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
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281

//! Integration tests for `rustyml::utils::standardize`, covering Global, Column,
//! and Row z-score standardization plus the error paths

use crate::common::assert_allclose;
use ndarray::{Array1, Array2, Array3, array};
use rustyml::error::Error;
use rustyml::utils::standardize::{StandardizationAxis, standardize};

// Global axis

/// Global standardization of a 1-D array yields population z-scores
#[test]
fn test_global_closed_form_1d() {
    let data: Array1<f64> = array![1.0, 3.0];
    let result = standardize(&data, StandardizationAxis::Global).unwrap();
    let expected: Array1<f64> = array![-1.0, 1.0];
    assert_allclose(&result, &expected, 1e-7);
}

/// Global standardization of a 1x2 matrix preserves shape
#[test]
fn test_global_closed_form_2d() {
    let data: Array2<f64> = array![[1.0, 3.0]];
    let result = standardize(&data, StandardizationAxis::Global).unwrap();
    let expected: Array2<f64> = array![[-1.0, 1.0]];
    assert_allclose(&result, &expected, 1e-7);
}

/// A healthy feature is divided by the raw sqrt(variance), giving exact z-scores with no
/// stability bias, since a healthy feature is never perturbed
#[test]
fn test_healthy_feature_exact_zscores() {
    let data: Array1<f64> = array![1.0, 3.0];
    // variance = 1.0, so the divisor is exactly 1.0 and the z-scores are exactly +/-1.0
    let result = standardize(&data, StandardizationAxis::Global).unwrap();
    let expected: Array1<f64> = array![-1.0, 1.0];
    assert_allclose(&result, &expected, 1e-12);
}

/// Constant detection is machine-precision and variance-based, so a tiny but real
/// variance is still normalized
#[test]
fn test_small_but_real_variance_is_normalized_like_sklearn() {
    let data: Array1<f64> = array![1.0, 1.0 + 1e-8];
    let result = standardize(&data, StandardizationAxis::Global).unwrap();
    let expected: Array1<f64> = array![-1.0, 1.0];
    assert_allclose(&result, &expected, 1e-4);
}

/// All-identical values give zero variance and all-zero output without NaN
#[test]
fn test_global_zero_variance_all_zeros() {
    let data: Array2<f64> = array![[3.0, 3.0], [3.0, 3.0]];
    let result = standardize(&data, StandardizationAxis::Global).unwrap();
    let expected: Array2<f64> = array![[0.0, 0.0], [0.0, 0.0]];
    assert_allclose(&result, &expected, 1e-12);
    assert!(result.iter().all(|v| v.is_finite()));
}

/// Single-element array standardizes to zero (variance = 0)
#[test]
fn test_global_single_element() {
    let data: Array2<f64> = array![[5.0]];
    let result = standardize(&data, StandardizationAxis::Global).unwrap();
    let expected: Array2<f64> = array![[0.0]];
    assert_allclose(&result, &expected, 1e-12);
}

/// Global standardization of [1,2,3,4,5] matches exact sqrt(2)-based z-scores
#[test]
fn test_global_1d_five_elements() {
    let data: Array1<f64> = array![1.0, 2.0, 3.0, 4.0, 5.0];
    let result = standardize(&data, StandardizationAxis::Global).unwrap();
    use std::f64::consts::{FRAC_1_SQRT_2, SQRT_2};
    let expected: Array1<f64> = array![-SQRT_2, -FRAC_1_SQRT_2, 0.0, FRAC_1_SQRT_2, SQRT_2];
    assert_allclose(&result, &expected, 1e-8);
}

/// Global standardization flattens a 3-D array and preserves its shape
#[test]
fn test_global_3d_array() {
    let data: Array3<f64> = array![[[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]]];
    let result = standardize(&data, StandardizationAxis::Global).unwrap();
    let expected: Array3<f64> = array![[
        [-1.4638501094, -0.8783100656],
        [-0.2927700219, 0.2927700219],
        [0.8783100656, 1.4638501094]
    ]];
    assert_allclose(&result, &expected, 1e-7);
    assert_eq!(result.shape(), data.shape());
}

// Column axis

/// Column standardization computes per-column population z-scores
#[test]
fn test_column_axis_closed_form() {
    let data: Array2<f64> = array![[1.0, 4.0], [2.0, 5.0], [3.0, 6.0]];
    let result = standardize(&data, StandardizationAxis::Column).unwrap();
    let expected: Array2<f64> = array![
        [-1.2247448714, -1.2247448714],
        [0.0, 0.0],
        [1.2247448714, 1.2247448714]
    ];
    assert_allclose(&result, &expected, 1e-7);
}

/// Each standardized column has mean ~= 0 and population std ~= 1
#[test]
fn test_column_axis_invariants() {
    let data: Array2<f64> = array![[1.0, 4.0], [2.0, 5.0], [3.0, 6.0]];
    let result = standardize(&data, StandardizationAxis::Column).unwrap();
    let n = result.nrows() as f64;
    for col in result.columns() {
        let mean: f64 = col.sum() / n;
        let variance: f64 = col.iter().map(|&x| (x - mean).powi(2)).sum::<f64>() / n;
        assert!(mean.abs() < 1e-10, "column mean {mean} not near 0");
        assert!(
            (variance - 1.0).abs() < 1e-7,
            "column variance {variance} not near 1"
        );
    }
}

