rand_distr/
zipf.rs

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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
// Copyright 2021 Developers of the Rand project.
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// https://www.apache.org/licenses/LICENSE-2.0> or the MIT license
// <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your
// option. This file may not be copied, modified, or distributed
// except according to those terms.

//! The Zipf distribution.

use crate::{Distribution, Standard};
use core::fmt;
use num_traits::Float;
use rand::Rng;

/// The Zipf (Zipfian) distribution `Zipf(n, s)`.
///
/// The samples follow [Zipf's law](https://en.wikipedia.org/wiki/Zipf%27s_law):
/// The frequency of each sample from a finite set of size `n` is inversely
/// proportional to a power of its frequency rank (with exponent `s`).
///
/// For large `n`, this converges to the [`Zeta`](crate::Zeta) distribution.
///
/// For `s = 0`, this becomes a [`uniform`](crate::Uniform) distribution.
///
/// # Plot
///
/// The following plot illustrates the Zipf distribution for `n = 10` and
/// various values of `s`.
///
/// ![Zipf distribution](https://raw.githubusercontent.com/rust-random/charts/main/charts/zipf.svg)
///
/// # Example
/// ```
/// use rand::prelude::*;
/// use rand_distr::Zipf;
///
/// let val: f64 = rand::rng().sample(Zipf::new(10, 1.5).unwrap());
/// println!("{}", val);
/// ```
///
/// # Implementation details
///
/// Implemented via [rejection sampling](https://en.wikipedia.org/wiki/Rejection_sampling),
/// due to Jason Crease[1].
///
/// [1]: https://jasoncrease.medium.com/rejection-sampling-the-zipf-distribution-6b359792cffa
#[derive(Clone, Copy, Debug, PartialEq)]
pub struct Zipf<F>
where
    F: Float,
    Standard: Distribution<F>,
{
    s: F,
    t: F,
    q: F,
}

/// Error type returned from [`Zipf::new`].
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum Error {
    /// `s < 0` or `nan`.
    STooSmall,
    /// `n < 1`.
    NTooSmall,
}

impl fmt::Display for Error {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.write_str(match self {
            Error::STooSmall => "s < 0 or is NaN in Zipf distribution",
            Error::NTooSmall => "n < 1 in Zipf distribution",
        })
    }
}

#[cfg(feature = "std")]
impl std::error::Error for Error {}

impl<F> Zipf<F>
where
    F: Float,
    Standard: Distribution<F>,
{
    /// Construct a new `Zipf` distribution for a set with `n` elements and a
    /// frequency rank exponent `s`.
    ///
    /// For large `n`, rounding may occur to fit the number into the float type.
    #[inline]
    pub fn new(n: u64, s: F) -> Result<Zipf<F>, Error> {
        if !(s >= F::zero()) {
            return Err(Error::STooSmall);
        }
        if n < 1 {
            return Err(Error::NTooSmall);
        }
        let n = F::from(n).unwrap(); // This does not fail.
        let q = if s != F::one() {
            // Make sure to calculate the division only once.
            F::one() / (F::one() - s)
        } else {
            // This value is never used.
            F::zero()
        };
        let t = if s != F::one() {
            (n.powf(F::one() - s) - s) * q
        } else {
            F::one() + n.ln()
        };
        debug_assert!(t > F::zero());
        Ok(Zipf { s, t, q })
    }

    /// Inverse cumulative density function
    #[inline]
    fn inv_cdf(&self, p: F) -> F {
        let one = F::one();
        let pt = p * self.t;
        if pt <= one {
            pt
        } else if self.s != one {
            (pt * (one - self.s) + self.s).powf(self.q)
        } else {
            (pt - one).exp()
        }
    }
}

impl<F> Distribution<F> for Zipf<F>
where
    F: Float,
    Standard: Distribution<F>,
{
    #[inline]
    fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> F {
        let one = F::one();
        loop {
            let inv_b = self.inv_cdf(rng.sample(Standard));
            let x = (inv_b + one).floor();
            let mut ratio = x.powf(-self.s);
            if x > one {
                ratio = ratio * inv_b.powf(self.s)
            };

            let y = rng.sample(Standard);
            if y < ratio {
                return x;
            }
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    fn test_samples<F: Float + fmt::Debug, D: Distribution<F>>(distr: D, zero: F, expected: &[F]) {
        let mut rng = crate::test::rng(213);
        let mut buf = [zero; 4];
        for x in &mut buf {
            *x = rng.sample(&distr);
        }
        assert_eq!(buf, expected);
    }

    #[test]
    #[should_panic]
    fn zipf_s_too_small() {
        Zipf::new(10, -1.).unwrap();
    }

    #[test]
    #[should_panic]
    fn zipf_n_too_small() {
        Zipf::new(0, 1.).unwrap();
    }

    #[test]
    #[should_panic]
    fn zipf_nan() {
        Zipf::new(10, f64::NAN).unwrap();
    }

    #[test]
    fn zipf_sample() {
        let d = Zipf::new(10, 0.5).unwrap();
        let mut rng = crate::test::rng(2);
        for _ in 0..1000 {
            let r = d.sample(&mut rng);
            assert!(r >= 1.);
        }
    }

    #[test]
    fn zipf_sample_s_1() {
        let d = Zipf::new(10, 1.).unwrap();
        let mut rng = crate::test::rng(2);
        for _ in 0..1000 {
            let r = d.sample(&mut rng);
            assert!(r >= 1.);
        }
    }

    #[test]
    fn zipf_sample_s_0() {
        let d = Zipf::new(10, 0.).unwrap();
        let mut rng = crate::test::rng(2);
        for _ in 0..1000 {
            let r = d.sample(&mut rng);
            assert!(r >= 1.);
        }
        // TODO: verify that this is a uniform distribution
    }

    #[test]
    fn zipf_sample_large_n() {
        let d = Zipf::new(u64::MAX, 1.5).unwrap();
        let mut rng = crate::test::rng(2);
        for _ in 0..1000 {
            let r = d.sample(&mut rng);
            assert!(r >= 1.);
        }
        // TODO: verify that this is a zeta distribution
    }

    #[test]
    fn zipf_value_stability() {
        test_samples(Zipf::new(10, 0.5).unwrap(), 0f32, &[10.0, 2.0, 6.0, 7.0]);
        test_samples(Zipf::new(10, 2.0).unwrap(), 0f64, &[1.0, 2.0, 3.0, 2.0]);
    }

    #[test]
    fn zipf_distributions_can_be_compared() {
        assert_eq!(Zipf::new(1, 2.0), Zipf::new(1, 2.0));
    }
}