#tensor #type #complex-numbers #data-science

rapl

A crate that makes numerical scripting with Rust simple and enjoyable

5 unstable releases

0.3.0 Jun 13, 2023
0.2.1 Apr 22, 2023
0.2.0 Apr 21, 2023
0.1.1 Mar 22, 2023
0.1.0 Mar 22, 2023

#638 in Math

42 downloads per month

MIT license

5MB
3.5K SLoC

rapl

Documentation Crate

Note: rapl is in early development and is not optimized for performance, is not recommended for production applications.

rapl is computing Rust library that provides a simple way of working with N-dimensional array, along with a wide range of mathematical functions to manipulate them. It takes inspiration from NumPy and APL, with the primary aim of achieving maximum ergonomics and user-friendliness while maintaining generality.

Our goal is to make Rust scripting as productive as possible, and make Rust a real option when it comes to numerical computing and data science. Check out the examples.

Out of the box rapl provides features like co-broadcasting, rank type checking, native complex number support, among many others:

use rapl::*;
fn main() {
    let a = Ndarr::from([1 + 1.i(), 2 + 1.i()]);
    let b = Ndarr::from([[1, 2], [3, 4]]);
    let r = a + b - 2;
    assert_eq!(r, Ndarr::from([[1.i(), 2 + 1.i()],[2 + 1.i(), 4 + 1.i()]]));
}

Array initialization

There are multiple handy ways of initializing N-dimensional arrays (or Ndarr).

  • From Native Rust arrays to Ndarr.
let a = Ndarr::from(["a","b","c"]); 
let b = Ndarr::from([[1,2],[3,4]]);
  • From ranges.
let a = Ndarr::from(1..7).reshape(&[2,3])
  • From &str
let chars = Ndarr::from("Hello rapl!"); //Ndarr<char,U1>
  • Others:
let ones: Ndarr<f32, 2> = Ndarr::ones(&[4,4]);
let zeros : Ndarr<i32, 3>= Ndarr::zeros(&[2,3,4]);
let letter_a = Ndarr::fill("a", &[5]);
let fold = Ndarr::new(data: &[0, 1, 2, 3], shape: [2, 2]).expect("Error initializing");
  • linspace, logspace, geomspace
    let linear = Ndarr::linspace(0, 9, 10);
    assert_eq!(linear,Ndarr::from(0..10));

    let logarithmic = Ndarr::logspace(0.,9., 10., 10);
    assert!(logarithmic.approx(&Ndarr::from([1.,1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7, 1e8, 1e9])));

    let geom = Ndarr::geomspace(1.,256., 9);
    assert!(geom.approx(&Ndarr::from([1., 2., 4., 8., 16., 32., 64., 128., 256.])));

Random array creation

You can easily create random array of any shape:

//Normal distribution
let arr_norm = NdarrRand::normal(low: 0f32, high: 1f32, shape: [2, 2], Seed: Some(1234));
//Normal distribution
let arr_uniform = NdarrRand::uniform(low: 0f32, high: 1f32, shape: [10], Seed: None);
//Choose between values
let arr_choose = NdarrRand::choose(&[1, 2, 3, 4, 5], [3, 3], Some(1234));

Element wise operations

  • Arithmetic operation with with scalars
let ones: Ndarr<i32, 2> = Ndarr::ones(&[4,4]);
let twos = ones + 1;
let sixes = twos * 3;
  • Arithmetic operation between Ndarrs,
let a = Ndarr::from([[1,2],[3,4]]);
let b = Ndarr::from([[1,2],[-3,-4]]);

assert_eq!(a + b, Ndarr::from([[2,4],[0,0]]))

Note: If the shapes are not equal rapl will automatically broadcast the arrays into a compatible shape (if it exist) and perform the operation.

