Scientific computing, native to JavaScript.

The suite

Seven packages, one architecture.

Small, dependency-free foundations under opinionated applications. The foundations are MIT and useful on their own; the applications compose them into the tools you reach for. Nothing depends on anything heavier than the package above it.

Applications GPL

tangent/dsData scienceStatistics, machine learning, multivariate analysis and plots, declarative and tidy.

Run example ›

tangent/mcBayesian inferencePyMC-style models with NUTS and HMC, on analytic gradients.

Run example ›

tangent/semStructural equation modelingConfirmatory factor analysis, path analysis and full SEM with lavaan-style syntax.experimental

Run example ›

↓

Foundations MIT

tangent/optOptimizationMinimize, L-BFGS, curve fitting with bounds and robust losses, root finding.

Run example ›

tangent/probaProbabilityDistributions with analytic log-density gradients, cdf, quantiles, sampling.

Run example ›

tangent/linaLinear algebraSolve, SVD, symmetric eigendecomposition, QR, Cholesky, least squares.

Run example ›

tangent/odeDifferential equationsAdaptive Dormand-Prince, a stiff Rosenbrock solver, dense output, events.

Run example ›

See it run

Every example is a live notebook.

Open any example at note.tangent.to and run it in the browser, no install. The same code you read is the code that runs.

Why trust it

Checked against the tools you already believe.

A scientific library is only as good as its agreement with the references. Every package ships a comparison suite that runs its own results against the Python or R equivalent on shared problems. These are measured agreements, not aspirations.

1e-13proba densities and quantiles versus scipy.stats, across the body and both tails.

1e-15lina factorizations versus numpy, at machine precision on seeded matrices.

exactsem matches R’s lavaan on the Holzinger-Swineford benchmark: the same chi-square of 85.306 and the same fit indices.