1 // Copyright 2018 Developers of the Rand project.
2 // Copyright 2013 The Rust Project Developers.
3 //
4 // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
5 // https://www.apache.org/licenses/LICENSE-2.0> or the MIT license
6 // <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your
7 // option. This file may not be copied, modified, or distributed
8 // except according to those terms.
9 
10 //! The normal and derived distributions.
11 
12 use Rng;
13 use distributions::{ziggurat_tables, Distribution, Open01};
14 use distributions::utils::ziggurat;
15 
16 /// Samples floating-point numbers according to the normal distribution
17 /// `N(0, 1)` (a.k.a. a standard normal, or Gaussian). This is equivalent to
18 /// `Normal::new(0.0, 1.0)` but faster.
19 ///
20 /// See `Normal` for the general normal distribution.
21 ///
22 /// Implemented via the ZIGNOR variant[^1] of the Ziggurat method.
23 ///
24 /// [^1]: Jurgen A. Doornik (2005). [*An Improved Ziggurat Method to
25 ///       Generate Normal Random Samples*](
26 ///       https://www.doornik.com/research/ziggurat.pdf).
27 ///       Nuffield College, Oxford
28 ///
29 /// # Example
30 /// ```
31 /// use rand::prelude::*;
32 /// use rand::distributions::StandardNormal;
33 ///
34 /// let val: f64 = SmallRng::from_entropy().sample(StandardNormal);
35 /// println!("{}", val);
36 /// ```
37 #[derive(Clone, Copy, Debug)]
38 pub struct StandardNormal;
39 
40 impl Distribution<f64> for StandardNormal {
sample<R: Rng + ?Sized>(&self, rng: &mut R) -> f6441     fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> f64 {
42         #[inline]
43         fn pdf(x: f64) -> f64 {
44             (-x*x/2.0).exp()
45         }
46         #[inline]
47         fn zero_case<R: Rng + ?Sized>(rng: &mut R, u: f64) -> f64 {
48             // compute a random number in the tail by hand
49 
50             // strange initial conditions, because the loop is not
51             // do-while, so the condition should be true on the first
52             // run, they get overwritten anyway (0 < 1, so these are
53             // good).
54             let mut x = 1.0f64;
55             let mut y = 0.0f64;
56 
57             while -2.0 * y < x * x {
58                 let x_: f64 = rng.sample(Open01);
59                 let y_: f64 = rng.sample(Open01);
60 
61                 x = x_.ln() / ziggurat_tables::ZIG_NORM_R;
62                 y = y_.ln();
63             }
64 
65             if u < 0.0 { x - ziggurat_tables::ZIG_NORM_R } else { ziggurat_tables::ZIG_NORM_R - x }
66         }
67 
68         ziggurat(rng, true, // this is symmetric
69                  &ziggurat_tables::ZIG_NORM_X,
70                  &ziggurat_tables::ZIG_NORM_F,
71                  pdf, zero_case)
72     }
73 }
74 
75 /// The normal distribution `N(mean, std_dev**2)`.
76 ///
77 /// This uses the ZIGNOR variant of the Ziggurat method, see [`StandardNormal`]
78 /// for more details.
79 ///
80 /// Note that [`StandardNormal`] is an optimised implementation for mean 0, and
81 /// standard deviation 1.
82 ///
83 /// # Example
84 ///
85 /// ```
86 /// use rand::distributions::{Normal, Distribution};
87 ///
88 /// // mean 2, standard deviation 3
89 /// let normal = Normal::new(2.0, 3.0);
90 /// let v = normal.sample(&mut rand::thread_rng());
91 /// println!("{} is from a N(2, 9) distribution", v)
92 /// ```
93 ///
94 /// [`StandardNormal`]: crate::distributions::StandardNormal
95 #[derive(Clone, Copy, Debug)]
96 pub struct Normal {
97     mean: f64,
98     std_dev: f64,
99 }
100 
101 impl Normal {
102     /// Construct a new `Normal` distribution with the given mean and
103     /// standard deviation.
104     ///
105     /// # Panics
106     ///
107     /// Panics if `std_dev < 0`.
108     #[inline]
new(mean: f64, std_dev: f64) -> Normal109     pub fn new(mean: f64, std_dev: f64) -> Normal {
110         assert!(std_dev >= 0.0, "Normal::new called with `std_dev` < 0");
111         Normal {
112             mean,
113             std_dev
114         }
115     }
116 }
117 impl Distribution<f64> for Normal {
sample<R: Rng + ?Sized>(&self, rng: &mut R) -> f64118     fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> f64 {
119         let n = rng.sample(StandardNormal);
120         self.mean + self.std_dev * n
121     }
122 }
123 
124 
125 /// The log-normal distribution `ln N(mean, std_dev**2)`.
126 ///
127 /// If `X` is log-normal distributed, then `ln(X)` is `N(mean, std_dev**2)`
128 /// distributed.
129 ///
130 /// # Example
131 ///
132 /// ```
133 /// use rand::distributions::{LogNormal, Distribution};
134 ///
135 /// // mean 2, standard deviation 3
136 /// let log_normal = LogNormal::new(2.0, 3.0);
137 /// let v = log_normal.sample(&mut rand::thread_rng());
138 /// println!("{} is from an ln N(2, 9) distribution", v)
139 /// ```
140 #[derive(Clone, Copy, Debug)]
141 pub struct LogNormal {
142     norm: Normal
143 }
144 
145 impl LogNormal {
146     /// Construct a new `LogNormal` distribution with the given mean
147     /// and standard deviation.
148     ///
149     /// # Panics
150     ///
151     /// Panics if `std_dev < 0`.
152     #[inline]
new(mean: f64, std_dev: f64) -> LogNormal153     pub fn new(mean: f64, std_dev: f64) -> LogNormal {
154         assert!(std_dev >= 0.0, "LogNormal::new called with `std_dev` < 0");
155         LogNormal { norm: Normal::new(mean, std_dev) }
156     }
157 }
158 impl Distribution<f64> for LogNormal {
sample<R: Rng + ?Sized>(&self, rng: &mut R) -> f64159     fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> f64 {
160         self.norm.sample(rng).exp()
161     }
162 }
163 
164 #[cfg(test)]
165 mod tests {
166     use distributions::Distribution;
167     use super::{Normal, LogNormal};
168 
169     #[test]
test_normal()170     fn test_normal() {
171         let norm = Normal::new(10.0, 10.0);
172         let mut rng = ::test::rng(210);
173         for _ in 0..1000 {
174             norm.sample(&mut rng);
175         }
176     }
177     #[test]
178     #[should_panic]
test_normal_invalid_sd()179     fn test_normal_invalid_sd() {
180         Normal::new(10.0, -1.0);
181     }
182 
183 
184     #[test]
test_log_normal()185     fn test_log_normal() {
186         let lnorm = LogNormal::new(10.0, 10.0);
187         let mut rng = ::test::rng(211);
188         for _ in 0..1000 {
189             lnorm.sample(&mut rng);
190         }
191     }
192     #[test]
193     #[should_panic]
test_log_normal_invalid_sd()194     fn test_log_normal_invalid_sd() {
195         LogNormal::new(10.0, -1.0);
196     }
197 }
198