changeforest 1.2.1

Random Forests for Change Point Detection
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

use crate::gain::GainResult;
use crate::optimizer::OptimizerResult;
use crate::{Control, Gain, ModelSelectionResult, Optimizer};

pub struct GridSearch<T: Gain> {
    pub gain: T,
}

impl<T> Optimizer for GridSearch<T>
where
    T: Gain,
{
    fn n(&self) -> usize {
        self.gain.n()
    }

    fn control(&self) -> &Control {
        self.gain.control()
    }

    fn find_best_split(&self, start: usize, stop: usize) -> Result<OptimizerResult, &str> {
        let split_candidates = self.split_candidates(start, stop)?;

        let mut full_gain = self.gain.gain_full(start, stop, &split_candidates);

        let mut best_split = 0;
        let mut max_gain = -f64::INFINITY;

        for index in split_candidates {
            if full_gain.gain[index - start] > max_gain {
                best_split = index;
                max_gain = full_gain.gain[index - start];
            }
        }

        full_gain.max_gain = Some(max_gain);
        full_gain.best_split = Some(best_split);

        Ok(OptimizerResult {
            start,
            stop,
            best_split,
            max_gain,
            gain_results: vec![GainResult::FullGainResult(full_gain)],
        })
    }

    fn model_selection(&self, optimizer_result: &OptimizerResult) -> ModelSelectionResult {
        self.gain.model_selection(optimizer_result)
    }
}

#[cfg(test)]
mod tests {

    use super::*;
    use crate::testing;
    use rstest::*;

    #[rstest]
    #[case(0, 7, 2)]
    #[case(1, 7, 4)]
    #[case(2, 7, 4)]
    #[case(3, 7, 4)]
    #[case(1, 5, 2)]
    #[case(1, 6, 4)]
    #[case(1, 7, 4)]
    #[case(2, 6, 4)]
    #[case(2, 7, 4)]
    #[case(3, 6, 4)]
    fn test_change_in_mean_find_best_split(
        #[case] start: usize,
        #[case] stop: usize,
        #[case] expected: usize,
    ) {
        let X = ndarray::array![
            [0., 1.],
            [0., 1.],
            [1., -1.],
            [1., -1.],
            [-1., -1.],
            [-1., -1.],
            [-1., -1.]
        ];
        let X_view = X.view();
        assert_eq!(X_view.shape(), &[7, 2]);
        let control = Control::default().with_minimal_relative_segment_length(0.1);

        let gain = testing::ChangeInMean::new(&X_view, &control);
        let grid_search = GridSearch { gain };
        assert_eq!(
            grid_search.find_best_split(start, stop).unwrap().best_split,
            expected
        );
    }

    #[rstest]
    #[case(0, 10, Some(vec![(0, 3)]), 0.09, vec![4, 5, 6, 7, 8])]
    #[case(1, 10, Some(vec![(6, 10)]), 0.15, vec![3, 4, 5, 6])]
    #[case(0, 10, Some(vec![(2, 4), (5, 7)]), 0.09, vec![1, 2, 5, 8])]
    #[case(1, 7, Some(vec![(2, 4), (5, 7)]), 0.09, vec![2, 5])]
    fn test_split_candidates(
        #[case] start: usize,
        #[case] stop: usize,
        #[case] forbidden_segments: Option<Vec<(usize, usize)>>,
        #[case] delta: f64,
        #[case] expected: Vec<usize>,
    ) {
        let X = ndarray::array![
            [0.0],
            [0.0],
            [0.0],
            [0.0],
            [-0.0],
            [-0.0],
            [-0.0],
            [-0.0],
            [-0.0],
            [-0.0]
        ];
        let X_view = X.view();
        let control = Control::default()
            .with_minimal_relative_segment_length(delta)
            .with_forbidden_segments(forbidden_segments);
        let gain = testing::ChangeInMean::new(&X_view, &control);
        let grid_search = GridSearch { gain };
        assert_eq!(grid_search.split_candidates(start, stop).unwrap(), expected);
    }
}