/// A zero-variance column standardizes to all zeros while other columns are unaffected
#[test]
fn test_column_axis_zero_variance_column() {
    let data: Array2<f64> = array![[3.0, 1.0], [3.0, 3.0], [3.0, 5.0]];
    let result = standardize(&data, StandardizationAxis::Column).unwrap();
    assert!(
        result.iter().all(|v| v.is_finite()),
        "no NaN or Inf expected"
    );
    let col0: Vec<f64> = result.column(0).to_vec();
    for v in &col0 {
        assert!(
            v.abs() < 1e-6,
            "zero-variance column value should be 0, got {v}"
        );
    }
    let expected_col1: Array1<f64> = array![-1.2247448714, 0.0, 1.2247448714];
    let actual_col1: Array1<f64> = result.column(1).to_owned();
    assert_allclose(&actual_col1, &expected_col1, 1e-7);
}

// Row axis

/// Row standardization computes per-row population z-scores
#[test]
fn test_row_axis_closed_form() {
    let data: Array2<f64> = array![[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]];
    let result = standardize(&data, StandardizationAxis::Row).unwrap();
    let expected: Array2<f64> = array![
        [-1.2247448714, 0.0, 1.2247448714],
        [-1.2247448714, 0.0, 1.2247448714]
    ];
    assert_allclose(&result, &expected, 1e-7);
}

/// Each standardized row has mean ~= 0 and population std ~= 1
#[test]
fn test_row_axis_invariants() {
    let data: Array2<f64> = array![[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]];
    let result = standardize(&data, StandardizationAxis::Row).unwrap();
    let n = result.ncols() as f64;
    for row in result.rows() {
        let mean: f64 = row.sum() / n;
        let variance: f64 = row.iter().map(|&x| (x - mean).powi(2)).sum::<f64>() / n;
        assert!(mean.abs() < 1e-10, "row mean {mean} not near 0");
        assert!(
            (variance - 1.0).abs() < 1e-7,
            "row variance {variance} not near 1"
        );
    }
}

/// A zero-variance row standardizes to all zeros while other rows are unaffected
#[test]
fn test_row_axis_zero_variance_row() {
    let data: Array2<f64> = array![[3.0, 3.0, 3.0], [1.0, 3.0, 5.0]];
    let result = standardize(&data, StandardizationAxis::Row).unwrap();
    assert!(
        result.iter().all(|v| v.is_finite()),
        "no NaN or Inf expected"
    );
    let row0: Array1<f64> = result.row(0).to_owned();
    let expected_row0: Array1<f64> = array![0.0, 0.0, 0.0];
    assert_allclose(&row0, &expected_row0, 1e-6);
    let row1: Array1<f64> = result.row(1).to_owned();
    let expected_row1: Array1<f64> = array![-1.2247448714, 0.0, 1.2247448714];
    assert_allclose(&row1, &expected_row1, 1e-7);
}

// Immutability

/// standardize does not mutate its input array
#[test]
fn test_original_array_not_mutated() {
    let data: Array2<f64> = array![[1.0, 3.0], [5.0, 7.0]];
    let original = data.clone();
    let _result = standardize(&data, StandardizationAxis::Global).unwrap();
    assert_eq!(
        data, original,
        "standardize must not modify the input array"
    );
}

#[test]
fn test_original_array_not_mutated_column() {
    let data: Array2<f64> = array![[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]];
    let original = data.clone();
    let _result = standardize(&data, StandardizationAxis::Column).unwrap();
    assert_eq!(data, original);
}

// Error paths

/// Empty array returns EmptyInput
#[test]
fn test_error_empty_global() {
    let data: Array2<f64> = Array2::zeros((0, 0));
    let err = standardize(&data, StandardizationAxis::Global).unwrap_err();
    assert!(
        matches!(err, Error::EmptyInput(_)),
        "expected EmptyInput, got {err:?}"
    );
}

#[test]
fn test_error_empty_column() {
    let data: Array2<f64> = Array2::zeros((0, 3));
    let err = standardize(&data, StandardizationAxis::Column).unwrap_err();
    assert!(
        matches!(err, Error::EmptyInput(_)),
        "expected EmptyInput, got {err:?}"
    );
}

/// NaN in input returns NonFinite
#[test]
fn test_error_nan_input() {
    let data: Array2<f64> = array![[1.0, f64::NAN], [3.0, 4.0]];
    let err = standardize(&data, StandardizationAxis::Global).unwrap_err();
    assert!(
        matches!(err, Error::NonFinite(_)),
        "expected NonFinite, got {err:?}"
    );
}

/// Infinite value in input returns NonFinite
#[test]
fn test_error_inf_input() {
    let data: Array2<f64> = array![[1.0, f64::INFINITY], [3.0, 4.0]];
    let err = standardize(&data, StandardizationAxis::Global).unwrap_err();
    assert!(
        matches!(err, Error::NonFinite(_)),
        "expected NonFinite, got {err:?}"
    );
}

/// Row standardization on a 1-D array returns InvalidInput (requires >= 2 dims)
#[test]
fn test_error_row_on_1d_array() {
    let data: Array1<f64> = array![1.0, 2.0, 3.0];
    let err = standardize(&data, StandardizationAxis::Row).unwrap_err();
    assert!(
        matches!(err, Error::InvalidInput(_)),
        "expected InvalidInput for Row on 1-D array, got {err:?}"
    );
}

/// Column standardization on a 1-D array returns InvalidInput (requires >= 2 dims)
#[test]
fn test_error_column_on_1d_array() {
    let data: Array1<f64> = array![1.0, 2.0, 3.0];
    let err = standardize(&data, StandardizationAxis::Column).unwrap_err();
    assert!(
        matches!(err, Error::InvalidInput(_)),
        "expected InvalidInput for Column on 1-D array, got {err:?}"
    );
}