touchstone-rs 0.3.0

A Rust library for benchmarking streaming anomaly detectors against labeled time-series datasets
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

use super::Metric;

/// Precision-Recall Area Under Curve metric.
pub struct PrAuc;

/// Average Precision metric: area under the precision-recall curve.
pub struct AveragePrecision;

impl Metric for PrAuc {
    fn name(&self) -> &str {
        "PR-AUC"
    }

    fn score(&self, labels: &[u8], scores: &[f32]) -> f64 {
        pr_auc(labels, scores)
    }
}

impl Metric for AveragePrecision {
    fn name(&self) -> &str {
        "AvgPrec"
    }

    fn score(&self, labels: &[u8], scores: &[f32]) -> f64 {
        average_precision(labels, scores)
    }
}

/// Pairs each score with its label and sorts descending by score.
fn sorted_pairs(labels: &[u8], scores: &[f32]) -> Vec<(f32, u8)> {
    let mut pairs: Vec<(f32, u8)> = scores
        .iter()
        .zip(labels.iter())
        .map(|(&s, &l)| (s, l))
        .collect();
    pairs.sort_by(|a, b| b.0.total_cmp(&a.0));
    pairs
}

/// Computes the Precision-Recall AUC score using the trapezoidal rule.
pub(crate) fn pr_auc(labels: &[u8], scores: &[f32]) -> f64 {
    let n_pos = labels.iter().filter(|&&l| l == 1).count();
    if n_pos == 0 {
        return f64::NAN;
    }

    let pairs = sorted_pairs(labels, scores);
    let n = pairs.len();

    // Walk from highest score down, computing precision/recall at each threshold change.
    let mut auc = 0.0_f64;
    let mut tp = 0_usize;
    let mut fp = 0_usize;
    let mut prev_rec = 0.0_f64;
    let mut prev_prec = 1.0_f64;
    let mut i = 0;

    while i < n {
        let thresh = pairs[i].0;
        while i < n && pairs[i].0 == thresh {
            if pairs[i].1 == 1 {
                tp += 1;
            } else {
                fp += 1;
            }
            i += 1;
        }
        let prec = tp as f64 / (tp + fp) as f64;
        let rec = tp as f64 / n_pos as f64;
        auc += (rec - prev_rec) * (prec + prev_prec) / 2.0;
        prev_rec = rec;
        prev_prec = prec;
    }
    auc
}

/// Computes PR-AUC with dilated anomaly ranges.
#[allow(dead_code)]
pub(crate) fn pr_auc_buffered(labels: &[u8], scores: &[f32], buffer: usize) -> f64 {
    use super::roc_auc::dilate;
    let dilated = dilate(labels, buffer);
    pr_auc(&dilated, scores)
}

/// Computes the Average Precision metric by integrating precision at recall transitions.
fn average_precision(labels: &[u8], scores: &[f32]) -> f64 {
    let n_pos = labels.iter().filter(|&&l| l == 1).count();
    if n_pos == 0 {
        return f64::NAN;
    }

    let pairs = sorted_pairs(labels, scores);
    let mut ap = 0.0_f64;
    let mut tp = 0_usize;
    let mut fp = 0_usize;
    let mut prev_rec = 0.0_f64;

    for (_, label) in &pairs {
        if *label == 1 {
            tp += 1;
        } else {
            fp += 1;
        }
        let prec = tp as f64 / (tp + fp) as f64;
        let rec = tp as f64 / n_pos as f64;
        ap += (rec - prev_rec) * prec;
        prev_rec = rec;
    }
    ap
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn perfect_pr_auc() {
        let labels = vec![0, 0, 1, 1];
        let scores = vec![0.1, 0.2, 0.8, 0.9];
        let auc = pr_auc(&labels, &scores);
        assert!((auc - 1.0).abs() < 1e-9, "got {auc}");
    }

    #[test]
    fn average_precision_perfect() {
        let labels = vec![0, 0, 1, 1];
        let scores = vec![0.1, 0.2, 0.8, 0.9];
        let ap = average_precision(&labels, &scores);
        assert!((ap - 1.0).abs() < 1e-9, "got {ap}");
    }
}