Introduction

Deepbox v1.0.0 is a zero-runtime-dependency TypeScript framework that unifies numerical computing, tabular data, machine learning, neural networks, statistics, datasets, and plotting under one package.

v1.0.0

Zero runtime dependencies

ESM + CommonJS

Typed subpath exports

Modules

Examples

Projects

Source Files

315

What v1.0.0 includes

A single package with stable subpath exports such as deepbox/ndarray, deepbox/ml, and deepbox/nn.
Expanded numerical coverage: FFT, einsum, advanced solvers, matrix functions, sparse tensors, and richer random distributions.
Production-oriented APIs for DataFrame IO/styling, datasets, calibration, anomaly detection, pipelines, model selection, and plotting diagnostics.
A larger docs and learning surface with 50 runnable examples and 9 end-to-end projects.
A dedicated Getting Started page on devices and execution (CPU by default, optional WebGPU/WASM registration, and how Module.to(device) relates to backends).

quickstart.ts

import { parameter, tensor } from "deepbox/ndarray";import { DataFrame } from "deepbox/dataframe";import { LinearRegression } from "deepbox/ml";const x = parameter([[1, 2], [3, 4]]);const w = parameter([[0.5], [1]]);const y = x.matmul(w).sum();y.backward();const df = new DataFrame({  name: ["Alice", "Bob", "Charlie"],  score: [91, 84, 96],});const model = new LinearRegression();model.fit(tensor([[1], [2], [3], [4]]), tensor([2, 4, 6, 8]));

One package, many workflows

Deepbox keeps tensors, statistics, ML estimators, neural networks, datasets, and plotting in one import graph.

Source-compatible subpaths

Import only the module you need and keep bundle boundaries explicit with deepbox/<module> entry points.

Typed surface area

The package ships declaration files for the root namespace and every submodule, making editor tooling reliable across the full framework.

Docs built around v1.0.0

The documentation now reflects only the published v1.0.0 API surface rather than mixed legacy and current content.

Installation