solvr 0.2.0

Advanced computing library for real-world problem solving - optimization, differential equations, interpolation, statistics, and more
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

//! CPU implementation of robust statistics algorithms.

use crate::stats::impl_generic::{
    median_abs_deviation_impl, siegelslopes_impl, theilslopes_impl, trim_mean_impl,
    winsorized_mean_impl,
};
use crate::stats::traits::{RobustRegressionResult, RobustStatisticsAlgorithms};
use numr::error::Result;
use numr::runtime::cpu::{CpuClient, CpuRuntime};
use numr::tensor::Tensor;

impl RobustStatisticsAlgorithms<CpuRuntime> for CpuClient {
    fn trim_mean(
        &self,
        x: &Tensor<CpuRuntime>,
        proportiontocut: f64,
    ) -> Result<Tensor<CpuRuntime>> {
        trim_mean_impl(self, x, proportiontocut)
    }

    fn winsorized_mean(
        &self,
        x: &Tensor<CpuRuntime>,
        proportiontocut: f64,
    ) -> Result<Tensor<CpuRuntime>> {
        winsorized_mean_impl(self, x, proportiontocut)
    }

    fn median_abs_deviation(
        &self,
        x: &Tensor<CpuRuntime>,
        scale: bool,
    ) -> Result<Tensor<CpuRuntime>> {
        median_abs_deviation_impl(self, x, scale)
    }

    fn siegelslopes(
        &self,
        x: &Tensor<CpuRuntime>,
        y: &Tensor<CpuRuntime>,
    ) -> Result<RobustRegressionResult<CpuRuntime>> {
        siegelslopes_impl(self, x, y)
    }

    fn theilslopes(
        &self,
        x: &Tensor<CpuRuntime>,
        y: &Tensor<CpuRuntime>,
    ) -> Result<RobustRegressionResult<CpuRuntime>> {
        theilslopes_impl(self, x, y)
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::stats::helpers::extract_scalar;
    use numr::runtime::cpu::CpuDevice;

    fn setup() -> (CpuClient, CpuDevice) {
        let device = CpuDevice::new();
        let client = CpuClient::new(device.clone());
        (client, device)
    }

    #[test]
    fn test_trim_mean() {
        let (client, device) = setup();
        // Data: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
        let data = Tensor::<CpuRuntime>::from_slice(
            &[1.0f64, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0],
            &[10],
            &device,
        );

        // 0% trim = regular mean
        let result = client.trim_mean(&data, 0.0).unwrap();
        let val = extract_scalar(&result).unwrap();
        assert!((val - 5.5).abs() < 1e-10);

        // 20% trim: cut 2 from each end -> mean(3,4,5,6,7,8) = 5.5
        let result = client.trim_mean(&data, 0.2).unwrap();
        let val = extract_scalar(&result).unwrap();
        assert!((val - 5.5).abs() < 1e-10);

        // 10% trim: cut 1 from each end -> mean(2..9) = 5.5
        let result = client.trim_mean(&data, 0.1).unwrap();
        let val = extract_scalar(&result).unwrap();
        assert!((val - 5.5).abs() < 1e-10);
    }

    #[test]
    fn test_winsorized_mean() {
        let (client, device) = setup();
        let data = Tensor::<CpuRuntime>::from_slice(
            &[1.0f64, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 100.0],
            &[10],
            &device,
        );

        // Without winsorizing
        let regular_mean = 145.0 / 10.0; // 14.5

        // 10% winsorize: replace 1->2, 100->9 -> mean(2,2,3,4,5,6,7,8,9,9) = 5.5
        let result = client.winsorized_mean(&data, 0.1).unwrap();
        let val = extract_scalar(&result).unwrap();
        assert!(val < regular_mean); // Should be much less due to outlier removal
        assert!((val - 5.5).abs() < 1e-10);
    }

    #[test]
    fn test_median_abs_deviation() {
        let (client, device) = setup();
        let data = Tensor::<CpuRuntime>::from_slice(&[1.0f64, 2.0, 3.0, 4.0, 5.0], &[5], &device);

        // Median = 3, |x - 3| = [2, 1, 0, 1, 2], median(|x-3|) = 1
        let result = client.median_abs_deviation(&data, false).unwrap();
        let val = extract_scalar(&result).unwrap();
        assert!((val - 1.0).abs() < 1e-10);

        // Scaled: 1.0 * 1.4826
        let result = client.median_abs_deviation(&data, true).unwrap();
        let val = extract_scalar(&result).unwrap();
        assert!((val - 1.4826).abs() < 1e-4);
    }

    #[test]
    fn test_theilslopes() {
        let (client, device) = setup();
        let x = Tensor::<CpuRuntime>::from_slice(&[1.0f64, 2.0, 3.0, 4.0, 5.0], &[5], &device);
        let y = Tensor::<CpuRuntime>::from_slice(&[2.0f64, 4.0, 6.0, 8.0, 10.0], &[5], &device);

        let result = client.theilslopes(&x, &y).unwrap();
        let slope = extract_scalar(&result.slope).unwrap();
        let intercept = extract_scalar(&result.intercept).unwrap();

        // Perfect linear: y = 2x, slope = 2, intercept = 0
        assert!((slope - 2.0).abs() < 1e-10);
        assert!((intercept - 0.0).abs() < 1e-10);
    }

    #[test]
    fn test_siegelslopes() {
        let (client, device) = setup();
        let x = Tensor::<CpuRuntime>::from_slice(&[1.0f64, 2.0, 3.0, 4.0, 5.0], &[5], &device);
        let y = Tensor::<CpuRuntime>::from_slice(&[2.0f64, 4.0, 6.0, 8.0, 10.0], &[5], &device);

        let result = client.siegelslopes(&x, &y).unwrap();
        let slope = extract_scalar(&result.slope).unwrap();
        let intercept = extract_scalar(&result.intercept).unwrap();

        assert!((slope - 2.0).abs() < 1e-10);
        assert!((intercept - 0.0).abs() < 1e-10);
    }
}