Accessing Tensor Elements in F# QuickStart Sample
Illustrates different ways of accessing elements of a tensor and sub-tensors using classes in the Numerics.NET.Tensors namespace in F#.
View this sample in: C# Visual Basic
// Illustrates different ways of getting and setting
// elements of a tensor.
open System
// Tensor classes reside in the Numerics.NET.Tensors
// namespace.
open Numerics.NET.Tensors
open Numerics.NET.FSharp
// The license is verified at runtime. We're using
// a 30 day trial key here. For more information, see
// https://numerics.net/trial-key
let licensed = Numerics.NET.License.Verify("64542-18980-57619-62268")
//
// Accessing tensor elements
//
// Let's create a few tensors to work with:
let t = Tensor.CreateFromFunction(new TensorShape(3, 4), fun i j -> 11 + 10 * i + j)
// t -> [ [ 11, 12, 13, 14 ],
// [ 21, 22, 23, 24 ],
// [ 31, 32, 33, 34 ] ]
// Actually, let's use something a little bigger:
let t = Tensor.CreateFromFunction(new TensorShape(3, 4, 5), fun i j k -> 100 * i + 10 * j + k)
// Tensors have indexer properties which get or set all or part
// of a tensor, including individual values.
// Important: All indexers return Tensor<T> objects,
// even if it contains just a single element!
let t123 = t[1, 2, 3]
// t123 -> [ 123 ]
Console.WriteLine($"Type of t[1, 2, 3] -> {t123.GetType()}")
#if PREVIEW
// As of C# 8.0, you can use the caret (^) operator to
// specify that the index should be counted from the end
// of the dimension:
let t_2_2_2 = t[^2, ^2, ^2]
// t_2_2_2 -> [ 123 ]
#endif
// Single values can be set using the indexer, but you
// have to assign a scalar tensor:
t[1, 2, 3] <- Tensor.CreateScalar(999)
let t123 = t[1, 2, 3]
// t123 -> [ 999 ]
// To get an element's value, and not a scalar tensor,
// use the GetValue method:
let tValue = t.GetValue(1, 2, 3)
// tValue -> [ 999 ]
// A corresponding SetValue method lets you set the value:
t.SetValue(99, 1, 2, 3)
let tValue = t.GetValue(1, 2, 3)
// tValue -> [ 99 ]
// When you leave out dimensions, the entire dimensions
// are returned:
let t12x = t[1, 2]
// t12x -> [ 120, 121, 122, 999, 124 ]
// You can use ranges and slices to get or set sub-tensors.
// You can use either Numerics.NET.Range or System.Range:
let r12 = new Numerics.NET.Range(1, 2)
let trrr = t[r12, r12, r12]
// trrr -> [[[ 111, 112], [121, 122]], [211, 212], [221, 222]]]
let trrr = t[1..3, 1..3, 1..3]
// You can mix and match:
let s = Tensor.CreateFromFunction(new TensorShape(3, 3), fun i j -> 11 + 10 * i + j)
// s -> [[ 11 12 13 ]
// [ 21 22 23 ]
// [ 31 32 33 ]]
let row1 = s[0, *]
// row1 -> [ 11 12 13 ]
let column1 = s[*, ^2]
// column1 -> [ 12 22 32 ]
let row2 = s[1, 1..]
// row2 -> [ 22 23 ]
// C#'s ranges do not support strides. For that, you have to use
// either Numerics.NET.Range or Numerics.NET.Slice:
let row3 = s[1, new Numerics.NET.Range(0, 2, 2)]
// row3 -> [ 21 23 ]
row3 = s[1, new Numerics.NET.Slice(2, 0, 2)]
// row3 -> [ 21 23 ]
// You can even have ranges with negative strides:
let x = Tensor.CreateRange(3)
// x -> [ 0 1 2 ]
let reverse = x[new Numerics.NET.Range(2, 0, -1)]
// reverse -> [ 2 1 0 ]
let reverse = x[new Numerics.NET.Slice(2, 2, -1)]
// reverse -> [ 2 1 0 ]
// You can set values using ranges and slices:
s[1, 1..3] <- Tensor.CreateFromArray(new[] { 88, 99 })
// s -> [[ 11 12 13 ]
// [ 21 88 99 ]
// [ 31 32 33 ]]
s[..^1, ^2] <- Tensor.CreateFromArray(new[] { 1, 2 })
// s -> [[ 11 1 13 ]
// [ 21 2 99 ]
// [ 31 32 33 ]]
// TODO: s[1, new Range(0, 2, 2)] = Tensor.CreateFromArray(new[] { 77, 66 })
// s -> [[ 11 1 13 ]
// [ 77 2 66 ]
// [ 31 32 33 ]]
//
// Advanced indexes:
//
// You can use sets of integers to specify only those elements:
int[] indexes = { 0, 3 }
let t1 = t[1, indexes, 3..5]
// t1 -> [[ 103 133 ]
// [ 104 134 ]]
// You can also use a mask, an array of booleans, that are true
// for the elements you want to select:
bool[] mask = { true, false, false, true }
let t2 = t[1, mask, 3..5]
// t2 -> [[ 103 133 ]
// [ 104 134 ]]
//
// Copying and cloning tensors
//
// A shallow copy of a tensor constructs a tensor
// that shares the component storage with the original.
// This is done using an indexer:
Console.WriteLine("Shallow copy vs. clone:")
let t10 = t2[TensorIndex.All]
// The Copy method creates a full copy.
let t11 = t2.Copy()
// When we change t2, t10 changes, but t11 is left
// unchanged:
Console.WriteLine($"t2[1,1] = {t2[1, 1]}")
t2.SetValue(-2, 1, 1)
Console.WriteLine($"t10[1,1] = {t10[1, 1]}")
Console.WriteLine($"t11[1,1] = {t11[1, 1]}")
#if CLONEDATA
// We can give a tensor its own component storage
// by calling the CloneData method:
Console.WriteLine("CloneData:")
// TODO: t11.CloneData()
// Now, changing the original v2 no longer changes v7:
// t2[1, 1] = 4
Console.WriteLine($"t11[1,1] = {t11[1, 1]}")
#endif
Console.Write("Press Enter key to exit...")
Console.ReadLine() |> ignore