9 releases
0.1.8 | Apr 17, 2024 |
---|---|
0.1.7 | Nov 3, 2022 |
0.1.6 | Apr 23, 2022 |
#406 in Data structures
474 downloads per month
70KB
1.5K
SLoC
LSPH - Learned SPatial HashMap
fast 2d point query powered by hashmap and statistic model
The original paper of LSPH can be found here.
The LSPH uses a learned model such as a linear regression model as the hash function to predict the index in a hashmap. As a result, the learned model is more fitted to the data that stored in the hashmap, and reduces the chance of hashing collisions. Moreover, if the learned model is monotonic function(e.g. linear regression), the hash indexes are increasing as the input data increases. This property can be used to create a sorted order of buckets in a hashmap, which allow us to do range searches in a hashmap.
The LSPH supports:
- Point Query
- Rectange Query
- Radius Range Query
- Nearest Neighbor Query
Example:
use lsph::{LearnedHashMap, LinearModel};
let point_data = vec![[1., 1.], [2., 1.], [3., 2.], [4., 4.]];
let (mut map, points) = LearnedHashMap::<LinearModel<f32>, f64>::with_data(&point_data).unwrap();
assert_eq!(map.get(&[1., 1.]).is_some(), true);
assert_eq!(map.get(&[3., 1.]).is_none(), true);
assert_eq!(map.range_search(&[0., 0.], &[3., 3.]).is_some(), true);
assert_eq!(map.radius_range(&[2., 1.], 1.).is_some(), true);
assert_eq!(map.nearest_neighbor(&[2., 1.]).is_some(), true);
To Run Benchmark:
cargo bench
License
Licensed under either of
- Apache License, Version 2.0, (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)
at your option.
Dependencies
~220KB