2 releases

0.1.1 Jul 31, 2024
0.1.0 Jul 22, 2024

#386 in Database implementations

23 downloads per month

MIT license

16KB
180 lines

Grafite

Grafite is a range filter with a simple design and clear theoretical guarantees that hold regardless of the input data and query distribution.

This library is a Rust implementation of the data structure introduced by this paper: Grafite: Taming Adversarial Queries with Optimal Range Filters.

The authors of this paper also created a C++ implementation for Grafite, which can be found on one of the author's GitHub: grafite.

The Grafite data structure relies on the Elias-Fano encoding of non-decreasing integer sequences, and this library uses the vers_vecs implementation of the encoding.

Examples

use grafite::{OrderPreservingHasher, RangeFilter};

let values = [1, 2, 3, 7, 8, 9, 15, 20];

let epsilon = 0.01;
let max_query_range = 20;
let hasher = OrderPreservingHasher::new(values.len(), epsilon, max_query_range)
    .expect("The input parameters should be valid");

let rf = RangeFilter::new(values.iter().copied(), hasher);

// If there are any values in the range, it will return `true`.
assert!(rf.query(..));
assert!(rf.query(..42));
assert!(rf.query(10..));
assert!(rf.query(0..20));

// Start is inclusive.
assert!(rf.query(3..5));
assert!(rf.query(9..16));

// End is exclusive. Note that false positives are possible depending on the input `epsilon`.
assert!(!rf.query(10..15));
assert!(rf.query(10..=15));

TODO

Dependencies

~2.5MB
~47K SLoC