linalg-wasm

A WebAssembly module for linear algebra operations optimized for Contextual Bandit algorithms.

Features

• UCB Value Calculation: Bulk computation of Upper Confidence Bound values for LinUCB algorithm
• Model Updates: Efficient bandit model updates using Sherman-Morrison formula
• Cosine Similarity: Fast cosine similarity calculations for vector comparisons
• Similarity Matrix: Computation of similarity matrices for multiple vectors

Installation

npm install linalg-wasm

Usage

import init, {
  get_ucb_values_bulk,
  update_bandit_model,
  cosine_similarity
} from 'linalg-wasm';

async function main() {
  // Initialize the WASM module
  await init();

  // Your code here
}

API

get_ucb_values_bulk(model, articles, user_ctr)

Calculates UCB values for multiple articles.

update_bandit_model(model, embedding, reward)

Updates the bandit model with new reward information.

cosine_similarity(vec1, vec2)

Computes cosine similarity between two vectors.

cosine_similarity_bulk(vec1s, vec2s)

Computes cosine similarities for multiple vector pairs.

calculate_similarity_matrix(vectors)

Computes a similarity matrix for a set of vectors.

cosine_similarity_one_to_many(target, candidates)

Computes similarities between one target vector and multiple candidates.

License

MIT