NumPy: Slicing ndarray
In Python, you can use slice [start:stop:step] to select a part of a sequence object such as a list, string, or tuple to get a value or assign another value.
It is also possible to select a subarray by slicing for the NumPy array numpy.ndarray and extract a value or assign another value.
This article describes the following:
- Basics of slicing
- Slice for one-dimensional
numpy.ndarray - Slice for multi-dimensional
numpy.ndarray - Views and copies
- With fancy indexing
See the following article for a detailed explanation of how to select elements and subarrays of numpy.ndarray.
See the following article for how to extract rows and columns that satisfy the conditions.
Basics of slicing
In Python, you can use slice [start:stop:step] to select a part of a sequence object such as a list, string, or tuple to get a value or assign another value.
import numpy as np
l = list(range(10))
print(l)
# [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
print(l[4:8])
# [4, 5, 6, 7]
print(l[-5:-2])
# [5, 6, 7]
print(l[::-1])
# [9, 8, 7, 6, 5, 4, 3, 2, 1, 0]
Slicing is a standard feature in Python. See the following article for details.
Slice for one-dimensional numpy.ndarray
Get a subarray
The slice for the one-dimensional numpy.ndarray is the same as the basic Python slice as in the example above.
a = np.arange(10)
print(a)
# [0 1 2 3 4 5 6 7 8 9]
print(a[4:8])
# [4 5 6 7]
print(a[-5:-2])
# [5 6 7]
print(a[::-1])
# [9 8 7 6 5 4 3 2 1 0]
Assign another value
The behavior of assignment using slices is different for list and for numpy.ndarray.
See the following article for assignment to list by slice. The number of elements selected by slice and the number of elements to be assigned do not have to match.
In numpy.ndarray, the value on the right side is converted by broadcasting and assigned.
If the right side is a scalar value, all elements selected by the slice will be replaced with that scalar value. If the right side is a one-dimensional array, it will be replaced with that array.
a[3:6] = 100
print(a)
# [ 0 1 2 100 100 100 6 7 8 9]
a[3:6] = [100, 200, 300]
print(a)
# [ 0 1 2 100 200 300 6 7 8 9]
When assigning an array, if the number of elements in the array to be assigned and the number of elements selected in the slice do not match, ValueError is raised.
# a[3:6] = [100, 200, 300, 400]
# ValueError: cannot copy sequence with size 4 to array axis with dimension 3
The same applies to slices for which step is specified.
a = np.arange(10)
print(a)
# [0 1 2 3 4 5 6 7 8 9]
print(a[2:8:2])
# [2 4 6]
a[2:8:2] = 100
print(a)
# [ 0 1 100 3 100 5 100 7 8 9]
a[2:8:2] = [100, 200, 300]
print(a)
# [ 0 1 100 3 200 5 300 7 8 9]
Slice for multi-dimensional numpy.ndarray
For multi-dimensional numpy.ndarray, slices of each dimension can be specified separated by commas.
Take the following two-dimensional array as an example.
a = np.arange(12).reshape((3, 4))
print(a)
# [[ 0 1 2 3]
# [ 4 5 6 7]
# [ 8 9 10 11]]
Get a subarray
Specify slices for each dimension separated by commas.
print(a[1:, 1:3])
# [[ 5 6]
# [ 9 10]]
Select rows
You can select rows using :. In this case, the trailing : can be omitted.
print(a[1:, :])
# [[ 4 5 6 7]
# [ 8 9 10 11]]
print(a[1:])
# [[ 4 5 6 7]
# [ 8 9 10 11]]
When selecting one row, if you specify the index by the scalar value instead of the slice, it will be selected as a one-dimensional array, but if you select one row by the slice, it will be selected as a two-dimensional array.
print(a[1])
# [4 5 6 7]
print(a[1].shape)
# (4,)
print(a[1:2])
# [[4 5 6 7]]
print(a[1:2].shape)
# (1, 4)
Be careful in cases that shape is important, such as matrix operations.
Select columns
The same applies to column selection. In this case, the first : cannot be omitted.
print(a[:, 1:3])
# [[ 1 2]
# [ 5 6]
# [ 9 10]]
print(a[:, 1])
# [1 5 9]
print(a[:, 1].shape)
# (3,)
print(a[:, 1:2])
# [[1]
# [5]
# [9]]
print(a[:, 1:2].shape)
# (3, 1)
You can use ... if : repeats. See the following article.
