sampling_demo/
sampling_demo.rs

1//! Sampling and bootstrapping utilities demonstration
2//!
3//! This example demonstrates the use of random sampling and stratified sampling
4//! utilities provided by scirs2-datasets.
5
6use ndarray::Array1;
7use scirs2_datasets::{load_iris, random_sample, stratified_sample, Dataset};
8
9#[allow(dead_code)]
10fn main() {
11    println!("=== Sampling and Bootstrapping Demonstration ===\n");
12
13    // Load the Iris dataset for demonstration
14    let iris = load_iris().unwrap();
15    let n_samples = iris.n_samples();
16
17    println!("Original Iris dataset:");
18    println!("- Samples: {n_samples}");
19    println!("- Features: {}", iris.n_features());
20
21    if let Some(target) = &iris.target {
22        let class_counts = count_classes(target);
23        println!("- Class distribution: {class_counts:?}\n");
24    }
25
26    // Demonstrate random sampling without replacement
27    println!("=== Random Sampling (without replacement) ===");
28    let samplesize = 30;
29    let random_indices = random_sample(n_samples, samplesize, false, Some(42)).unwrap();
30
31    println!("Sampled {samplesize} indices from {n_samples} total samples");
32    println!(
33        "Sample indices: {:?}",
34        &random_indices[..10.min(random_indices.len())]
35    );
36
37    // Create a subset dataset
38    let sampledata = iris.data.select(ndarray::Axis(0), &random_indices);
39    let sample_target = iris
40        .target
41        .as_ref()
42        .map(|t| t.select(ndarray::Axis(0), &random_indices));
43    let sampledataset = Dataset::new(sampledata, sample_target)
44        .with_description("Random sample from Iris dataset".to_string());
45
46    println!(
47        "Random sample dataset: {} samples, {} features",
48        sampledataset.n_samples(),
49        sampledataset.n_features()
50    );
51
52    if let Some(target) = &sampledataset.target {
53        let sample_class_counts = count_classes(target);
54        println!("Sample class distribution: {sample_class_counts:?}\n");
55    }
56
57    // Demonstrate bootstrap sampling (with replacement)
58    println!("=== Bootstrap Sampling (with replacement) ===");
59    let bootstrapsize = 200; // More than original dataset size
60    let bootstrap_indices = random_sample(n_samples, bootstrapsize, true, Some(42)).unwrap();
61
62    println!("Bootstrap sampled {bootstrapsize} indices from {n_samples} total samples");
63    println!(
64        "Bootstrap may have duplicates - first 10 indices: {:?}",
65        &bootstrap_indices[..10]
66    );
67
68    // Count frequency of each index in bootstrap sample
69    let mut index_counts = vec![0; n_samples];
70    for &idx in &bootstrap_indices {
71        index_counts[idx] += 1;
72    }
73    let max_count = *index_counts.iter().max().unwrap();
74    let zero_count = index_counts.iter().filter(|&&count| count == 0).count();
75
76    println!("Bootstrap statistics:");
77    println!("- Maximum frequency of any sample: {max_count}");
78    println!("- Number of original samples not selected: {zero_count}\n");
79
80    // Demonstrate stratified sampling
81    println!("=== Stratified Sampling ===");
82    if let Some(target) = &iris.target {
83        let stratifiedsize = 30;
84        let stratified_indices = stratified_sample(target, stratifiedsize, Some(42)).unwrap();
85
86        println!("Stratified sampled {stratifiedsize} indices maintaining class proportions");
87
88        // Create stratified subset
89        let stratifieddata = iris.data.select(ndarray::Axis(0), &stratified_indices);
90        let stratified_target = target.select(ndarray::Axis(0), &stratified_indices);
91        let stratifieddataset = Dataset::new(stratifieddata, Some(stratified_target))
92            .with_description("Stratified sample from Iris dataset".to_string());
93
94        println!(
95            "Stratified sample dataset: {} samples, {} features",
96            stratifieddataset.n_samples(),
97            stratifieddataset.n_features()
98        );
99
100        let stratified_class_counts = count_classes(&stratifieddataset.target.unwrap());
101        println!("Stratified sample class distribution: {stratified_class_counts:?}");
102
103        // Verify proportions are maintained
104        let original_proportions = calculate_proportions(&count_classes(target));
105        let stratified_proportions = calculate_proportions(&stratified_class_counts);
106
107        println!("Class proportion comparison:");
108        for (&class, &original_prop) in &original_proportions {
109            let stratified_prop = stratified_proportions.get(&class).unwrap_or(&0.0);
110            println!(
111                "  Class {}: Original {:.2}%, Stratified {:.2}%",
112                class,
113                original_prop * 100.0,
114                stratified_prop * 100.0
115            );
116        }
117    }
118
119    // Demonstrate practical use case: creating training/validation splits
120    println!("\n=== Practical Example: Multiple Train/Validation Splits ===");
121    for i in 1..=3 {
122        let split_indices = random_sample(n_samples, 100, false, Some(42 + i)).unwrap();
123        let (train_indices, val_indices) = split_indices.split_at(80);
124
125        println!(
126            "Split {}: {} training samples, {} validation samples",
127            i,
128            train_indices.len(),
129            val_indices.len()
130        );
131    }
132
133    println!("\n=== Sampling Demo Complete ===");
134}
135
136/// Count the number of samples in each class
137#[allow(dead_code)]
138fn count_classes(targets: &Array1<f64>) -> std::collections::HashMap<i64, usize> {
139    let mut counts = std::collections::HashMap::new();
140    for &target in targets.iter() {
141        let class = target.round() as i64;
142        *counts.entry(class).or_insert(0) += 1;
143    }
144    counts
145}
146
147/// Calculate class proportions
148#[allow(dead_code)]
149fn calculate_proportions(
150    counts: &std::collections::HashMap<i64, usize>,
151) -> std::collections::HashMap<i64, f64> {
152    let total: usize = counts.values().sum();
153    counts
154        .iter()
155        .map(|(&class, &count)| (class, count as f64 / total as f64))
156        .collect()
157}
sampling_demo/sampling_demo.rs

sampling_demo/
sampling_demo.rs
