deepbox/ndarray

Tensor Creation

Tensor construction, dtype-aware buffers, and the core Tensor/AnyTensor abstractions.

Creation

Tensor core

Overview

Use tensor, zeros, ones, eye, arange, linspace, geomspace, and related factories to create dense arrays.
The Tensor class is the common runtime object for dense numerical data, while AnyTensor covers Tensor and GradTensor together.
Complex arrays, Float16/BFloat16 buffers, and TensorCreateOptions support newer numerical workflows introduced in v1.0.0.

type AnyTensor

export type AnyTensor = TensorClass | GradTensorClass;

Union type representing either a Tensor or GradTensor.

type NestedArray

export type NestedArray = number | boolean | NestedArray[];

Recursive type for nested number arrays.

type SliceRange

export type SliceRange = number | { readonly start?: number; readonly end?: number; readonly step?: number; };

A single index or a start/end/step range for tensor slicing.

type TensorCreateOptions

export type TensorCreateOptions = { readonly dtype?: DType; readonly device?: Device; };

Options for the creation function.

type TensorOptions

export type TensorOptions = { readonly dtype: DType; readonly device: Device; };

Options for tensor dtype and device configuration.

type expandDims

export declare const expandDims: typeof unsqueeze;

expandDims is a public const in deepbox/ndarray.

Complex

Complex number types and typed arrays for complex64/complex128 support.

Complex64Array

Complex64 typed array — each element is two float32 values (real, imag).

Complex128Array

Complex128 typed array — each element is two float64 values (real, imag).

BFloat16Array

Software BFloat16 (Brain Floating Point) array.

Float16Array

Software IEEE 754 half-precision floating-point array.

Tensor

Multi-dimensional array (tensor) with typed storage.

arange

export declare function arange(start: number, stop?: number, step?: number, opts?: TensorCreateOptions): Tensor;

Range.

empty

export declare function empty(shape: Shape, opts?: TensorCreateOptions): Tensor;

Fill with a scalar value.

eye

export declare function eye(n: number, m?: number, k?: number, opts?: TensorCreateOptions): Tensor;

Identity matrix.

flatten

export declare function flatten(t: Tensor): Tensor;

Flatten to 1D.

full

export declare function full(shape: Shape, value: number | string, opts?: TensorCreateOptions): Tensor;

full is exported by deepbox/ndarray.

gather

export declare function gather(t: Tensor, indices: Tensor, axis: Axis): Tensor;

Gather values along an axis specified by indices.

geomspace

export declare function geomspace(start: number, stop: number, num?: number, endpoint?: boolean, opts?: TensorCreateOptions): Tensor;

Numbers spaced evenly on a log scale (geometric progression).

linspace

export declare function linspace(start: number, stop: number, num?: number, endpoint?: boolean, opts?: TensorCreateOptions): Tensor;

Evenly spaced numbers over a specified interval.

logspace

export declare function logspace(start: number, stop: number, num?: number, base?: number, endpoint?: boolean, opts?: TensorCreateOptions): Tensor;

Numbers spaced evenly on a log scale.

ones

export declare function ones(shape: Shape, opts?: TensorCreateOptions): Tensor;

All ones.

randn

export declare function randn(shape: Shape, opts?: TensorCreateOptions): Tensor;

Return a tensor filled with random samples from a standard normal distribution.

reshape

export declare function reshape(t: Tensor, rawShape: Shape): Tensor;

Change shape (view) without copying.

slice

export declare function slice(t: Tensor, ...ranges: SliceRange[]): Tensor;

Slice a tensor.

tensor

export declare function tensor(data: NestedArray | StringNestedArray | TypedArray, opts?: TensorCreateOptions): Tensor;

tensor is exported by deepbox/ndarray.

transpose

export declare function transpose(t: Tensor, axes?: readonly number[]): Tensor;

Transpose tensor dimensions.

zeros

export declare function zeros(shape: Shape, opts?: TensorCreateOptions): Tensor;

All zeros.

squeeze

export declare function squeeze(t: Tensor, axis?: Axis | Axis[]): Tensor;

Remove single-dimensional entries from the shape.

unsqueeze

export declare function unsqueeze(t: Tensor, axis: number): Tensor;

Expand the shape by inserting a new axis.

ndarray-tensor.ts

import {  Complex,  arange,  eye,  reshape,  tensor,  zeros,} from "deepbox/ndarray";const a = tensor([[1, 2], [3, 4]]);const b = zeros([2, 2]);const c = reshape(arange(0, 6), [2, 3]);const i = eye(3);const z = new Complex(1, -2);console.log(a.shape, b.dtype, c.toString(), i.toString(), z.toString());

Utilities, Serialization & Parallelism

Math & Reductions