Starlight Classifier

Starlight Classifier is a lightweight text classification library for Starlight, built on top of starlight-vec and starlight-ml. It uses TF-IDF vectorization and a simple Naive Bayes approach for classifying documents quickly and efficiently.

Features

• Vectorizes text using TF-IDF (via starlight-vec)
• Simple Naive Bayes classifier for text
• Predict labels for single or multiple documents
• Works seamlessly with Starlight ML ecosystem
• Customizable stopwords support

Installation

npm install starlight-classifier starlight-vec starlight-ml

Usage

Importing

import { NaiveBayesClassifier, classify } from "starlight-classifier";
import * as ml from "starlight-ml";

Example

// Sample documents
const docs = [
  "I love machine learning",
  "Starlight ML is amazing",
  "Python is great for AI",
  "I enjoy coding in JavaScript"
];

// Corresponding labels
const labels = ["tech", "tech", "tech", "programming"];

// Stopwords (optional)
const stopwords = ["is", "in"];

// Train classifier
const clf = classify(docs, labels, stopwords);

// Predict a new document
const newDoc = "I love AI and machine learning";
const prediction = clf.predict(newDoc);

console.log("Predicted class:", prediction); // e.g., "tech"

// Predict multiple documents
const batch = ["I code every day", "Starlight ML is cool"];
const batchPredictions = clf.predictBatch(batch);

console.log("Batch predictions:", batchPredictions);

Advanced Usage

import { NaiveBayesClassifier } from "starlight-classifier";

// Create classifier with stopwords
const clf = new NaiveBayesClassifier(["is", "and"]);

// Fit on documents
clf.fit(docs, labels);

// Transform and analyze similarity
const vec1 = clf.vectorizer.transform(docs[0]);
const vec2 = clf.vectorizer.transform(docs[1]);
const similarity = NaiveBayesClassifier.cosine(vec1, vec2);

console.log("Cosine similarity between doc1 and doc2:", similarity);

API

NaiveBayesClassifier(stopwords = [])

• Creates a classifier instance.
• stopwords — optional array of words to ignore during vectorization.

fit(docs, labels)

• Train classifier on an array of documents with corresponding labels.

predict(doc)

• Predict the class of a single document.

predictBatch(docs)

• Predict the classes for multiple documents.

classify(docs, labels, stopwords)

• Convenience function: creates a classifier, fits it, and returns it.

Dependencies

• starlight-vec — TF-IDF vectorization for Starlight
• starlight-ml — Text preprocessing utilities

License

MIT © Dominex Macedon