skjson-js
v1.0.3
Published
Zero-dependency universal inference library for skjson models
Maintainers
hongyaokReadme
skjson-js
skjson-js is a zero-dependency, universal Javascript inference library for Python skjson exported models. It parses the JSON representation of trained scikit-learn models and accurately executes inference directly in Node.js or any browser environment.
Features
- Zero dependencies: No web assembly, no native bindings, pure Javascript/TypeScript.
- Universal compatibility: Works effortlessly in browsers and Node.js.
- 100% Mathematical Parity: Strict test suites guarantee identical predictions compared to Python scikit-learn outcomes.
Installation
npm install skjson-jsSupported Models
Currently, skjson-js supports inference for the following model types:
- Linear Models:
LinearRegression,LogisticRegression,Ridge,Lasso - Trees & Forests:
DecisionTreeClassifier,DecisionTreeRegressor,RandomForestClassifier,RandomForestRegressor - Naive Bayes:
GaussianNB - Support Vector Machines:
SVC,SVR(with linear kernels) - Nearest Neighbors:
KNeighborsClassifier,KNeighborsRegressor - Preprocessing:
StandardScaler,MinMaxScaler
Usage Example
First, export your model in Python using skjson:
from sklearn.datasets import load_iris
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
import skjson
# Load the Iris dataset and split
X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
# Train a Random Forest classifier
clf = RandomForestClassifier(random_state=42)
clf.fit(X_train, y_train)
# Export the trained model to JSON
skjson.save(clf, 'model.json')Then, load and predict in Javascript:
import { loadModel } from 'skjson-js';
import myModelJson from './model.json'; // Or fetch() in the browser
// Instantiate the predictor
const predictor = loadModel(myModelJson);
// Perform inference
const predictions = predictor.predict([
[2.3, -1.0, 0.4, 5.0],
[0.5, 0.2, 0.1, -1.1]
]);
console.log('Predictions:', predictions);
// For classification models, you can also get probabilities
if (predictor.predict_proba) {
const probas = predictor.predict_proba([
[2.3, -1.0, 0.4, 5.0]
]);
console.log('Probabilities:', probas);
}Preprocessing Example
import { loadModel } from 'skjson-js';
import scalerJson from './scaler.json';
const scaler = loadModel(scalerJson);
// Transformers use the `.transform()` method instead of predict()
const transformed = scaler.transform([
[2.3, -1.0, 0.4, 5.0],
[0.5, 0.2, 0.1, -1.1]
]);
console.log('Transformed:', transformed);HTML-Only Usage (Browser via CDN)
You can use skjson-js directly in plain HTML without any bundlers by utilizing modern ES modules and a CDN like unpkg.
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>skjson-js HTML Demo</title>
</head>
<body>
<h1>Prediction Result: <span id="result">Loading...</span></h1>
<script type="module">
// Import directly from a CDN
import { loadModel } from 'https://unpkg.com/skjson-js/dist/index.js';
async function run() {
// Fetch your JSON model from a local path or URL
const response = await fetch('./model.json');
const modelJson = await response.json();
// Load and predict
const predictor = loadModel(modelJson);
const preds = predictor.predict([[2.3, -1.0, 0.4, 5.0]]);
document.getElementById('result').innerText = preds[0];
}
run();
</script>
</body>
</html>Security
This package operates with 0 runtime dependencies to ensure full immunity against supply-chain attacks. CI enforces npm audit on all development tools prior to publishing.
AI use
This project uses AI-assisted development with Gemini and Claude Opus.
License
ISC