metabench/vectorious

High performance linear algebra.

A high performance linear algebra library, written in JavaScript and optimized with C++ bindings to BLAS.

Usage

In node.js

$ npm install vectorious

var v = require('vectorious'),
    Matrix = v.Matrix,
    Vector = v.Vector,
    BLAS = v.BLAS; // access BLAS routines

Will use your local BLAS copy (if any). Some notes for different operating systems:

• OSX - by default included in the Accelerate framework
• Debian/Ubuntu - different options, easiest is to apt-get install libblas-dev
• Windows - https://icl.cs.utk.edu/lapack-for-windows/

In browser

Download a release and use it like this:

<script src="vectorious-4.x.x.min.js"></script>

<script>
  var A = new Matrix([[1], [2], [3]]),
      B = new Matrix([[1, 3, 5]]),
      C = A.multiply(B);

  console.log('C:', C.toArray());
  /* C: [
    [1, 3, 5],
    [2, 6, 10],
    [3, 9, 15]
  ] */
</script>

Examples

Basic

• Solving linear systems of equations
• Using low-level BLAS routines

Machine learning

• Neural network (by @lucidrains)
• Logistic regression

Documentation

The documentation is located in the wiki section of this repository.

Go to wiki.

Benchmarks

Internal benchmarks are located in the wiki section of this repository.

Go to wiki.

Compared to other libraries

The following benchmarks compare Vectorious 4.1.0 with three popular matrix/vector libraries:

• Numeric.js
• Sylvester
• Math.js

The graphs show operations per second on the vertical (y) axis.

Below is a graph comparing the vector operations add, angle, dot, magnitude (aka L2-norm), normalize and scale.

The operations were performed on vectors generated with Vector.random(1048576).

Below is a graph comparing the matrix operations add, scale and transpose.

The operations were performed on matrices generated with Matrix.random(512, 512).