Crate tensorism

Types for handling arrays with multiple indexes.

Provides types for arrays with multiple indexes (up to 8 indexes) that are:

• convenient to manipulate thanks to associated macros;
• safe to use as dimensions are encoded in the type system;
• minimising the number of runtime checks about validity of indexes.

Tensor types and dimensions

Tensor types (Tensor0, Tensor1, Tensor2, Tensor3, Tensor4, Tensor5, Tensor6, Tensor7 and Tensor8) are generic types not only over the type of data they contains but also over their multiple dimensions. This means dimensions are encoded in the type system.

// A tensor of usizes of rank 3 with static dimensions 3, 2 and 5.
let tensor_a: Tensor3<StaticDimTag<3>, StaticDimTag<2>, StaticDimTag<5>, usize> =
  TensorBuilding::with_static::<3>()
    .with_static::<2>()
    .with_static::<5>()
    .define(|(i, j, k)| (3 * i + 2 * j + 5 * k) % 17);
// A tensor of strings of rank 1 with static dimensions 130.
let tensor_b: Tensor1<StaticDimTag<130>, String> =
  TensorBuilding::with_static::<130>()
    .prepare()
    .append_vec(&mut string_vector)
    .generate();

This is not only true for compile-time dimensions. You can also create tensors some dimensions of which are not known until runtime and still telling the compiler where these dimensions are equal, independently of their actual numerical value.

let n: usize = …;
let m: usize = …;
// A tensor of floats of rank 4 with dimensions n, 3, m and n.
// Its type is similar to `Tensor4<TagN, StaticDimTag<3>, TagM, TagN, f64>`.
let tensor_c: Tensor4<_, StaticDimTag<3>, _, _, f64> =
  TensorBuilding::with(new_dynamic_dim!(n))
    .with_static::<3>()
    .with(new_dynamic_dim!(m))
    .with_first()
    .define(|(i, j, k, l)| … as f64);

// A tensor of booleans of rank 2 with dimensions m and n.
// Its type is similar to `Tensor4<TagM, TagN, bool>`.
let tensor_d: Tensor2<_, _, bool> =
  TensorBuilding::with(tensor_c.shape().2)
    .with(tensor_c.shape().0)
    .define(|(i, j)| … as bool);

Manipulating tensors

Sibling crate tensorism-gen contains procedural macros that greatly simplify the effective manipulation of tensors via a domain specific language. Please refer to this crate for additional documentation.

Modules

building

dimensions

Abstractions for dimensions used in a tensor.

tensors

The mathematical word tensor is used here to describe more broadly arrays with multiple indexes. Elements can be equated and accessed throught tuple of usize.

Macros

from_vec_to_tensor1

Transform a Vec into a Tensor1.

new_dynamic_dim

Generate a new dimension at runtime.