fdars_core/depth/
mod.rs

1//! Depth measures for functional data.
2//!
3//! This module provides various depth measures for assessing the centrality
4//! of functional observations within a reference sample.
5
6use crate::matrix::FdMatrix;
7use crate::maybe_par_chunks_mut_enumerate;
8use rand::prelude::*;
9use rand_distr::StandardNormal;
10
11pub mod band;
12pub mod fraiman_muniz;
13pub mod modal;
14pub mod random_projection;
15pub mod random_tukey;
16pub mod spatial;
17
18#[cfg(test)]
19mod tests;
20
21// Re-export all public functions
22pub use band::{band_1d, modified_band_1d, modified_epigraph_index_1d};
23pub use fraiman_muniz::{fraiman_muniz_1d, fraiman_muniz_2d};
24pub use modal::{modal_1d, modal_2d};
25pub use random_projection::{
26    random_projection_1d, random_projection_1d_seeded, random_projection_2d,
27};
28pub use random_tukey::{random_tukey_1d, random_tukey_1d_seeded, random_tukey_2d};
29pub use spatial::{
30    functional_spatial_1d, functional_spatial_2d, kernel_functional_spatial_1d,
31    kernel_functional_spatial_2d,
32};
33
34// ---------------------------------------------------------------------------
35// Shared helpers
36// ---------------------------------------------------------------------------
37
38/// Generate `nproj` unit-norm random projection vectors of dimension `m`.
39///
40/// Returns a flat buffer of length `nproj * m` where projection `p` occupies
41/// `[p*m .. (p+1)*m]`.
42///
43/// If `seed` is Some, uses a deterministic RNG seeded from the given value.
44pub(super) fn generate_random_projections(nproj: usize, m: usize, seed: Option<u64>) -> Vec<f64> {
45    let mut rng: Box<dyn RngCore> = match seed {
46        Some(s) => Box::new(StdRng::seed_from_u64(s)),
47        None => Box::new(rand::thread_rng()),
48    };
49    let mut projections = vec![0.0; nproj * m];
50    for p_idx in 0..nproj {
51        let base = p_idx * m;
52        let mut norm_sq = 0.0;
53        for t in 0..m {
54            let v: f64 = rng.sample(StandardNormal);
55            projections[base + t] = v;
56            norm_sq += v * v;
57        }
58        let inv_norm = 1.0 / norm_sq.sqrt();
59        for t in 0..m {
60            projections[base + t] *= inv_norm;
61        }
62    }
63    projections
64}
65
66/// Project each reference curve onto each projection direction and sort.
67///
68/// Returns a flat buffer of length `nproj * nori` where the sorted projections
69/// for direction `p` occupy `[p*nori .. (p+1)*nori]`.
70pub(super) fn project_and_sort_reference(
71    data_ori: &FdMatrix,
72    projections: &[f64],
73    nproj: usize,
74    nori: usize,
75    m: usize,
76) -> Vec<f64> {
77    let mut sorted = vec![0.0; nproj * nori];
78    maybe_par_chunks_mut_enumerate!(sorted, nori, |(p_idx, spo): (usize, &mut [f64])| {
79        let proj = &projections[p_idx * m..(p_idx + 1) * m];
80        for j in 0..nori {
81            let mut dot = 0.0;
82            for t in 0..m {
83                dot += data_ori[(j, t)] * proj[t];
84            }
85            spo[j] = dot;
86        }
87        spo.sort_unstable_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));
88    });
89    sorted
90}
91
92/// Shared implementation for random projection-based depth measures.
93///
94/// Generates `nproj` random projections, projects both object and reference
95/// curves to scalars, and computes univariate depth via binary-search ranking.
96/// The `aggregate` and `finalize` closures control how per-projection depths
97/// are combined (e.g. average for RP depth, minimum for Tukey depth).
98pub(super) fn random_depth_core(
99    data_obj: &FdMatrix,
100    data_ori: &FdMatrix,
101    nproj: usize,
102    seed: Option<u64>,
103    init: f64,
104    aggregate: impl Fn(f64, f64) -> f64 + Sync,
105    finalize: impl Fn(f64, usize) -> f64 + Sync,
106) -> Vec<f64> {
107    use crate::iter_maybe_parallel;
108    #[cfg(feature = "parallel")]
109    use rayon::iter::ParallelIterator;
110
111    let nobj = data_obj.nrows();
112    let nori = data_ori.nrows();
113    let m = data_obj.ncols();
114
115    if nobj == 0 || nori == 0 || m == 0 || nproj == 0 {
116        return Vec::new();
117    }
118
119    let projections = generate_random_projections(nproj, m, seed);
120    let sorted_proj_ori = project_and_sort_reference(data_ori, &projections, nproj, nori, m);
121    let denom = nori as f64 + 1.0;
122
123    iter_maybe_parallel!(0..nobj)
124        .map(|i| {
125            let mut acc = init;
126            for p_idx in 0..nproj {
127                let proj = &projections[p_idx * m..(p_idx + 1) * m];
128                let sorted_ori = &sorted_proj_ori[p_idx * nori..(p_idx + 1) * nori];
129
130                let mut proj_i = 0.0;
131                for t in 0..m {
132                    proj_i += data_obj[(i, t)] * proj[t];
133                }
134
135                let below = sorted_ori.partition_point(|&v| v < proj_i);
136                let above = nori - sorted_ori.partition_point(|&v| v <= proj_i);
137                let depth = (below.min(above) as f64 + 1.0) / denom;
138                acc = aggregate(acc, depth);
139            }
140            finalize(acc, nproj)
141        })
142        .collect()
143}
fdars_core/depth/mod.rs

fdars_core/depth/
mod.rs
