Cosinity

Cosinity is a lightweight, zero-dependency NPM package for calculating the cosine similarity between two vectors. It's particularly useful for working with vector embeddings, such as those obtained from the OpenAI Embedding API, enabling applications in semantic similarity, recommendation systems, and more.

Table of Contents

• Features
• Installation
• Usage
  • ES6 Modules
  • CommonJS
• API
• Examples
  • Basic Example
  • Using with OpenAI Embeddings
• Error Handling
• Performance Considerations
• Contributing
• License

Features

• Lightweight: Minimalistic implementation without external dependencies.
• Flexible: Works with both ES6 Modules (import) and CommonJS (require).
• TypeScript Support: Built-in TypeScript type definitions for type safety.
• Zero Dependencies: No additional packages required.
• High Performance: Optimized for performance with large vectors.

Installation

You can install Cosinity via NPM:

1npm install cosinity

Or with Yarn:

1yarn add cosinity

Usage

ES6 Modules

1import cosineSimilarity from "cosinity";
2
3const vectorA = [1, 2, 3];
4const vectorB = [4, 5, 6];
5
6const similarity = cosineSimilarity(vectorA, vectorB);
7console.log("Cosine Similarity:", similarity);

CommonJS

1const cosineSimilarity = require("cosinity");
2
3const vectorA = [1, 2, 3];
4const vectorB = [4, 5, 6];
5
6const similarity = cosineSimilarity(vectorA, vectorB);
7console.log("Cosine Similarity:", similarity);

API

cosineSimilarity(vectorA, vectorB)

Calculates the cosine similarity between two vectors.

Parameters

• vectorA number[]: The first vector.
• vectorB number[]: The second vector.

Returns

• number: The cosine similarity between vectorA and vectorB. The value ranges from -1 (exact opposite) to 1 (exact same), where 0 indicates orthogonality (no similarity).

Throws

• Error: If the input vectors are not of the same length or are empty.

Examples

Basic Example

1import cosineSimilarity from "cosinity";
2
3const vectorA = [0, 1];
4const vectorB = [1, 0];
5
6const similarity = cosineSimilarity(vectorA, vectorB);
7console.log("Cosine Similarity:", similarity); // Output: 0

Using with OpenAI Embeddings

Cosinity can be integrated with OpenAI's Embedding API to calculate the similarity between text snippets.

1import OpenAI from "openai";
2import cosineSimilarity from "cosinity";
3
4const openai = new OpenAI({
5  apiKey: "YOUR_OPENAI_API_KEY", // Replace with your OpenAI API key
6});
7
8async function getEmbedding(text) {
9  const response = await openai.embeddings.create({
10    model: "text-embedding-3-small",
11    input: text,
12    encoding_format: "float",
13  });
14  return response.data[0].embedding;
15}
16
17async function compareTexts(text1, text2) {
18  const [embedding1, embedding2] = await Promise.all([
19    getEmbedding(text1),
20    getEmbedding(text2),
21  ]);
22
23  const similarity = cosineSimilarity(embedding1, embedding2);
24
25  console.log(
26    `Cosine Similarity between "${text1}" and "${text2}":`,
27    similarity
28  );
29}
30
31compareTexts("Hello, world!", "Hi, universe!");

Sample Output

Cosine Similarity between "Hello, world!" and "Hi, universe!": 0.87654321

Error Handling

The cosineSimilarity function performs input validation:

• Vector Length Mismatch: Throws an error if vectors vectorA and vectorB are not of the same length.
• Empty Vectors: Throws an error if either vector is empty.
• Non-Numeric Values: Throws an error if vectors contain non-numeric values.

Example:

1try {
2  cosineSimilarity([1, 2], [1, 2, 3]);
3} catch (error) {
4  console.error(error.message); // Output: Vectors must be of the same length and not empty.
5}

Performance Considerations

• Large Vectors: Cosinity is optimized for performance, but when working with extremely large vectors (e.g., embeddings with thousands of dimensions), consider batching or streaming if you experience performance issues.
• Floating-Point Precision: Be aware of floating-point precision limitations when dealing with very small or very large numbers.

Contributing

Contributions are welcome!

1. Fork the repository.

2. Clone your fork:

   1git clone https://github.com/develanet/cosinity.git

3. Create a new branch:

   1git checkout -b feature/my-new-feature

4. Commit your changes:

   1git commit -am 'Add new feature'

5. Push to the branch:

   1git push origin feature/my-new-feature

6. Submit a Pull Request.

Please make sure your code passes existing tests and add new tests for your features.

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Made with ❤️ by Zay

Acknowledgments

• Inspired by the need for a simple way to calculate cosine similarity for vector embeddings.
• Thanks to the OpenAI community for the support and collaboration.

Contact

For any questions or suggestions, feel free to open an issue or contact me at isaias@develanet.com.