Benchmarks

CPU Benchmarks

Speedup relative to R’s stats::lowess (higher is better):

Category	R baseline	Serial	Parallel
Clustered	2.34 ms	2.0×	2.5×
Constant Y	1.81 ms	1.7×	3.2×
Extreme Outliers	5.81 ms	1.5×	2.6×
Financial (500–5K)	0.65 ms	2.0×	1.4×
Fraction (0.05–0.67)	3.8 ms	1.6×	3.2×
Genomic (1K–100K)	11.2 ms	2.2×	2.4×
High Noise	7.08 ms	1.5×	3.6×
Iterations (0–10)	3.0 ms	1.9×	2.7×
Scale (1K–10K)	1.6 ms	1.5×	1.6×
Scientific (500–5K)	0.9 ms	1.4×	1.4×

The R column shows average time across scenarios in multi-scenario categories. Speedups are averages across the same range.

Note

The WebAssembly build runs single-threaded (no parallel option). The figures above reflect the native Node.js binding with worker-thread parallelism. WASM serial performance is comparable to the Serial column.

---

Reproducing Benchmarks

Use performance.now() to time serial WASM runs:

const { Lowess } = require('fastlowess-wasm');

function benchMs(fn, reps = 10) {
    fn(); // warm-up
    const { performance } = require('perf_hooks');
    const t0 = performance.now();
    for (let i = 0; i < reps; i++) fn();
    return (performance.now() - t0) / reps;
}

const n = 5000;
const x = Float64Array.from({ length: n }, (_, i) => (i / (n - 1)) * 10);
const y = Float64Array.from(x, (xi, i) =>
    Math.sin(xi) + (((i * 7 + 3) % 17) / 17 - 0.5) * 0.6
);

const ms = benchMs(() => new Lowess({ fraction: 0.67 }).fit(x, y));
console.log(`WASM: ${ms.toFixed(2)} ms`);

WASM: 17.73 ms