When would you choose between Cosine Similarity, Inner Product, and Euclidean Distance (L2) for your vector search?

Question

Accepted Answer

Cosine similarity is more expensive (in terms of compute) than Euclidean distance because it first normalizes a vector. You would use cosine if you don't already have normalized vectors, and Euclidean or Inner Product otherwise.

When would you choose between Cosine Similarity, Inner Product, and Euclidean Distance (L2) for your vector search?

Practice Your Response

Similar Questions in Context and Retrieval (RAG)