perpetual 2.1.0

A self-generalizing gradient boosting machine that doesn't need hyperparameter optimization
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

use crate::metrics::evaluation::EvaluationMetric;

pub struct QuantileLossMetric {}
impl EvaluationMetric for QuantileLossMetric {
    fn calculate_metric(y: &[f64], yhat: &[f64], sample_weight: &[f64], _group: &[u64], alpha: Option<f32>) -> f64 {
        quantile_loss(y, yhat, sample_weight, alpha)
    }
    fn maximize() -> bool {
        false
    }
}

pub struct RootMeanSquaredLogErrorMetric {}
impl EvaluationMetric for RootMeanSquaredLogErrorMetric {
    fn calculate_metric(y: &[f64], yhat: &[f64], sample_weight: &[f64], _group: &[u64], _alpha: Option<f32>) -> f64 {
        root_mean_squared_log_error(y, yhat, sample_weight)
    }
    fn maximize() -> bool {
        false
    }
}

pub struct RootMeanSquaredErrorMetric {}
impl EvaluationMetric for RootMeanSquaredErrorMetric {
    fn calculate_metric(y: &[f64], yhat: &[f64], sample_weight: &[f64], _group: &[u64], _alpha: Option<f32>) -> f64 {
        root_mean_squared_error(y, yhat, sample_weight)
    }
    fn maximize() -> bool {
        false
    }
}

pub fn quantile_loss(y: &[f64], yhat: &[f64], sample_weight: &[f64], alpha: Option<f32>) -> f64 {
    let mut w_sum = 0.;
    let res = y
        .iter()
        .zip(yhat)
        .zip(sample_weight)
        .map(|((y_, yhat_), w_)| {
            w_sum += *w_;
            let _alpha = alpha.unwrap() as f64;
            let s = *y_ - *yhat_;
            let l = if s >= 0.0 { _alpha * s } else { (1.0 - _alpha) * s };
            l * *w_
        })
        .sum::<f64>();
    res / w_sum
}

pub fn root_mean_squared_log_error(y: &[f64], yhat: &[f64], sample_weight: &[f64]) -> f64 {
    let mut w_sum = 0.;
    let res = y
        .iter()
        .zip(yhat)
        .zip(sample_weight)
        .map(|((y_, yhat_), w_)| {
            w_sum += *w_;
            (y_.ln_1p() - yhat_.ln_1p()).powi(2) * *w_
        })
        .sum::<f64>();
    (res / w_sum).sqrt()
}

pub fn root_mean_squared_error(y: &[f64], yhat: &[f64], sample_weight: &[f64]) -> f64 {
    let mut w_sum = 0.;
    let res = y
        .iter()
        .zip(yhat)
        .zip(sample_weight)
        .map(|((y_, yhat_), w_)| {
            w_sum += *w_;
            (y_ - yhat_).powi(2) * *w_
        })
        .sum::<f64>();
    (res / w_sum).sqrt()
}

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

    #[test]
    fn test_quantile_loss_median() {
        let y = vec![1.0, 2.0, 3.0, 4.0];
        let yhat = vec![1.5, 2.5, 2.5, 3.5];
        let w = vec![1.0; 4];
        let loss = quantile_loss(&y, &yhat, &w, Some(0.5));
        assert!(loss.is_finite());
        // median quantile: symmetric penalty
        assert!((loss - 0.0).abs() < 1.0); // sanity check
    }

    #[test]
    fn test_quantile_loss_low_alpha() {
        let y = vec![1.0, 2.0, 3.0];
        let yhat = vec![2.0, 3.0, 4.0]; // all over-predicted
        let w = vec![1.0; 3];
        let loss = quantile_loss(&y, &yhat, &w, Some(0.1));
        assert!(loss.is_finite());
    }

    #[test]
    fn test_quantile_loss_high_alpha() {
        let y = vec![1.0, 2.0, 3.0];
        let yhat = vec![0.0, 1.0, 2.0]; // all under-predicted
        let w = vec![1.0; 3];
        let loss = quantile_loss(&y, &yhat, &w, Some(0.9));
        assert!(loss.is_finite());
    }

