tavianator / acap Goto Github PK

Hello.

As the search results consists of values instead of indices, I have tried to use such kind of wrapper to access to the indice:

        struct IndexedCoordinates {
            index: usize,
            coords: Vec<f64> // already implment all the necessary traits
        }

and implement the appropriate traits to make it useable.
Although here is no issue to implement the Coordinates trait:

        impl Coordinates for IndexedCoordinates {
            type Value = f64;

            fn dims(&self) -> usize {
                self.coords.dims()
            }

            fn coord(&self, i: usize) -> Self::Value {
                self.coords.coord(i)
            }
        }

I am unable to implement the traits that allow to search a neighbors.
If I am right, I have to implement Proximity for Euclidean<IndexedCoordinates> which is not possible.

Am I wrong in the design of my indexed coordinates ? If so, could you guide me ? If no, is there a possibility to add such ability to the library ?
Thanks

I want to be able to iterate over all the points in the tree. Since those structures store them as flat arrays, it seems like it would be easy enough to expose a slice reference and let user do whatever they want on the read-only items.

See https://github.com/erikbern/ann-benchmarks

The state of the art is currently HNSW.

Based on https://github.com/tavianator/kd-forest/blob/adaafdd7043507cbceae65e78c38954e47103b5c/src/metric/approx.rs but generic.

Related to the VP trees with "stretched" triangle inequality from https://github.com/nmslib/nmslib