# skaters
[](https://www.npmjs.com/package/skaters)
[](https://www.npmjs.com/package/skaters)
[](https://github.com/microprediction/skaters/blob/main/LICENSE)
### ▶ [Watch it race live in your browser](https://skaters.microprediction.org/demos/race.html)
The actual JavaScript port racing `arima`, `@bsull/augurs` (ETS and MSTL), and
Prophet (Stan compiled to WASM) on 150 real FRED series — every method scored on
held-out log-likelihood, rolling one-step-ahead, no server. `laplace` keeps up
with a moving average while the refit baselines grind. **[Start here →](https://skaters.microprediction.org/demos/race.html)**
---
Fast univariate **online distributional** time-series forecasting. Zero
dependencies, runs in Node or the browser. This is the JavaScript port of the
Python [`skaters`](https://pypi.org/project/skaters/) package, numerically
identical to it within `1e-6` (enforced on every release by a parity checker
against the Python reference).
Every prediction is a full predictive **distribution** (a `Dist`), so it can be
scored on log-likelihood — not just a point or an interval.
## Install
```
npm install skaters
```
## Quickstart
```js
import { laplace } from "skaters";
const f = laplace(1); // online, O(1) per step, the general-purpose default
let state = null;
for (const y of stream) {
let dists;
[dists, state] = f(y, state);
const d = dists[0];
d.mean; // point forecast
d.std; // uncertainty
d.quantile(0.975); // 97.5th percentile
d.logpdf(y); // a real density — scorable on log-likelihood
d.crps(y); // ...and on CRPS
}
```
In the browser, import the hosted module directly — no build step:
```html
```
## What's exported
`laplace` and `buildCandidates`; the `Dist` object; transforms (`difference`,
`standardize`, `garch`, `ar`, `holtLinear`, `powerTransform`, …); ensembles
(`precisionWeightedEnsemble`, `bayesianEnsemble`, `terminalLeafEnsemble`); leaves
(`scaleMixtureLeaf`, `crpsLeaf`, `garchLeaf`); `multiscale`, `sticky`, periodicity
and covariance helpers; and spec (de)serialisers. See `index.mjs` for the full
surface.
## Notes
- `laplace` is the general-purpose default; its signature is
`laplace(k = 1, objective = "crps", sticky = true, scales = null)`. On
price/return series with volatility clustering there is no free lunch — a
dedicated GARCH-t model is the right tool there.
- The port is a faithful mirror of the Python package; the two agree to `1e-6`.
## Links
- Docs & benchmarks:
- Paper: *Transforms All the Way Down — Automatic Online Distributional Forecasting by Conjugation*
- Python package:
- Source:
MIT © Peter Cotton