    #[test]
    fn test_quantile_loss_weighted() {
        let y = vec![1.0, 2.0, 3.0];
        let yhat = vec![1.5, 2.5, 2.5];
        let w = vec![2.0, 1.0, 3.0];
        let loss = quantile_loss(&y, &yhat, &w, Some(0.5));
        assert!(loss.is_finite());
    }

    #[test]
    fn test_quantile_loss_metric_trait() {
        let y = vec![1.0, 2.0, 3.0];
        let yhat = vec![1.5, 2.5, 2.5];
        let w = vec![1.0; 3];
        let group = vec![0u64; 3];
        let metric = QuantileLossMetric::calculate_metric(&y, &yhat, &w, &group, Some(0.5));
        assert!(metric.is_finite());
        assert!(!QuantileLossMetric::maximize());
    }

    #[test]
    fn test_rmsle_perfect() {
        let y = vec![1.0, 2.0, 3.0];
        let yhat = vec![1.0, 2.0, 3.0];
        let w = vec![1.0; 3];
        let rmsle = root_mean_squared_log_error(&y, &yhat, &w);
        assert!((rmsle - 0.0).abs() < 1e-10);
    }

    #[test]
    fn test_rmsle_imperfect() {
        let y = vec![1.0, 2.0, 3.0];
        let yhat = vec![1.5, 2.5, 3.5];
        let w = vec![1.0; 3];
        let rmsle = root_mean_squared_log_error(&y, &yhat, &w);
        assert!(rmsle > 0.0);
        assert!(rmsle.is_finite());
    }

    #[test]
    fn test_rmsle_metric_trait() {
        let y = vec![1.0, 2.0, 3.0];
        let yhat = vec![1.5, 2.5, 3.5];
        let w = vec![1.0; 3];
        let group = vec![0u64; 3];
        let metric = RootMeanSquaredLogErrorMetric::calculate_metric(&y, &yhat, &w, &group, None);
        assert!(metric > 0.0);
        assert!(!RootMeanSquaredLogErrorMetric::maximize());
    }

    #[test]
    fn test_rmse_perfect() {
        let y = vec![1.0, 2.0, 3.0];
        let yhat = vec![1.0, 2.0, 3.0];
        let w = vec![1.0; 3];
        let rmse = root_mean_squared_error(&y, &yhat, &w);
        assert!((rmse - 0.0).abs() < 1e-10);
    }

    #[test]
    fn test_rmse_imperfect() {
        let y = vec![1.0, 2.0, 3.0];
        let yhat = vec![2.0, 3.0, 4.0];
        let w = vec![1.0; 3];
        let rmse = root_mean_squared_error(&y, &yhat, &w);
        // Each error is 1.0, so MSE = 1.0, RMSE = 1.0
        assert!((rmse - 1.0).abs() < 1e-10);
    }

    #[test]
    fn test_rmse_weighted() {
        let y = vec![1.0, 2.0];
        let yhat = vec![2.0, 2.0]; // errors: 1.0 and 0.0
        let w = vec![2.0, 1.0]; // weight = 2.0 on error=1
        let rmse = root_mean_squared_error(&y, &yhat, &w);
        // weighted MSE = (2.0*1.0 + 1.0*0.0) / 3.0 = 2/3
        let expected = (2.0_f64 / 3.0).sqrt();
        assert!((rmse - expected).abs() < 1e-10);
    }

    #[test]
    fn test_rmse_metric_trait() {
        let y = vec![1.0, 2.0, 3.0];
        let yhat = vec![2.0, 3.0, 4.0];
        let w = vec![1.0; 3];
        let group = vec![0u64; 3];
        let metric = RootMeanSquaredErrorMetric::calculate_metric(&y, &yhat, &w, &group, None);
        assert!((metric - 1.0).abs() < 1e-10);
        assert!(!RootMeanSquaredErrorMetric::maximize());
    }
}