Julia, like most technical computing languages, provides a first-class array implementation with very important functions that make working with N-Dimensional arrays quite easy. One of the latest features, the dot(.) based broadcasting makes most of the “repetitive” array operations, a one-lined code.
Array Indexing
Indexing into arrays in Julia is similar to it’s counterparts.
Syntax:
x = arr[index_1, index_2, …, index_n]
where each index_k may be a scalar integer, an array of integers, or an object that represents an array of scalar indices.
Array Indexing in Julia is of two types:
- Cartesian Indexing
- Logical Indexing
Cartesian Indexing
Cartesian Coordinates give the location of a point in 1D, 2D or 3D plane. Cartesian Indices have similar behavior. They give the value of element stored in a 1D, 2D, 3D or n-D array.
For scalar indices, CartesianIndex{N}s, behave like an N-tuple of integers spanning multiple dimensions while array of scalar indices include Arrays of CartesianIndex{N}
Syntax:
CartesianIndex(i, j, k...) -> I
CartesianIndex((i, j, k...)) -> I
The above syntax creates a multidimensional index I, which can be used for indexing a multidimensional array arr. We can say that, arr[I] is equivalent to arr[i, j, k…]. It is allowed to mix integer and CartesianIndex indices.
Example :
arr[Ipre, i, Ipost]
(where Ipre and Ipost are CartesianIndex indices and i is an Int)
can be a useful expression when writing algorithms that work along a single dimension
of an array of arbitrary dimensionality.
Scalar CartesianIndexing
CartesianIndexing for 1D arrays:
arr = reshape(Vector( 1 : 2 : 10 ), ( 5 ))
arr[CartesianIndex( 1 )]
arr[CartesianIndex( 3 )]
|
Output:
CartesianIndexing for a 2D array:
arr = reshape(Vector( 1 : 2 : 12 ), ( 3 , 2 ))
arr[CartesianIndex( 2 , 2 )]
arr[CartesianIndex( 3 , 1 )]
|
Output:
CartesianIndexing for 3D Array:
arr = reshape(Vector( 1 : 2 : 8 ), ( 2 , 1 , 2 ))
arr[CartesianIndex( 1 , 1 , 1 )]
arr[CartesianIndex( 1 , 1 , 2 )]
|
Output:
CartesianIndexing for nD array:
arr = reshape(Vector( 1 : 2 : 32 ), ( 2 , 2 , 1 , 2 , 2 ))
arr[CartesianIndex( 2 , 2 , 1 , 2 , 2 )]
|
Output:
From the above examples, it becomes pretty clear that CartesianIndex simply gathers multiple integers together into one object that represents a single multidimensional index.
Having discussed scalar representations, let’s talk about arrays of CartesianIndex{N}, which represent a collection of scalar indices each spanning N dimensions, such form of indexing is referred to as “pointwise” indexing.
Array based CartesianIndexing
Array based CartesianIndexing for 1D Array:
arr = reshape(Vector( 2 : 6 ), 5 )
arr[[CartesianIndex( 2 ),
CartesianIndex( 3 ), CartesianIndex( 5 )]]
|
Output:
Array based CartesianIndexing for 2D Array:
arr = reshape(Vector( 1 : 30 ), ( 5 , 6 ))
arr[[CartesianIndex( 5 , 2 ),
CartesianIndex( 3 , 6 ), CartesianIndex( 1 , 4 )]]
|
Output:
Array based CartesianIndexing for 3D array:
arr = reshape(Vector( 1 : 12 ), ( 2 , 3 , 2 ))
arr[[CartesianIndex( 1 , 2 , 1 ),
CartesianIndex( 1 , 3 , 1 ), CartesianIndex( 2 , 2 , 2 )]]
|
Output:
Array based CartesianIndexing for nD array:
arr = reshape(Vector( 1 : 2 : 32 ), ( 2 , 2 , 1 , 2 , 2 ))
arr[[CartesianIndex( 1 , 2 , 1 , 1 , 1 ),
CartesianIndex( 1 , 2 , 1 , 2 , 2 ),
CartesianIndex( 2 , 1 , 1 , 2 , 1 )]]
|
Output:
This is really useful, in case we want to select specific elements of an array if they are not in order. But what if we have an array of 1000 and we need to extract 100 elements, it’ll be quite tedious!
This is where dot(.) broadcasting comes into play.
arr = reshape(Vector( 1 : 27 ), ( 3 , 3 , 3 ))
arr[[CartesianIndex( 1 , 1 , 1 ), CartesianIndex( 2 , 2 , 1 ),
CartesianIndex( 3 , 3 , 1 ), CartesianIndex( 1 , 1 , 2 ),
CartesianIndex( 2 , 2 , 2 ), CartesianIndex( 3 , 3 , 2 ),
CartesianIndex( 1 , 1 , 3 ), CartesianIndex( 2 , 2 , 3 ),
CartesianIndex( 3 , 3 , 3 )]]
arr[CartesianIndex.(axes(arr, 1 ), axes(arr, 2 ), :)]
|
Output:
How efficiently we have been able to reduce our code in the third step!!
What should we do with Cartesian Indexing ?
We observe, CartesianIndices bear close resemblance to Cartesian coordinates and function in a similar way.
Example:
We have a 3D shape in space, and it is mapped out by storing the coordinates of it's
edges in an array and we might want to mask out a region from the shape. It becomes
easy to deal with by using CartesianIndices, since they resemble the coordinate system.
Logical Indexing
In computing/electronics, the basis is a logic that is deterministic in nature. It is:
true, false
one, zero
on, off
It is basically a selection of elements at the indices where the values of our logical indexing array are true.
A logical indexing array is called a “mask” since it masks out the values that are false. A mask is of type bool(boolean).
Relation with CartesianIndexing:
Indexing by a N-dimensional boolean array is equivalent to indexing by the vector
of CartesianIndex{N}s where its values are true.
Example : Implementation of a logical mask
arr = reshape(Vector( 1 : 25 ), ( 5 , 5 ))
arr[[true, false, true, false, true], :]
|
Output:
So we see how only the rows whose index matches with the index of trues in our mask[true, false, true, false, true] are selected.
Any condition that yields a boolean value can be used as a mask.
Example for 1D array:
arr = reshape(Vector( 4 : 2 : 22 ), 10 )
mask = map (ispow2, arr)
arr[mask]
|
Output:
Example for 2D array:
arr = reshape(Vector( 1 : 100 ), ( 10 , 10 ))
using Primes
mask = map (isprime, arr)
arr[mask]
|
Output:
Example for 3D array:
arr = reshape(Vector( 2 : 19 ), ( 3 , 2 , 3 ))
using Primes
mask = map (isprime, arr)
arr[mask]
|
Output:
One of the most used cases of logical indexing is when we want to filter elements out of an array on the basis of multiple conditions.
Example:
# Create a 3D array of size 3x2x3
arr = reshape(Vector(2:19), (3, 2, 3))
# Create a function that checks for
# both prime number or power of two
# Julia library for working with prime numbers
using Primes
function prime_and_2pow(arr)
if isprime(arr) || ispow2(arr)
return true
else
return false
end
end
# Create a logic mask and map it
# with the size of arr
mask = map(prime_and_2pow, arr)
# Apply it to the array
arr[mask]
|
Output:
Last Updated :
01 Dec, 2022
Like Article
Save Article
Share your thoughts in the comments
Please Login to comment...