ci_core/ci_tests/
chi_squared.rs

1use crate::strategy::{CITest, CITestDataType, TestResult};
2use crate::utils::power_divergence::power_divergence;
3
4use ndarray::{Array1, Array2};
5
6const CHI_SQUARED_LAMBDA: f64 = 1.0;
7
8/// Pearson chi-squared conditional independence test (λ = 1).
9///
10/// Operates on discrete data only. Delegates to the power-divergence family
11/// with λ = 1, which is the classical chi-squared statistic.
12#[derive(Debug, Clone, PartialEq)]
13pub struct ChiSquared {
14    pub boolean: bool,
15    pub significance_level: f64,
16}
17
18impl ChiSquared {
19    #[must_use]
20    pub fn new(boolean: bool, significance_level: f64) -> Self {
21        Self {
22            boolean,
23            significance_level,
24        }
25    }
26}
27
28impl CITest for ChiSquared {
29    fn run_test(
30        &self,
31        x_values: Array1<f64>,
32        y_values: Array1<f64>,
33        z: Array2<f64>,
34    ) -> anyhow::Result<TestResult> {
35        power_divergence(
36            &x_values,
37            &y_values,
38            &z,
39            self.boolean,
40            self.significance_level,
41            CHI_SQUARED_LAMBDA,
42        )
43    }
44
45    fn data_types(&self) -> &'static [CITestDataType] {
46        &[CITestDataType::Discrete]
47    }
48}
49
50#[cfg(test)]
51#[allow(clippy::many_single_char_names)]
52mod tests {
53    use super::*;
54    use crate::utils::EPS;
55    use ndarray::{array, Array2};
56
57    fn unwrap_correlated(r: &TestResult) -> (f64, f64, usize) {
58        match r {
59            TestResult::Statistic(a, b, c) => (*a, *b, *c),
60            _ => panic!("expected Correlated2"),
61        }
62    }
63
64    #[test]
65    fn uncond_independent_data_accepted() {
66        let t = ChiSquared {
67            boolean: false,
68            significance_level: 0.05,
69        };
70        let x = array![1., 1., 2., 2., 1., 1., 2., 2.];
71        let y = array![1., 2., 1., 2., 1., 2., 1., 2.];
72        let empty = Array2::<f64>::zeros((0, 0));
73
74        let (p, stat, dof) = unwrap_correlated(&t.run_test(x, y, empty).unwrap());
75        assert!(stat.abs() < EPS, "stat should be ~0, got {stat}");
76        assert!(p > 0.99);
77        assert_eq!(dof, 1);
78    }
79
80    #[test]
81    fn cond_independent_data_accepted() {
82        let t = ChiSquared {
83            boolean: false,
84            significance_level: 0.05,
85        };
86        let x = array![1., 1., 2., 2., 1., 1., 2., 2.];
87        let y = array![1., 2., 1., 2., 1., 2., 1., 2.];
88        let z = array![[1.], [1.], [1.], [1.], [2.], [2.], [2.], [2.]];
89
90        let (p, stat, dof) = unwrap_correlated(&t.run_test(x, y, z).unwrap());
91        assert!(stat.abs() < EPS, "stat should be ~0, got {stat}");
92        assert!(p > 0.99);
93        assert_eq!(dof, 2);
94    }
95
96    #[test]
97    fn uncond_dependent_data_rejected() {
98        let t = ChiSquared {
99            boolean: false,
100            significance_level: 0.05,
101        };
102        let x = array![1., 1., 1., 1., 2., 2., 2., 2.];
103        let y = array![1., 1., 1., 1., 2., 2., 2., 2.];
104        let empty = Array2::<f64>::zeros((0, 0));
105
106        let (p, stat, dof) = unwrap_correlated(&t.run_test(x, y, empty).unwrap());
107        assert!((stat - 8.0).abs() < EPS, "got {stat}");
108        assert!((p - 0.004_677_734_981_047_276).abs() < EPS, "got {p}");
109        assert_eq!(dof, 1);
110    }
111
112    #[test]
113    fn cond_dependent_data_rejected() {
114        let t = ChiSquared {
115            boolean: false,
116            significance_level: 0.05,
117        };
118        let x = array![1., 1., 2., 2., 1., 1., 2., 2.];
119        let y = array![1., 1., 2., 2., 1., 1., 2., 2.];
120        let z = array![[1.], [1.], [1.], [1.], [2.], [2.], [2.], [2.]];
121
122        let (p, stat, dof) = unwrap_correlated(&t.run_test(x, y, z).unwrap());
123        assert!((stat - 8.0).abs() < EPS, "stat {stat} should be larger");
124        assert!(
125            (p - 0.018_315_638_888_734_193).abs() < EPS,
126            "rejected p value {p}"
127        );
128        assert_eq!(dof, 2);
129    }
130
131    #[test]
132    fn uncond_boolean_mode() {
133        let t = ChiSquared {
134            boolean: true,
135            significance_level: 0.05,
136        };
137        let empty = Array2::<f64>::zeros((0, 0));
138        let x = array![1., 1., 2., 2., 1., 1., 2., 2.];
139        let y = array![1., 2., 1., 2., 1., 2., 1., 2.];
140        let r = t.run_test(x, y, empty.clone()).unwrap();
141        assert!(matches!(r, TestResult::Boolean(true)));
142
143        let x = array![1., 1., 1., 1., 2., 2., 2., 2.];
144        let y = array![1., 1., 1., 1., 2., 2., 2., 2.];
145        let r = t.run_test(x, y, empty).unwrap();
146        assert!(matches!(r, TestResult::Boolean(false)));
147    }
148
149    #[test]
150    fn cond_boolean_mode() {
151        let t = ChiSquared {
152            boolean: true,
153            significance_level: 0.05,
154        };
155        let z = array![[1.], [1.], [1.], [1.], [2.], [2.], [2.], [2.]];
156        let x = array![1., 1., 2., 2., 1., 1., 2., 2.];
157        let y = array![1., 2., 1., 2., 1., 2., 1., 2.];
158        let r = t.run_test(x, y, z).unwrap();
159        assert!(matches!(r, TestResult::Boolean(true)));
160
161        let x = array![1., 1., 2., 2., 1., 1., 2., 2.];
162        let y = array![1., 1., 2., 2., 1., 1., 2., 2.];
163        let z = array![[1.], [1.], [1.], [1.], [2.], [2.], [2.], [2.]];
164        let r = t.run_test(x, y, z).unwrap();
165        assert!(matches!(r, TestResult::Boolean(false)));
166    }
167}
ci_core/ci_tests/chi_squared.rs

ci_core/ci_tests/
chi_squared.rs
