GitHub

Introduction

Deepbox is a comprehensive, type-safe TypeScript framework that unifies numerical computing, tabular data workflows, and machine learning into a single modular package. Zero runtime dependencies. 4,344 tests. Production-ready.
TypeScript
Zero Dependencies
4,344 Tests
MIT License
Modules
13
Tests
4,344
Dependencies
0
Node.js
≥ 24.13

Core Capabilities

  • deepbox/ndarray — N-dimensional arrays, broadcasting, 90+ mathematical operations
  • deepbox/dataframe — DataFrames, Series, tabular data manipulation
  • deepbox/nn + deepbox/optim — Autograd, neural network modules, optimizers
  • deepbox/ml + deepbox/preprocess + deepbox/metrics — Classical ML models, preprocessing, metrics
  • deepbox/plot — SVG/PNG plotting and visualization

Module Overview

  • deepbox/core — Types, errors, validation, dtype helpers, configuration
  • deepbox/ndarray — N-D tensors with autograd, broadcasting, 90+ ops, sparse matrices
  • deepbox/linalg — SVD, QR, LU, Cholesky, eigenvalue decomposition, solvers, norms
  • deepbox/dataframe — DataFrame + Series with 50+ operations, CSV I/O
  • deepbox/stats — Descriptive stats, correlations, hypothesis tests
  • deepbox/metrics — 40+ ML metrics (classification, regression, clustering)
  • deepbox/preprocess — Scalers, encoders, normalizers, cross-validation splits
  • deepbox/ml — Classical ML (Linear, Ridge, Lasso, Logistic, Trees, SVM, KNN, Naive Bayes, Ensembles)
  • deepbox/nn — Neural networks (Linear, Conv, RNN/LSTM/GRU, Attention, Normalization, Losses)
  • deepbox/optim — Optimizers (SGD, Adam, AdamW, RMSprop, etc.) + LR schedulers
  • deepbox/random — Distributions (uniform, normal, binomial, gamma, beta, etc.) + sampling
  • deepbox/datasets — Built-in datasets (Iris, Digits, Breast Cancer, etc.) + synthetic generators
  • deepbox/plot — SVG/PNG plotting (scatter, line, bar, hist, heatmap, contour, ML plots)

Links

  • Website — https://deepbox.dev
  • Documentation — https://deepbox.dev/docs
  • Examples — https://deepbox.dev/examples
  • Projects — https://deepbox.dev/projects
  • GitHub — https://github.com/jehaad1/Deepbox
  • npm — https://www.npmjs.com/package/deepbox

Design Principles

  • Type Safety — Full TypeScript strict mode, no `any` types, noUncheckedIndexedAccess enabled
  • Immutability — Tensor operations return new tensors, never mutate (except shuffle)
  • Functional + OOP — Stateless math functions + stateful classes for models, scalers, optimizers
  • Full broadcasting — Automatic shape broadcasting across all element-wise operations
  • TypedArray backing — Float64Array, Float32Array, Int32Array for performance
  • Tree-shakeable — Per-module imports for minimal bundle size
Next
Installation