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 dispatch;
13pub mod fraiman_muniz;
14pub mod modal;
15pub mod random_projection;
16pub mod random_tukey;
17pub mod rpd;
18pub mod spatial;
19
20#[cfg(test)]
21mod tests;
22
23// Re-export all public functions
24pub use band::{band_1d, modified_band_1d, modified_epigraph_index_1d};
25pub use dispatch::{functional_boxplot, functional_depth, DepthMethod, FunctionalBoxplotResult};
26pub use fraiman_muniz::{fraiman_muniz_1d, fraiman_muniz_2d};
27pub use modal::{modal_1d, modal_2d};
28pub use random_projection::{
29    random_projection_1d, random_projection_1d_seeded, random_projection_2d,
30};
31pub use random_tukey::{random_tukey_1d, random_tukey_1d_seeded, random_tukey_2d};
32pub use rpd::{rpd_depth_1d, rpd_depth_1d_seeded};
33pub use spatial::{
34    functional_spatial_1d, functional_spatial_2d, kernel_functional_spatial_1d,
35    kernel_functional_spatial_2d,
36};
37
38// ---------------------------------------------------------------------------
39// Shared helpers
40// ---------------------------------------------------------------------------
41
42/// Generate `nproj` unit-norm random projection vectors of dimension `m`.
43///
44/// Returns a flat buffer of length `nproj * m` where projection `p` occupies
45/// `[p*m .. (p+1)*m]`.
46///
47/// If `seed` is Some, uses a deterministic RNG seeded from the given value.
48pub(super) fn generate_random_projections(nproj: usize, m: usize, seed: Option<u64>) -> Vec<f64> {
49    let mut rng: Box<dyn RngCore> = match seed {
50        Some(s) => Box::new(StdRng::seed_from_u64(s)),
51        None => Box::new(rand::thread_rng()),
52    };
53    let mut projections = vec![0.0; nproj * m];
54    for p_idx in 0..nproj {
55        let base = p_idx * m;
56        let mut norm_sq = 0.0;
57        for t in 0..m {
58            let v: f64 = rng.sample(StandardNormal);
59            projections[base + t] = v;
60            norm_sq += v * v;
61        }
62        let inv_norm = 1.0 / norm_sq.sqrt();
63        for t in 0..m {
64            projections[base + t] *= inv_norm;
65        }
66    }
67    projections
68}
69
70/// Project each reference curve onto each projection direction and sort.
71///
72/// Returns a flat buffer of length `nproj * nori` where the sorted projections
73/// for direction `p` occupy `[p*nori .. (p+1)*nori]`.
74pub(super) fn project_and_sort_reference(
75    data_ori: &FdMatrix,
76    projections: &[f64],
77    nproj: usize,
78    nori: usize,
79    m: usize,
80) -> Vec<f64> {
81    let mut sorted = vec![0.0; nproj * nori];
82    maybe_par_chunks_mut_enumerate!(sorted, nori, |(p_idx, spo): (usize, &mut [f64])| {
83        let proj = &projections[p_idx * m..(p_idx + 1) * m];
84        for j in 0..nori {
85            let mut dot = 0.0;
86            for t in 0..m {
87                dot += data_ori[(j, t)] * proj[t];
88            }
89            spo[j] = dot;
90        }
91        spo.sort_unstable_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));
92    });
93    sorted
94}
95
96/// Shared implementation for random projection-based depth measures.
97///
98/// Generates `nproj` random projections, projects both object and reference
99/// curves to scalars, and computes univariate depth via binary-search ranking.
100/// The `aggregate` and `finalize` closures control how per-projection depths
101/// are combined (e.g. average for RP depth, minimum for Tukey depth).
102pub(super) fn random_depth_core(
103    data_obj: &FdMatrix,
104    data_ori: &FdMatrix,
105    nproj: usize,
106    seed: Option<u64>,
107    init: f64,
108    aggregate: impl Fn(f64, f64) -> f64 + Sync,
109    finalize: impl Fn(f64, usize) -> f64 + Sync,
110) -> Vec<f64> {
111    use crate::iter_maybe_parallel;
112    #[cfg(feature = "parallel")]
113    use rayon::iter::ParallelIterator;
114
115    let nobj = data_obj.nrows();
116    let nori = data_ori.nrows();
117    let m = data_obj.ncols();
118
119    if nobj == 0 || nori == 0 || m == 0 || nproj == 0 {
120        return Vec::new();
121    }
122
123    let projections = generate_random_projections(nproj, m, seed);
124    let sorted_proj_ori = project_and_sort_reference(data_ori, &projections, nproj, nori, m);
125    let denom = nori as f64 + 1.0;
126
127    iter_maybe_parallel!(0..nobj)
128        .map(|i| {
129            let mut acc = init;
130            for p_idx in 0..nproj {
131                let proj = &projections[p_idx * m..(p_idx + 1) * m];
132                let sorted_ori = &sorted_proj_ori[p_idx * nori..(p_idx + 1) * nori];
133
134                let mut proj_i = 0.0;
135                for t in 0..m {
136                    proj_i += data_obj[(i, t)] * proj[t];
137                }
138
139                let below = sorted_ori.partition_point(|&v| v < proj_i);
140                let above = nori - sorted_ori.partition_point(|&v| v <= proj_i);
141                let depth = (below.min(above) as f64 + 1.0) / denom;
142                acc = aggregate(acc, depth);
143            }
144            finalize(acc, nproj)
145        })
146        .collect()
147}
fdars_core/depth/mod.rs

fdars_core/depth/
mod.rs
