numeris 0.5.13

Pure-Rust numerical algorithms library — high performance with SIMD support while also supporting no-std for embedded and WASM targets.
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

use super::{normal_quantile_standard, quantile_newton, ContinuousDistribution, StatsError};
use crate::special::{gamma_inc, lgamma};
use crate::FloatScalar;

/// Gamma distribution with shape α and rate β.
///
/// f(x) = β^α x^{α−1} e^{−βx} / Γ(α) for x > 0.
///
/// The scale parameter is θ = 1/β.
///
/// # Example
///
/// ```
/// use numeris::stats::{Gamma, ContinuousDistribution};
///
/// let g = Gamma::new(2.0_f64, 1.0).unwrap();
/// assert!((g.mean() - 2.0).abs() < 1e-14);
/// assert!((g.variance() - 2.0).abs() < 1e-14);
/// ```
#[derive(Debug, Clone, Copy)]
pub struct Gamma<T> {
    shape: T,   // α > 0
    rate: T,    // β > 0
    ln_norm: T, // α·ln β − lgamma(α), cached at construction
}

impl<T: FloatScalar> Gamma<T> {
    /// Create a Gamma distribution with `shape` α and `rate` β.
    /// Requires both > 0. Scale = 1/rate.
    pub fn new(shape: T, rate: T) -> Result<Self, StatsError> {
        if shape <= T::zero() || rate <= T::zero() {
            return Err(StatsError::InvalidParameter);
        }
        let ln_norm = shape * rate.ln() - lgamma(shape);
        Ok(Self {
            shape,
            rate,
            ln_norm,
        })
    }
}

impl<T: FloatScalar> Gamma<T> {
    /// Draw a random sample from this distribution (Marsaglia & Tsang's method).
    pub fn sample(&self, rng: &mut super::Rng) -> T {
        // Gamma(shape, rate) = Gamma(shape, 1) / rate
        rng.next_gamma(self.shape) / self.rate
    }

    impl_sample_array!(T, T::zero());
}

impl<T: FloatScalar> ContinuousDistribution<T> for Gamma<T> {
    fn pdf(&self, x: T) -> T {
        if x < T::zero() {
            return T::zero();
        }
        if x == T::zero() {
            let one = T::one();
            if self.shape == one {
                return self.rate; // Exponential special case
            } else if self.shape > one {
                return T::zero(); // mode is interior
            } else {
                return T::infinity(); // shape < 1: density blows up at 0
            }
        }
        self.ln_pdf(x).exp()
    }

    fn ln_pdf(&self, x: T) -> T {
        if x < T::zero() {
            return T::neg_infinity();
        }
        if x == T::zero() {
            return self.pdf(x).ln();
        }
        let one = T::one();
        self.ln_norm + (self.shape - one) * x.ln() - self.rate * x
    }

    fn cdf(&self, x: T) -> T {
        if x <= T::zero() {
            return T::zero();
        }
        gamma_inc(self.shape, self.rate * x).unwrap_or(T::nan())
    }

    fn quantile(&self, p: T) -> T {
        let mean = self.mean();
        let std = self.variance().sqrt();
        // Wilson-Hilferty initial guess for shape >= 1
        let x0 = if self.shape >= T::one() {
            let nine = T::from(9.0).unwrap();
            let z = normal_quantile_standard(p);
            let v = T::one() - T::one() / (nine * self.shape) + z / (nine * self.shape).sqrt();
            let wh = self.shape / self.rate * v * v * v;
            if wh > T::zero() {
                wh
            } else {
                mean
            }
        } else {
            mean
        };
        let hi = mean + T::from(40.0).unwrap() * std;
        quantile_newton(|x| self.cdf(x), |x| self.pdf(x), p, x0, T::zero(), hi)
    }

    fn mean(&self) -> T {
        self.shape / self.rate
    }

    fn variance(&self) -> T {
        self.shape / (self.rate * self.rate)
    }
}