Starlight Eval

Starlight Eval is a lightweight evaluation library for machine learning models in the Starlight ecosystem. It provides essential metrics for evaluating classification models such as accuracy, precision, recall, F1-score, and confusion matrices.

Designed to work seamlessly with starlight-classifier, starlight-vec, and starlight-ml.

Features

• Classification accuracy
• Precision, recall, and F1-score (per label)
• Confusion matrix generation
• Full classification report
• Zero dependencies
• Framework-agnostic

Installation

npm install starlight-eval

📚 Importing

JavaScript / ES Modules

import * as evalml from "starlight-eval";

Starlight Language

import * as evalml from "starlight-eval";

Basic Usage

const yTrue = ["tech", "tech", "programming", "programming"];
const yPred = ["tech", "programming", "programming", "programming"];

console.log(evalml.accuracy(yTrue, yPred));
console.log(evalml.confusionMatrix(yTrue, yPred));

Classification Report

const report = evalml.classificationReport(yTrue, yPred);
console.log(report);

Example output:

{
  tech: { precision: 0.5, recall: 0.5, f1: 0.5 },
  programming: { precision: 0.67, recall: 1.0, f1: 0.8 },
  accuracy: 0.75
}

Available Functions

accuracy(yTrue, yPred)

Returns the overall classification accuracy.

confusionMatrix(yTrue, yPred)

Returns a label-to-label confusion matrix.

precision(yTrue, yPred, label)

Calculates precision for a given class label.

recall(yTrue, yPred, label)

Calculates recall for a given class label.

f1Score(yTrue, yPred, label)

Calculates F1-score for a given class label.

classificationReport(yTrue, yPred)

Generates precision, recall, F1-score per label, plus overall accuracy.

Works Great With

• starlight-ml – Tokenization & text processing
• starlight-vec – TF-IDF vectorization
• starlight-classifier – Document classification
• starlight-cluster – Unsupervised learning

Philosophy

Starlight Eval is built to be:

• Simple
• Transparent
• Educational
• Production-ready

Perfect for learning ML concepts or building real applications.

License

MIT License © Dominex Macedon