6 ms·
Which vector similarity metric should I use?
- messe 3y agoEven ignoring vector magnitudes, wouldn't cosine distance as a measure of similarity only make sense if you're working with a convex set? That seems like it's far from a guarantee working in a high-dimensional space.
- imaurer 3y agoYes, cosine distance works best in convex or normalized sets. Thinking about adding this caveat. Thanks for the question.
- sharemywin 3y agoDoes this seem right? | Task | Distance Measure | |-------------------------------|-----------------------| | Document classification | Cosine Distance | | Semantic search | Cosine Distance | | Recommendation systems | Cosine Distance | | Image recognition | Euclidean Distance (L2)| | Speech recognition | Euclidean Distance (L2)| | Handwriting analysis | Euclidean Distance (L2)| | Recommendation systems | Inner Product (Dot Product)| | Collaborative filtering | Inner Product (Dot Product)| | Matrix factorization | Inner Product (Dot Product)| | Image processing | L2-Squared Distance | | Error detection and correction| Hamming Distance | | DNA sequence comparison | Hamming Distance | | Taxicab geometry | Manhattan Distance | | Chessboard distance | Manhattan Distance |
- imaurer 3y agoYes