Assign another value
As in the case of one-dimensional numpy.ndarray, in the assignment to the multi-dimensional numpy.ndarray, the value on the right side is converted by broadcasting and assigned.
When assigning an array, ValueError is raised if the number of elements of the array to be assigned and the number of corresponding elements selected by the slice do not match.
a = np.arange(12).reshape((3, 4))
print(a)
# [[ 0 1 2 3]
# [ 4 5 6 7]
# [ 8 9 10 11]]
print(a[1:, 1:3])
# [[ 5 6]
# [ 9 10]]
a[1:, 1:3] = 100
print(a)
# [[ 0 1 2 3]
# [ 4 100 100 7]
# [ 8 100 100 11]]
a[1:, 1:3] = [100, 200]
print(a)
# [[ 0 1 2 3]
# [ 4 100 200 7]
# [ 8 100 200 11]]
a[1:, 1:3] = [[100, 200], [300, 400]]
print(a)
# [[ 0 1 2 3]
# [ 4 100 200 7]
# [ 8 300 400 11]]
The same applies to slices for which step is specified.
a = np.arange(12).reshape((3, 4))
print(a)
# [[ 0 1 2 3]
# [ 4 5 6 7]
# [ 8 9 10 11]]
print(a[1:, ::2])
# [[ 4 6]
# [ 8 10]]
a[1:, ::2] = 100
print(a)
# [[ 0 1 2 3]
# [100 5 100 7]
# [100 9 100 11]]
a[1:, ::2] = [100, 200]
print(a)
# [[ 0 1 2 3]
# [100 5 200 7]
# [100 9 200 11]]
a[1:, ::2] = [[100, 200], [300, 400]]
print(a)
# [[ 0 1 2 3]
# [100 5 200 7]
# [300 9 400 11]]
Views and copies
The subarray extracted by slicing is a view of the original array, and changing the elements of the sub-array also changes the elements of the original array.
a = np.arange(12).reshape((3, 4))
print(a)
# [[ 0 1 2 3]
# [ 4 5 6 7]
# [ 8 9 10 11]]
a_slice = a[1:, 1:3]
print(a_slice)
# [[ 5 6]
# [ 9 10]]
a_slice[0, 0] = 100
print(a_slice)
# [[100 6]
# [ 9 10]]
print(a)
# [[ 0 1 2 3]
# [ 4 100 6 7]
# [ 8 9 10 11]]
You can make a copy of a subarray using copy(). Changing the elements of the copy does not change the elements of the original array.
a = np.arange(12).reshape((3, 4))
print(a)
# [[ 0 1 2 3]
# [ 4 5 6 7]
# [ 8 9 10 11]]
a_slice_copy = a[1:, 1:3].copy()
print(a_slice_copy)
# [[ 5 6]
# [ 9 10]]
a_slice_copy[0, 0] = 100
print(a_slice_copy)
# [[100 6]
# [ 9 10]]
print(a)
# [[ 0 1 2 3]
# [ 4 5 6 7]
# [ 8 9 10 11]]
See the following article for more information on views and copies in numpy.ndarray.
With fancy indexing
In NumPy, you can select a subarray from numpy.ndarray by a list of indexes.
Subarrays can be selected by combining fancy indexing and slicing.
a = np.arange(12).reshape((3, 4))
print(a)
# [[ 0 1 2 3]
# [ 4 5 6 7]
# [ 8 9 10 11]]
print(a[[0, 2], 1:3])
# [[ 1 2]
# [ 9 10]]
a[[0, 2], 1:3] = 100
print(a)
# [[ 0 100 100 3]
# [ 4 5 6 7]
# [ 8 100 100 11]]
a[[0, 2], 1:3] = [100, 200]
print(a)
# [[ 0 100 200 3]
# [ 4 5 6 7]
# [ 8 100 200 11]]
a[[0, 2], 1:3] = [[100, 200], [300, 400]]
print(a)
# [[ 0 100 200 3]
# [ 4 5 6 7]
# [ 8 300 400 11]]
Note that the subarray extracted by fancy indexing is a copy, not a view.
a_subset = a[[0, 2], 1:3]
print(a_subset)
# [[100 200]
# [300 400]]
a_subset[0, 0] = -1
print(a_subset)
# [[ -1 200]
# [300 400]]
print(a)
# [[ 0 100 200 3]
# [ 4 5 6 7]
# [ 8 300 400 11]]