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

fdars_core/classification/
mod.rs