  • Math operations including trigonometric and activation functions.
let x = Ndarr::from([-1.0 , -0.8, -0.6, -0.4, -0.2, 0.0, 0.2, 0.4, 0.6, 0.8, 1.0]);
let sin_x = x.sin();
let cos_x = x.cos();
let tanh_x = x.tanh();

let abs_x = x.abs();
let relu_x = x.relu();
  • Map function
let a = Ndarr::from([[1,2],[3,4]]);
let mapped = a.map(|x| x*2-1);

Monadic tensor operations

  • Transpose
let arr = Ndarr::from([[1,2,3],[4,5,6]]);	
assert_eq!(arr.shape(), [2,3]);
assert_eq!(arr.clone().t().shape, [3,2]); //transpose
  • Reshape
let a = Ndarr::from(1..7).reshape(&[2,3]).unwrap();
  • Slice
let arr = Ndarr::from([[1,2],[3,4]]);

assert_eq!(arr.slice_at(1)[0], Ndarr::from([1,3]))
  • Reduce
let sum_axis = arr.clone().reduce(1, |x,y| x + y).unwrap();
assert_eq!(sum_axis, Ndarr::from([6, 15])); //sum reduction
  • Scan right an left
 let s = Ndarr::from([1,2,3]);
 let cumsum = s.scanr( 0, |x,y| x + y);
 assert_eq!(cumsum, Ndarr::from([1,3,6]));
  • Roll
let a = Ndarr::from([[1, 2], [3, 4]]);
assert_eq!(a.roll(1, 1), Ndarr::from([[2, 1], [4, 3]]))

Dyatic tensor operations

  • Generalized matrix multiplication between compatible arrays
use rapl::*
use rapl::ops::{mat_mul};
let a = Ndarr::from(1..7).reshape(&[2,3]).unwrap();
let b = Ndarr::from(1..7).reshape(&[3,2]).unwrap();
    
let matmul = mat_mul(a, b))
  • APL inspired Inner Product.
    let a = Ndarr::from(1..7).reshape(&[2,3]).unwrap();
    let b = Ndarr::from(1..7).reshape(&[3,2]).unwrap();
    
    let inner = rapl::ops::inner_product(|x,y| x*y, |x,y| x+y, a.clone(), b.clone());
    assert_eq!(inner, rapl::ops::mat_mul(a, b))

  • Outer Product.
    let suits = Ndarr::from(["","","",""]);
    let ranks = Ndarr::from(["2","3","4","5","6","7","8","9","10","J","Q","K","A"]);

    let add_str = |x: &str, y: &str| (x.to_owned() + y);

    let deck = ops::outer_product( add_str, ranks, suits).flatten(); //All cards in a deck

Complex numbers

You can ergonomically do operations between native numeric types and complex types C<T> with a simple and clean interface.

use rapl::*;
// Complex sclars
    let z = 1 + 2.i();
    assert_eq!(z, C(1,2));
    assert_eq!(z - 3, -2 + 2.i());

Seamlessly work with complex numbers, and complex tensors.

use rapl::*;
// Complex tensors
let arr = Ndarr::from([1, 2, 3]);
let arr_z = arr + -1 + 2.i();
assert_eq!(arr_z, Ndarr::from([C(0,2), C(1,2), C(2,2)]));
assert_eq!(arr_z.im(), Ndarr::from([2,2,2]));

Dead Simple 1D and 2D FFT

    let signal = Ndarr::linspace(-10., 10., 100).sin();
    let signal_fft = signal.to_complex().fft();

Image to Array and Array to Image conversion

You can easily work with images of almost any format. rapl provides helpful functions to open images as both RGB and Luma Ndarr, and also save them to your preferred format.

use rapl::*;
use rapl::utils::rapl_img;

fn main() {
    //open RGB image as  Ndarr<u8,3>
    let img: Ndarr<u8,U3> = rapl_img::open_rgbu8(&"image_name.jpg").unwrap();
    //Split RGB channels by Slicing along 3'th axis.
    let channels: Vec<Ndarr<u8,U2>> = img.slice_at(2);
    //select blue channel and save it as black and white image.
    channels[2].save_as_luma(&"blue_channel.png", rapl_img::ImageFormat::Png);
}

Features in development:

  • Port to stable Rust
  • Native support for complex numbers.
  • Line space and meshigrid initialization.
  • Random array creation.
  • 1D and 2D FFT.
  • Matrix inversion.
  • Image to array conversion.
  • Array to image conversion.
  • APL-inspired rotate function.
  • Commonly use ML functions like Relu, Softmax etc.
  • Support for existing plotting libraries in rust.
  • Mutable slicing.
  • Other Linear algebra functionalities: Eigen, LU, Gauss Jordan, Etc.
  • Automatic differentiation.

Dependencies

~2.1–4.5MB
~80K SLoC