umap-rs 0.4.5

Fast, parallel, memory-efficient Rust implementation of UMAP
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62

use ndarray::Array1;
use ndarray::ArrayView1;
use std::fmt::Debug;

/// Metric type determines which optimization path to use.
#[derive(Debug, Clone, Copy, serde::Serialize, serde::Deserialize)]
pub enum MetricType {
  /// Euclidean-like metrics with fast squared distance.
  Euclidean,
  /// Generic metrics requiring gradient computation.
  Generic,
}

/// A distance metric for embedding spaces.
///
/// Metrics must be able to compute both distances and their gradients for use
/// in gradient descent optimization. Thread-safety (Send + Sync) is required
/// for parallel optimization.
pub trait Metric: Debug + Send + Sync {
  /// Returns the type of this metric for optimization path selection.
  fn metric_type(&self) -> MetricType {
    MetricType::Generic
  }
  /// Compute the distance between two points and its gradient.
  ///
  /// # Arguments
  ///
  /// * `a` - First point
  /// * `b` - Second point
  ///
  /// # Returns
  ///
  /// A tuple of (distance, gradient) where:
  /// - `distance` is the scalar distance value
  /// - `gradient` is ∂distance/∂a (gradient with respect to the first point)
  fn distance(&self, a: ArrayView1<f32>, b: ArrayView1<f32>) -> (f32, Array1<f32>);

  /// Distance threshold beyond which points are considered disconnected.
  ///
  /// For unbounded metrics like Euclidean, this is typically infinity.
  /// For bounded metrics (e.g., on spheres or in hyperbolic space), this
  /// should be the maximum meaningful distance.
  ///
  /// Default: f32::INFINITY
  fn disconnection_threshold(&self) -> f32 {
    f32::INFINITY
  }

  /// Optional optimized computation of squared distance without square root.
  ///
  /// For Euclidean and related metrics, computing squared distance is faster
  /// than the full distance (avoids sqrt). Return `Some(dist_squared)` if
  /// your metric supports this optimization, `None` otherwise.
  ///
  /// This is used to select between optimized (Euclidean) and generic
  /// optimization paths.
  ///
  /// Default: None
  fn squared_distance(&self, _a: ArrayView1<f32>, _b: ArrayView1<f32>) -> Option<f32> {
    None
  }
}