autoregressive/
lib.rs

1//! Create for sampling data from an autoregressive random process.
2//!
3//! For more information, please see https://en.wikipedia.org/wiki/Autoregressive_model.
4//!
5//! ```
6//! let mut ar = autoregressive::univariate::Autoregressive::new(0.0, 1.0, &[0.5]);
7//!
8//! // Sample next value
9//! let n = ar.step();
10//!
11//! // Take 10 values as an iterator
12//! let n10 = ar.take(10).collect::<Vec<f32>>();
13//! ```
14
15/// Univariate AR model
16///
17pub mod univariate {
18    use num_traits::Float;
19    use rand_distr::{Distribution, StandardNormal};
20
21    pub struct Autoregressive<F, const N: usize>
22    where
23        F: Float,
24        StandardNormal: Distribution<F>,
25    {
26        c: F,
27        x: [F; N],
28        phi: [F; N],
29        noise: rand_distr::Normal<F>,
30    }
31
32    impl<F, const N: usize> Autoregressive<F, N>
33    where
34        F: Float + std::iter::Sum,
35        StandardNormal: Distribution<F>,
36    {
37        /// Create a new instance
38        /// * `c`  parameter
39        /// * `noise_variance` Variance of the white noise (epsilon)
40        /// * `phi` model parameters
41        pub fn new(c: F, noise_variance: F, phi: &[F; N]) -> Self {
42            let x = [num_traits::identities::zero(); N];
43            let noise =
44                rand_distr::Normal::new(num_traits::identities::zero(), noise_variance).unwrap();
45            Self {
46                c,
47                phi: *phi,
48                x,
49                noise,
50            }
51        }
52
53        /// Next value from the AR model
54        pub fn step(&mut self) -> F {
55            let mut rng = rand::rng();
56            let epsilon: F = self.noise.sample(&mut rng);
57            let new_x = self.c
58                + self
59                    .x
60                    .iter()
61                    .zip(self.phi.iter())
62                    .map(|(x, p)| *x * *p)
63                    .sum::<F>()
64                + epsilon;
65            if !self.x.is_empty() {
66                self.x.rotate_right(1);
67                self.x[0] = new_x;
68            }
69            new_x
70        }
71    }
72
73    impl<F, const N: usize> Iterator for Autoregressive<F, N>
74    where
75        F: Float + std::iter::Sum,
76        StandardNormal: Distribution<F>,
77    {
78        type Item = F;
79
80        fn next(&mut self) -> Option<Self::Item> {
81            Some(self.step())
82        }
83    }
84}
85
86#[cfg(test)]
87mod test {
88    #[test]
89    fn bounded() {
90        const NUM: usize = 1_000_000;
91
92        let ar = super::univariate::Autoregressive::new(0.0, 1.0, &[]);
93        let avg = ar.take(NUM).sum::<f32>() / (NUM as f32);
94        assert!(avg.abs() < 1.0);
95
96        let ar = super::univariate::Autoregressive::new(0.0, 1.0, &[0.3]);
97        let avg = ar.take(NUM).sum::<f32>() / (NUM as f32);
98        assert!(avg.abs() < 1.0);
99
100        let ar = super::univariate::Autoregressive::new(0.0, 1.0, &[0.9]);
101        let avg = ar.take(NUM).sum::<f32>() / (NUM as f32);
102        assert!(avg.abs() < 1.0);
103
104        let ar = super::univariate::Autoregressive::new(0.0, 1.0, &[0.3, 0.3]);
105        let avg = ar.take(NUM).sum::<f32>() / (NUM as f32);
106        assert!(avg.abs() < 1.0);
107
108        let ar = super::univariate::Autoregressive::new(0.0, 1.0, &[0.9, -0.8]);
109        let avg = ar.take(NUM).sum::<f32>() / (NUM as f32);
110        assert!(avg.abs() < 1.0);
111    }
112}
autoregressive/lib.rs

autoregressive/
lib.rs
