NumPy: Ellipsis (...) for ndarray | note.nkmk.me (original) (raw)
In NumPy, you can use the Python built-in constant Ellipsis (represented as three dots, ...) within [] to omit specifying intermediate dimensions of an array (ndarray).
For information on how to set whether to omit elements when outputting ndarray with print(), refer to the following article.
Python built-in constant Ellipsis (three dots, ...)
In Python, Ellipsis is defined as a built-in constant.
`print(Ellipsis)
Ellipsis
`
Ellipsis can also be represented as ..., consisting of three dots.
Ellipsis and ... are the same ellipsis object.
`print(type(Ellipsis))
<class 'ellipsis'>
print(type(...))
<class 'ellipsis'>
print(Ellipsis is ...)
True
`
As of Python 3.12, Ellipsis (...) has no specific role in the basic syntax, but NumPy offers a practical application, as demonstrated below.
Ellipsis (...) in NumPy
In NumPy, Ellipsis (...) allows you to omit intermediate dimensions when specifying a range of an array (ndarray) with [].
Consider the following four-dimensional array as an example.
`import numpy as np
print(np.version)
1.26.1
a = np.arange(120).reshape(2, 3, 4, 5)
print(a.shape)
(2, 3, 4, 5)
`
For example, to specify only the last dimension, you can write it using : like below.
`print(a[:, :, :, 0])
[[[ 0 5 10 15]
[ 20 25 30 35]
[ 40 45 50 55]]
[[ 60 65 70 75]
[ 80 85 90 95]
[100 105 110 115]]]
`
Using ..., it can be written as follows.
`print(a[..., 0])
[[[ 0 5 10 15]
[ 20 25 30 35]
[ 40 45 50 55]]
[[ 60 65 70 75]
[ 80 85 90 95]
[100 105 110 115]]]
`
The same applies when specifying only the first and last dimensions. Intermediate dimensions can be omitted with ....
`print(a[0, :, :, 0])
[[ 0 5 10 15]
[20 25 30 35]
[40 45 50 55]]
print(a[0, ..., 0])
[[ 0 5 10 15]
[20 25 30 35]
[40 45 50 55]]
`
You can also use Ellipsis instead of ....
`print(a[Ellipsis, 0])
[[[ 0 5 10 15]
[ 20 25 30 35]
[ 40 45 50 55]]
[[ 60 65 70 75]
[ 80 85 90 95]
[100 105 110 115]]]
print(a[0, Ellipsis, 0])
[[ 0 5 10 15]
[20 25 30 35]
[40 45 50 55]]
`
In a four-dimensional example, the difference between using : and ... is minimal. However, ... becomes more convenient for higher-dimensional arrays. While : requires matching the number of dimensions, ... does not have this requirement.
Using two or more ... creates ambiguity about where to omit and results in an error.
`# print(a[..., 0, ...])
IndexError: an index can only have a single ellipsis ('...')
`
If : is used for each dimension leading up to the last one, it can be entirely omitted. In such cases, there is no necessity to use ..., though including it does not result in an error.
`print(a[0, 0, :, :])
[[ 0 1 2 3 4]
[ 5 6 7 8 9]
[10 11 12 13 14]
[15 16 17 18 19]]
print(a[0, 0])
[[ 0 1 2 3 4]
[ 5 6 7 8 9]
[10 11 12 13 14]
[15 16 17 18 19]]
print(a[0, 0, ...])
[[ 0 1 2 3 4]
[ 5 6 7 8 9]
[10 11 12 13 14]
[15 16 17 18 19]]
`