fdars_core/classification/
mod.rs

1//! Functional classification with mixed scalar/functional predictors.
2//!
3//! Implements supervised classification for functional data using:
4//! - [`fclassif_lda`] / [`fclassif_qda`] — FPC + LDA/QDA pipeline
5//! - [`fclassif_knn`] — FPC + k-NN classifier
6//! - [`fclassif_kernel`] — Nonparametric kernel classifier with mixed predictors
7//! - [`fclassif_dd`] — Depth-based DD-classifier
8//! - [`fclassif_cv`] — Cross-validated error rate
9//!
10//! ## Parameter ordering convention
11//! All classifiers: `(data, y, [argvals,] [scalar_covariates,] method-params)`
12
13use crate::error::FdarError;
14use crate::matrix::FdMatrix;
15use crate::regression::fdata_to_pc_1d;
16
17pub mod cv;
18pub mod dd;
19pub mod fit;
20pub mod kernel;
21pub mod knn;
22pub mod lda;
23pub mod qda;
24
25#[cfg(test)]
26mod tests;
27
28// ---------------------------------------------------------------------------
29// Shared types
30// ---------------------------------------------------------------------------
31
32/// Classification result.
33#[derive(Debug, Clone, PartialEq)]
34#[non_exhaustive]
35pub struct ClassifResult {
36    /// Predicted class labels (length n)
37    pub predicted: Vec<usize>,
38    /// Posterior/membership probabilities (n x G) — if available
39    pub probabilities: Option<FdMatrix>,
40    /// Training accuracy
41    pub accuracy: f64,
42    /// Confusion matrix (G x G): row = true, col = predicted
43    pub confusion: Vec<Vec<usize>>,
44    /// Number of classes
45    pub n_classes: usize,
46    /// Number of FPC components used
47    pub ncomp: usize,
48}
49
50/// Cross-validation result.
51#[derive(Debug, Clone, PartialEq)]
52#[non_exhaustive]
53pub struct ClassifCvResult {
54    /// Mean error rate across folds
55    pub error_rate: f64,
56    /// Per-fold error rates
57    pub fold_errors: Vec<f64>,
58    /// Best ncomp (if tuned)
59    pub best_ncomp: usize,
60}
61
62// ---------------------------------------------------------------------------
63// Utility helpers
64// ---------------------------------------------------------------------------
65
66/// Count distinct classes and remap labels to 0..G-1.
67pub(crate) fn remap_labels(y: &[usize]) -> (Vec<usize>, usize) {
68    let mut labels: Vec<usize> = y.to_vec();
69    let mut unique: Vec<usize> = y.to_vec();
70    unique.sort_unstable();
71    unique.dedup();
72    let g = unique.len();
73    for label in &mut labels {
74        *label = unique.iter().position(|&u| u == *label).unwrap_or(0);
75    }
76    (labels, g)
77}
78
79/// Build confusion matrix (G x G).
80fn confusion_matrix(true_labels: &[usize], pred_labels: &[usize], g: usize) -> Vec<Vec<usize>> {
81    let mut cm = vec![vec![0usize; g]; g];
82    for (&t, &p) in true_labels.iter().zip(pred_labels.iter()) {
83        if t < g && p < g {
84            cm[t][p] += 1;
85        }
86    }
87    cm
88}
89
90/// Compute per-class means, counts, and priors from labeled features.
91pub(crate) fn class_means_and_priors(
92    features: &FdMatrix,
93    labels: &[usize],
94    g: usize,
95) -> (Vec<Vec<f64>>, Vec<usize>, Vec<f64>) {
96    let n = features.nrows();
97    let d = features.ncols();
98    let mut counts = vec![0usize; g];
99    let mut class_means = vec![vec![0.0; d]; g];
100    for i in 0..n {
101        let c = labels[i];
102        counts[c] += 1;
103        for j in 0..d {
104            class_means[c][j] += features[(i, j)];
105        }
106    }
107    for c in 0..g {
108        if counts[c] > 0 {
109            for j in 0..d {
110                class_means[c][j] /= counts[c] as f64;
111            }
112        }
113    }
114    let priors: Vec<f64> = counts.iter().map(|&c| c as f64 / n as f64).collect();
115    (class_means, counts, priors)
116}
117
118/// Accuracy from labels.
119fn compute_accuracy(true_labels: &[usize], pred_labels: &[usize]) -> f64 {
120    let n = true_labels.len();
121    if n == 0 {
122        return 0.0;
123    }
124    let correct = true_labels
125        .iter()
126        .zip(pred_labels.iter())
127        .filter(|(&t, &p)| t == p)
128        .count();
129    correct as f64 / n as f64
130}
131
132/// Extract FPC scores and append optional scalar covariates.
133pub(crate) fn build_feature_matrix(
134    data: &FdMatrix,
135    scalar_covariates: Option<&FdMatrix>,
136    ncomp: usize,
137) -> Result<(FdMatrix, Vec<f64>, FdMatrix), FdarError> {
138    let fpca = fdata_to_pc_1d(data, ncomp)?;
139    let n = data.nrows();
140    let d_pc = fpca.scores.ncols();
141    let d_cov = scalar_covariates.map_or(0, super::matrix::FdMatrix::ncols);
142    let d = d_pc + d_cov;
143
144    let mut features = FdMatrix::zeros(n, d);
145    for i in 0..n {
146        for j in 0..d_pc {
147            features[(i, j)] = fpca.scores[(i, j)];
148        }
149        if let Some(cov) = scalar_covariates {
150            for j in 0..d_cov {
151                features[(i, d_pc + j)] = cov[(i, j)];
152            }
153        }
154    }
155
156    Ok((features, fpca.mean, fpca.rotation))
157}
158
159// ---------------------------------------------------------------------------
160// Re-exports — preserves the external API
161// ---------------------------------------------------------------------------
162
163pub use cv::fclassif_cv;
164pub use dd::fclassif_dd;
165pub(crate) use fit::classif_predict_probs;
166pub use fit::{
167    fclassif_cv_with_config, fclassif_knn_fit, fclassif_lda_fit, fclassif_qda_fit, ClassifCvConfig,
168    ClassifFit, ClassifMethod,
169};
170pub use kernel::fclassif_kernel;
171pub use knn::fclassif_knn;
172pub use lda::fclassif_lda;
173pub use qda::fclassif_qda;
fdars_core/classification/mod.rs

fdars_core/classification/
mod.rs
