Crate ddsketchy 

ddsketchy

A Rust implementation of the DDSketch quantile sketch algorithm.

DDSketch is a fast, fully-mergeable quantile sketch with relative-error guarantees: every quantile estimate is within a user-chosen multiplicative factor alpha of the true value. This makes it well-suited to latency distributions and other data that spans many orders of magnitude, where absolute-error sketches behave poorly. Sketches are cheap to merge, making them a natural fit for distributed aggregation.

Quick start

use ddsketchy::DDSketch;

let mut sketch = DDSketch::new(0.01).expect("valid alpha");
for v in [1.0, 2.0, 3.0, 4.0, 5.0] {
    sketch.add(v);
}

// Quantiles are accurate to within `alpha` (1%) relative error.
let median = sketch.quantile(0.5).unwrap();
assert!((median - 3.0).abs() <= 3.0 * 0.01);
assert_eq!(sketch.count(), 5);

Features

• serde — enables Serialize / Deserialize for DDSketch.
• python — enables PyO3 bindings (used by the Python wheel build).

More examples

See the examples/ directory in the repository for runnable examples, including serialization round-trips and the code snippets shown in the README.

Structs

DDSketch

A DDSketch quantile estimator with configurable relative accuracy.

DDSketchBuilder

Builder for configuring and constructing a DDSketch.

Enums

DDSketchError

Errors returned by DDSketch operations.