rustyml 0.13.0

A high-performance machine learning & deep learning library in pure Rust, offering ML algorithms and neural network support
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221

//! Train/test split utilities
//!
//! Provides the [`TrainTestSplit`] type alias, the [`train_test_split()`] function for a plain
//! random partition, and [`train_test_split_stratified()`] for a partition that preserves each
//! class proportion across the two subsets

use crate::error::Error;
use ahash::AHashMap;
use ndarray::{Array1, Array2, Axis};
use ndarray_rand::rand::seq::SliceRandom;
use std::hash::Hash;

/// The four arrays produced by [`train_test_split`], in order:
/// `(x_train, x_test, y_train, y_test)`
pub type TrainTestSplit<A> = (Array2<f64>, Array2<f64>, Array1<A>, Array1<A>);

/// Splits a dataset into training and test sets
///
/// # Parameters
///
/// - `x` - Feature matrix with shape (n_samples, n_features)
/// - `y` - Target values with shape (n_samples); the label type `A` is generic, so integer,
///   float, or string labels all work (only `Clone` is required)
/// - `test_size` - Size of the test set, default is 0.3 (30%)
/// - `random_state` - Random seed, default is None
///
/// # Type Parameters
///
/// - `A` - The label element type; must be `Clone` so rows can be gathered into the output arrays
///
/// # Returns
///
/// - `Result<TrainTestSplit<A>, Error>` - `(x_train, x_test, y_train, y_test)` on success
///
/// # Errors
///
/// - [`Error::EmptyInput`] if the dataset is empty
/// - [`Error::DimensionMismatch`] if `x` and `y` have different lengths
/// - [`Error::InvalidParameter`] if `test_size` is not between 0 and 1
/// - [`Error::InvalidInput`] if the dataset is too small to split
///
/// # Examples
///
/// ```rust
/// use ndarray::{Array1, Array2};
/// use rustyml::utils::train_test_split::train_test_split;
///
/// let x = Array2::from_shape_vec((5, 2), vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0]).unwrap();
/// // Integer class labels work just as well as floats
/// let y = Array1::from(vec![0, 1, 0, 1, 0]);
/// let (x_train, x_test, y_train, y_test) = train_test_split(x, y, Some(0.4), Some(42)).unwrap();
/// ```
pub fn train_test_split<A>(
    x: Array2<f64>,
    y: Array1<A>,
    test_size: Option<f64>,
    random_state: Option<u64>,
) -> Result<TrainTestSplit<A>, Error>
where
    A: Clone,
{
    let n_samples = x.nrows();

    if n_samples == 0 {
        return Err(Error::empty_input("dataset"));
    }

    if n_samples != y.len() {
        return Err(Error::dimension_mismatch(n_samples, y.len()));
    }

    let test_size = test_size.unwrap_or(0.3);
    if test_size <= 0.0 || test_size >= 1.0 {
        return Err(Error::invalid_parameter(
            "test_size",
            format!(
                "test_size must be between 0 and 1 (exclusive), got {}",
                test_size
            ),
        ));
    }

    // Clamp so both train and test always keep at least one sample
    let n_test = if n_samples == 1 {
        return Err(Error::invalid_input(
            "Cannot split a dataset with only 1 sample into train and test sets",
        ));
    } else if n_samples == 2 {
        1 // with 2 samples, always put 1 in test regardless of test_size
    } else {
        let calculated = (n_samples as f64 * test_size).round() as usize;
        calculated.max(1).min(n_samples - 1)
    };

    let mut indices: Vec<usize> = (0..n_samples).collect();

    let mut rng = crate::random::make_rng(random_state);
    indices.shuffle(&mut rng);

    let (test_indices, train_indices) = indices.split_at(n_test);

    // select gathers rows by index in one pass
    let x_train = x.select(Axis(0), train_indices);
    let x_test = x.select(Axis(0), test_indices);
    let y_train = y.select(Axis(0), train_indices);
    let y_test = y.select(Axis(0), test_indices);

