aprender-core 0.29.1

Next-generation machine learning library in pure Rust
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

pub(crate) use super::*;

#[test]
fn test_lf_output() {
    let abstain = LFOutput::Abstain;
    let label = LFOutput::Label(1);

    assert_eq!(abstain.to_i32(), -1);
    assert_eq!(label.to_i32(), 1);

    assert_eq!(LFOutput::from_i32(-1), LFOutput::Abstain);
    assert_eq!(LFOutput::from_i32(2), LFOutput::Label(2));
}

#[test]
fn test_label_model_creation() {
    let model = LabelModel::new(3, 5);
    assert_eq!(model.n_classes, 3);
    assert_eq!(model.n_lfs, 5);
    assert_eq!(model.class_priors.len(), 3);
}

#[test]
fn test_label_model_fit() {
    let mut model = LabelModel::new(2, 3);

    // Simple test: LFs agree on labels
    let lf_matrix = vec![
        vec![LFOutput::Label(0), LFOutput::Label(0), LFOutput::Abstain],
        vec![LFOutput::Label(1), LFOutput::Label(1), LFOutput::Label(1)],
        vec![LFOutput::Label(0), LFOutput::Abstain, LFOutput::Label(0)],
        vec![LFOutput::Abstain, LFOutput::Label(1), LFOutput::Label(1)],
    ];

    model.fit(&lf_matrix, 10, 0.5);

    // Should have learned something
    assert!(model.accuracies[0][0] > 0.0);
}

#[test]
fn test_label_model_predict() {
    let mut model = LabelModel::new(2, 2);
    model.accuracies = vec![vec![0.9, 0.9], vec![0.8, 0.8]];

    let lf_matrix = vec![
        vec![LFOutput::Label(0), LFOutput::Label(0)],
        vec![LFOutput::Label(1), LFOutput::Label(1)],
    ];

    let predictions = model.predict(&lf_matrix);

    assert_eq!(predictions.len(), 2);
    assert_eq!(predictions[0], 0);
    assert_eq!(predictions[1], 1);
}

#[test]
fn test_label_model_proba() {
    let mut model = LabelModel::new(2, 2);
    model.accuracies = vec![vec![0.9, 0.9], vec![0.8, 0.8]];

    let lf_matrix = vec![vec![LFOutput::Label(0), LFOutput::Label(0)]];

    let probs = model.predict_proba(&lf_matrix);

    assert_eq!(probs.len(), 1);
    assert_eq!(probs[0].len(), 2);

    // Sum should be 1
    let sum: f32 = probs[0].as_slice().iter().sum();
    assert!((sum - 1.0).abs() < 1e-5);

    // Class 0 should be more likely
    assert!(probs[0][0] > probs[0][1]);
}

#[test]
fn test_label_model_coverage() {
    let model = LabelModel::new(2, 3);

    let lf_matrix = vec![
        vec![LFOutput::Label(0), LFOutput::Label(0), LFOutput::Abstain],
        vec![LFOutput::Label(1), LFOutput::Abstain, LFOutput::Abstain],
        vec![LFOutput::Abstain, LFOutput::Label(1), LFOutput::Label(1)],
        vec![LFOutput::Label(0), LFOutput::Label(0), LFOutput::Label(0)],
    ];

    let coverage = model.get_lf_coverage(&lf_matrix);

    assert_eq!(coverage.len(), 3);
    assert!((coverage[0] - 0.75).abs() < 1e-5); // 3/4
    assert!((coverage[1] - 0.75).abs() < 1e-5); // 3/4
    assert!((coverage[2] - 0.5).abs() < 1e-5); // 2/4
}

#[test]
fn test_confident_learning_creation() {
    let cl = ConfidentLearning::new(3);
    assert_eq!(cl.n_classes, 3);
    assert!((cl.threshold - 0.5).abs() < 1e-10);

    let cl2 = ConfidentLearning::with_threshold(3, 0.7);
    assert!((cl2.threshold - 0.7).abs() < 1e-10);
}

#[test]
fn test_confident_learning_find_issues() {
    let cl = ConfidentLearning::with_threshold(2, 0.6);

    let labels = vec![0, 1, 0, 1];
    let pred_probs = vec![
        Vector::from_slice(&[0.9, 0.1]), // Correct, confident
        Vector::from_slice(&[0.8, 0.2]), // Mislabeled (pred=0, label=1)
        Vector::from_slice(&[0.4, 0.6]), // Mislabeled (pred=1, label=0)
        Vector::from_slice(&[0.3, 0.7]), // Correct, confident
    ];

    let issues = cl.find_label_issues(&labels, &pred_probs);

    assert!(issues.contains(&1)); // Mislabeled
    assert!(issues.contains(&2)); // Mislabeled
    assert!(!issues.contains(&0)); // Correct
    assert!(!issues.contains(&3)); // Correct
}

#[test]
fn test_confident_joint() {
    let cl = ConfidentLearning::new(2);

    let labels = vec![0, 0, 1, 1];
    let pred_probs = vec![
        Vector::from_slice(&[0.9, 0.1]),
        Vector::from_slice(&[0.8, 0.2]),
        Vector::from_slice(&[0.2, 0.8]),
        Vector::from_slice(&[0.3, 0.7]),
    ];

    let joint = cl.compute_confident_joint(&labels, &pred_probs);

    assert_eq!(joint.len(), 2);
    assert_eq!(joint[0].len(), 2);

    // Sum should be 1
    let total: f32 = joint.iter().flat_map(|r| r.iter()).sum();
    assert!((total - 1.0).abs() < 0.1 || total == 0.0);
}

#[test]
fn test_noise_matrix_estimation() {
    let cl = ConfidentLearning::new(2);

    let labels = vec![0, 0, 1, 1, 0, 1];
    let pred_probs = vec![
        Vector::from_slice(&[0.9, 0.1]),
        Vector::from_slice(&[0.8, 0.2]),
        Vector::from_slice(&[0.1, 0.9]),
        Vector::from_slice(&[0.2, 0.8]),
        Vector::from_slice(&[0.7, 0.3]),
        Vector::from_slice(&[0.3, 0.7]),
    ];

    let noise = cl.estimate_noise_matrix(&labels, &pred_probs);

    // Each column should sum to ~1
    for j in 0..2 {
        let col_sum: f32 = (0..2).map(|i| noise[i][j]).sum();
        assert!((col_sum - 1.0).abs() < 0.1 || col_sum == 0.0);
    }
}