Cosim

Compute the cosine similarity of two vectors. Useful when working with LLMs. Fully typed, zero dependencies.

Install

npm install cosim

Usage

You can specify vectors as arrays, or as Javascript objects with string keys and number values.

Constraints:

• Both arguments must have the same type: you can't compare objects with arrays.
• Vectors must have the same length.
• Vector values must be numbers.

Example usage

With objects:

import { similarity } from "cosim"

const result = similarity(
  { x: 1, y: 3, z: -5}, 
  { x: 4, y: -2, z: -1 }
).toFixed(3); // 0.111

With arrays:

const result = similarity(
  [0.1, 0.5, 0.9], 
  [0.3, 0.8, 0.2]
).toFixed(3); // 0.672

The library also exposes its helper for computing the dot product of two vectors:

  const result = dotProduct({ x: -1, y: -2 }, { x: -3, y: -4 });
  console.log(result) // 11

Contributing

This uses pnpm for package management and tsup for bundling.

To contribute:

1. pnpm install
2. Make your changes in src
3. pnpm build
4. Add new tests in test
5. Submit a PR 🥳