) answers the question directly.

import numpy as np

arr = np.array([1, 2, 3, 4, 5])
x = arr.copy()
y = arr.view()
part = arr[1:3]

print(x.base)
print(y.base is arr)
print(part.base is arr)
print(np.shares_memory(arr, x), np.shares_memory(arr, part))

Output:

None
True
True
False True

flatten and ravel

Both turn a multi-dimensional array into 1-D, reading row by row. flatten always returns a copy. ravel returns a view when it can, which is faster and uses no extra memory. That also means a change to the raveled array can reach the original.

import numpy as np

arr = np.array([[1, 2, 3], [4, 5, 6]])

flat = arr.flatten()
flat[0] = 10
print(flat)
print(arr)

line = arr.ravel()
line[0] = 20
print(line)
print(arr)

Output:

[10  2  3  4  5  6]
[[1 2 3]
 [4 5 6]]
[20  2  3  4  5  6]
[[20  2  3]
 [ 4  5  6]]

Key takeaways

• A copy owns its data, so base is None

• A view and a basic slice share data with the original

• flatten copies. ravel returns a view when it can

• Use .copy() before editing a piece you want to keep separate

• Next: Sorting and Comparing

• Previous: Advanced Indexing