Introduction

Typed Matrices in Numz are an extension of Zikojs Matrix, designed to provide high-performance, memory-efficient, and type-safe matrix operations.

Unlike regular matrices that rely on standard JavaScript arrays, Typed Matrices are backed by TypedArray structures, enabling contiguous memory layout and faster numerical computation.

---

TypedMatrix Variants

Numz provides multiple TypedMatrix classes, each mapped to a specific JavaScript TypedArray implementation.

Floating-Point Matrices

Class	Backing Type	Precision	Use case
F16Matrix	Float16Array	16-bit	ML, GPU-style data
F32Matrix	Float32Array	32-bit	GPU, graphics, signals
F64Matrix	Float64Array	64-bit	Scientific computing

Integer Matrices

Class	Backing Type	Range	Use case
I8Matrix	Int8Array	−128 → 127	Compact data
U8Matrix	Uint8Array	0 → 255	Images, buffers
I16Matrix	Int16Array	−32K → 32K	DSP, audio
U16Matrix	Uint16Array	0 → 65K	Indexing
I32Matrix	Int32Array	32-bit	General integers
U32Matrix	Uint32Array	32-bit	Large indices

Big Integer Matrices

Class	Backing Type	Notes
BI64Matrix	BigInt64Array	Requires BigInt
BU64Matrix	BigUint64Array	Large integer arithmetic

---

Example

import { F64Matrix } from 'numz';

const A = new F64Matrix(2, 2, [
  1, 2,
  3, 4
]);

A.add(2).mul(3);

Note

During the construction phase, the provided data is accepted as a regular JavaScript array only as a temporary input. Internally, Numz copies the values into a TypedArray (Float64Array in this case), and the original array is discarded and no longer used.

This guarantees:

• contiguous memory layout
• predictable numeric type
• better performance for subsequent operations