cos-similarity

cos-similarity

Computes the cosine similarity between two vectors

• Optimized for dense vectors, that is, arrays
• Works with different length vectors, ignoring the additional dimensions of the longest one
• Works with zero vectors, returning 0 if any vector is 0, undefined or []

Installing

npm install cos-similarity

API

cosSimilarity(vectorA, vectorB)

Returns the cosine similarity between the given vectorA and vectorB. Returns 0 when given a zero vector, [], undefined or nothing.

import cosSimilarity from "cos-similarity";

cosSimilarity([1, 2, 4], [1, 0, 2]); // -> 0.8783100656536799
cosSimilarity([1, 2, 0], [1, 2, 0]); // -> 1
cosSimilarity([2, 0, 0], [0, 2, 0]); // -> 0
cosSimilarity([-1, -2, 0], [1, 2, 0]); // -> -1

Benchmark

cosine-similarity x 56,881 ops/sec ±0.25% (97 runs sampled)
compute-cosine-similarity x 1,036,728 ops/sec ±0.12% (98 runs sampled)
doc-similarity cosineSim x 7,230,600 ops/sec ±0.86% (94 runs sampled)
cos-similarity x 7,317,571 ops/sec ±0.73% (95 runs sampled)

Fastest is cos-similarity

To run the benchmark:

git clone https://github.com/rtomrud/cos-similarity.git
cd cos-similarity
npm install
npm run bench

License

MIT