    Ok((x_train, x_test, y_train, y_test))
}

/// Splits a dataset into training and test sets while preserving class proportions
///
/// Each class is split independently using `test_size`, so both subsets keep roughly the same
/// label distribution as the input. This avoids a class being absent from one side, which a plain
/// [`train_test_split`] can produce on imbalanced data
///
/// # Parameters
///
/// - `x` - Feature matrix with shape (n_samples, n_features)
/// - `y` - Class labels with shape (n_samples); the label type `A` must be hashable so samples can
///   be grouped by class
/// - `test_size` - Fraction of each class placed in the test set, default is 0.3 (30%)
/// - `random_state` - Random seed, default is None
///
/// # Type Parameters
///
/// - `A` - The label element type; must be `Clone + Eq + Hash` to group rows by class
///
/// # Returns
///
/// - `Result<TrainTestSplit<A>, Error>` - `(x_train, x_test, y_train, y_test)` on success
///
/// # Errors
///
/// - [`Error::EmptyInput`] if the dataset is empty
/// - [`Error::DimensionMismatch`] if `x` and `y` have different lengths
/// - [`Error::InvalidParameter`] if `test_size` is not between 0 and 1
/// - [`Error::InvalidInput`] if any class has fewer than 2 samples
///
/// # Examples
///
/// ```rust
/// use ndarray::{Array1, Array2};
/// use rustyml::utils::train_test_split::train_test_split_stratified;
///
/// let x = Array2::from_shape_vec((6, 1), vec![0.0, 1.0, 2.0, 3.0, 4.0, 5.0]).unwrap();
/// // 3 samples per class, so each side keeps both classes
/// let y = Array1::from(vec![0, 0, 0, 1, 1, 1]);
/// let (x_train, x_test, y_train, y_test) =
///     train_test_split_stratified(x, y, Some(0.34), Some(42)).unwrap();
/// ```
pub fn train_test_split_stratified<A>(
    x: Array2<f64>,
    y: Array1<A>,
    test_size: Option<f64>,
    random_state: Option<u64>,
) -> Result<TrainTestSplit<A>, Error>
where
    A: Clone + Eq + Hash,
{
    let n_samples = x.nrows();

    if n_samples == 0 {
        return Err(Error::empty_input("dataset"));
    }

    if n_samples != y.len() {
        return Err(Error::dimension_mismatch(n_samples, y.len()));
    }

    let test_size = test_size.unwrap_or(0.3);
    if test_size <= 0.0 || test_size >= 1.0 {
        return Err(Error::invalid_parameter(
            "test_size",
            format!(
                "test_size must be between 0 and 1 (exclusive), got {}",
                test_size
            ),
        ));
    }

    // Group sample indices by class label, keeping first-appearance order so the result is
    // deterministic for a given seed
    let mut group_of: AHashMap<A, usize> = AHashMap::new();
    let mut groups: Vec<Vec<usize>> = Vec::new();
    for (idx, label) in y.iter().enumerate() {
        match group_of.get(label) {
            Some(&g) => groups[g].push(idx),
            None => {
                group_of.insert(label.clone(), groups.len());
                groups.push(vec![idx]);
            }
        }
    }

    let mut rng = crate::random::make_rng(random_state);

    // Split every class on its own, so both sides keep at least one sample of each class
    let mut test_indices: Vec<usize> = Vec::new();
    let mut train_indices: Vec<usize> = Vec::new();
    for group in groups.iter_mut() {
        let class_size = group.len();
        if class_size < 2 {
            return Err(Error::invalid_input(
                "Stratified split requires at least 2 samples per class",
            ));
        }
        group.shuffle(&mut rng);
        let n_test = ((class_size as f64 * test_size).round() as usize).clamp(1, class_size - 1);
        let (test_part, train_part) = group.split_at(n_test);
        test_indices.extend_from_slice(test_part);
        train_indices.extend_from_slice(train_part);
    }

    let x_train = x.select(Axis(0), &train_indices);
    let x_test = x.select(Axis(0), &test_indices);
    let y_train = y.select(Axis(0), &train_indices);
    let y_test = y.select(Axis(0), &test_indices);

    Ok((x_train, x_test, y_train, y_test))
}