2 ms·
Can someone help explaining why this hash method could improve distance calculation for k-NN? What does it improve compared with Geohash or k-d tree structure?
by zocoi 9y ago
Can someone help explaining why this hash method could improve distance calculation for k-NN? What does it improve compared with Geohash or k-d tree structure?
- imron 9y ago> What does it improve compared with Geohash From what I can tell, it's the exact same algorithm used by Geohash.
- mmalone 9y agoYep. Geohash is just a fancy name for a z-curve.
- mmalone 9y ago1. A geohash is a z-curve. 2. It won't be better than a k-d tree. Dimensionality reduction is usually done when you have really truly huge numbers of dimensions that are sparsely populated and you don't care much about some information loss (e.g., for machine learning) or, in this case, when you have an easy way to create a single dimensional index and you want to force multi-dimensional data into it. In the general case a k-d tree would be objectively better in terms of performance.
- AstralStorm 9y agoAnd in very many or highly sparse dimensions with few or lazy updates, R-tree or derivatives.