humpday

23 derivative-free optimizers in zero-dependency JavaScript. This is the browser/Node twin of the Python humpday package; parity tests in the repository hold the two implementations in agreement, so the algorithm you run here is the algorithm the Python package ships.

Every optimizer minimises a black-box function on the unit cube [0,1]^n under a hard evaluation budget. No gradients, no dependencies, no build step.

Install

npm install humpday

Use

const { Alloy, NelderMead, DifferentialEvolution } = require('humpday');

const objective = (x) => (x[0] - 0.3) ** 2 + Math.abs(x[1] - 0.6);

const opt = new Alloy(objective, 200, 2);   // objective, nTrials, nDim
const { bestValue, bestX } = opt.optimize();

Or create by name:

const humpday = require('humpday');
const opt = new humpday.algorithms['CMAEvolutionStrategy'](objective, 200, 2);

The roster

PRIMA trust-region (UOBYQA, NEWUOA, BOBYQA), classic numerical (NelderMead, Powell, LBFGSB), evolutionary and swarm (DifferentialEvolution, ParticleSwarm, GeneticAlgorithm, CMAEvolutionStrategy, EvolutionStrategy), metaheuristics (SimulatedAnnealing, FireflyAlgorithm, AntColonyOpt, HarmonySearch), model-based (BayesianOpt), local and pattern search (Rechenberg, HillClimbing, CoordinateDescent, PatternSearch), baselines (RandomSearch, GridSearch), and Alloy, a machine-designed blend of five classical methods validated on held-out problems (see the papers).

Docs and live demos

Every optimizer can be raced in the browser on real physics and engineering problems at humpday.microprediction.org.

MIT licensed.