rill-ml 0.8.0

Lightweight, serializable online machine learning for Rust applications and streaming data.
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

//! Integration tests: pipeline behavior (transformer + model composition).
//!
//! Verifies that pipelines correctly chain preprocessing and model learning,
//! that transformers do not leak labels, and that dimension mismatches are
//! rejected at the pipeline level.

use rand::SeedableRng;
use rill_ml::OnlineRegressor;
use rill_ml::loss::RegressionLoss;
use rill_ml::models::{LinearRegression, LinearRegressionConfig};
use rill_ml::optim::{Optimizer, SgdConfig};
use rill_ml::pipeline::{ClassificationPipeline, RegressionPipeline};
use rill_ml::preprocessing::{Clipper, MinMaxScaler, StandardScaler};
use rill_ml::traits::{OnlineBinaryClassifier, Transformer};

fn make_regression_pipeline(d: usize) -> RegressionPipeline<StandardScaler, LinearRegression> {
    let scaler = StandardScaler::new(d).unwrap();
    let model = LinearRegression::new(
        d,
        LinearRegressionConfig {
            optimizer: Optimizer::sgd(d, SgdConfig::default()).unwrap(),
            loss: RegressionLoss::default(),
        },
    )
    .unwrap();
    RegressionPipeline::new(scaler, model).unwrap()
}

#[test]
fn pipeline_learns_linear_relation() {
    let d = 2;
    let mut pipeline = make_regression_pipeline(d);
    let mut rng = rand_chacha::ChaCha8Rng::seed_from_u64(55);

    for _ in 0..1000 {
        let x1 = rand::Rng::gen_range(&mut rng, 0.0..10.0);
        let x2 = rand::Rng::gen_range(&mut rng, 0.0..10.0);
        let y = 3.0 * x1 - 2.0 * x2 + 1.0;
        pipeline.learn(&[x1, x2], y).unwrap();
    }

    let pred = pipeline.predict(&[5.0, 2.5]).unwrap();
    let expected = 3.0 * 5.0 - 2.0 * 2.5 + 1.0;
    assert!(
        (pred - expected).abs() < 2.0,
        "pred = {pred}, expected = {expected}"
    );
}

#[test]
fn pipeline_predict_is_side_effect_free() {
    let mut pipeline = make_regression_pipeline(2);
    pipeline.learn(&[1.0, 2.0], 3.0).unwrap();
    let samples_before = pipeline.samples_seen();
    let _ = pipeline.predict(&[1.0, 2.0]).unwrap();
    let _ = pipeline.predict(&[3.0, 4.0]).unwrap();
    assert_eq!(pipeline.samples_seen(), samples_before);
}

#[test]
fn pipeline_transformer_does_not_see_target() {
    let d = 2;
    let scaler = StandardScaler::new(d).unwrap();
    let model = LinearRegression::new(
        d,
        LinearRegressionConfig {
            optimizer: Optimizer::sgd(d, SgdConfig::default()).unwrap(),
            loss: RegressionLoss::default(),
        },
    )
    .unwrap();
    let mut pipeline = RegressionPipeline::new(scaler, model).unwrap();

    let x = &[1.0, 2.0];
    let y_target = 100.0;
    pipeline.learn(x, y_target).unwrap();

    let scaler_after = pipeline.transformer();
    assert_eq!(scaler_after.samples_seen(), 1);
}

#[test]
fn pipeline_rejects_dimension_mismatch() {
    let mut pipeline = make_regression_pipeline(3);
    assert!(pipeline.predict(&[1.0, 2.0]).is_err());
    assert!(pipeline.learn(&[1.0, 2.0], 1.0).is_err());
}

#[test]
fn pipeline_with_minmax_scaler_works() {
    let d = 2;
    let scaler = MinMaxScaler::new(d).unwrap();
    let model = LinearRegression::new(
        d,
        LinearRegressionConfig {
            optimizer: Optimizer::sgd(
                d,
                SgdConfig {
                    learning_rate: 0.1,
                    l2: 0.0,
                },
            )
            .unwrap(),
            loss: RegressionLoss::default(),
        },
    )
    .unwrap();
    let mut pipeline = RegressionPipeline::new(scaler, model).unwrap();

    let mut rng = rand_chacha::ChaCha8Rng::seed_from_u64(88);
    for _ in 0..2000 {
        let x1 = rand::Rng::gen_range(&mut rng, 0.0..10.0);
        let x2 = rand::Rng::gen_range(&mut rng, 0.0..10.0);
        let y = 2.0 * x1 + 3.0 * x2;
        pipeline.learn(&[x1, x2], y).unwrap();
    }

    let pred = pipeline.predict(&[5.0, 5.0]).unwrap();
    let expected = 2.0 * 5.0 + 3.0 * 5.0;
    assert!(
        (pred - expected).abs() < 3.0,
        "pred = {pred}, expected = {expected}"
    );
}

#[test]
fn classification_pipeline_learns() {
    use rill_ml::models::{LogisticRegression, LogisticRegressionConfig};
    let d = 2;
    let scaler = StandardScaler::new(d).unwrap();
    let model = LogisticRegression::new(
        d,
        LogisticRegressionConfig {
            optimizer: Optimizer::sgd(
                d,
                SgdConfig {
                    learning_rate: 0.1,
                    l2: 0.0,
                },
            )
            .unwrap(),
            loss: Default::default(),
        },
    )
    .unwrap();
    let mut pipeline = ClassificationPipeline::new(scaler, model).unwrap();

    let mut rng = rand_chacha::ChaCha8Rng::seed_from_u64(33);
    for _ in 0..1000 {
        let x1 = rand::Rng::gen_range(&mut rng, -2.0..2.0);
        let x2 = rand::Rng::gen_range(&mut rng, -2.0..2.0);
        let y = x1 > x2;
        pipeline.learn(&[x1, x2], y).unwrap();
    }

    let p_pos = pipeline.predict_proba(&[2.0, -1.0]).unwrap();
    let p_neg = pipeline.predict_proba(&[-1.0, 2.0]).unwrap();
    assert!(p_pos > 0.7, "p_pos = {p_pos}");
    assert!(p_neg < 0.3, "p_neg = {p_neg}");
}

#[test]
fn pipeline_reset_clears_state() {
    let mut pipeline = make_regression_pipeline(2);
    pipeline.learn(&[1.0, 2.0], 3.0).unwrap();
    pipeline.learn(&[4.0, 5.0], 6.0).unwrap();
    assert_eq!(pipeline.samples_seen(), 2);
    pipeline.reset();
    assert_eq!(pipeline.samples_seen(), 0);
}

#[test]
fn clipper_in_pipeline_clamps_values() {
    let clipper = Clipper::new(1, -1.0, 1.0).unwrap();
    let model = LinearRegression::new(
        1,
        LinearRegressionConfig {
            optimizer: Optimizer::sgd(1, SgdConfig::default()).unwrap(),
            loss: RegressionLoss::default(),
        },
    )
    .unwrap();
    let mut pipeline = RegressionPipeline::new(clipper, model).unwrap();

    // Learning with extreme values should be clipped by the clipper.
    for _ in 0..100 {
        pipeline.learn(&[100.0], 1.0).unwrap();
    }

    // The clipper is stateless, so samples_seen() is always 0.
    // Verify that predictions are produced successfully.
    let pred = pipeline.predict(&[100.0]).unwrap();
    assert!(pred.is_finite());
}