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
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
// Copyright 2018 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.

//! Weighted index sampling

use crate::distributions::uniform::{SampleBorrow, SampleUniform, UniformSampler};
use crate::distributions::Distribution;
use crate::Rng;
use core::cmp::PartialOrd;
use core::fmt;

// Note that this whole module is only imported if feature="alloc" is enabled.
use alloc::vec::Vec;

#[cfg(feature = "serde1")]
use serde::{Serialize, Deserialize};

/// A distribution using weighted sampling of discrete items
///
/// Sampling a `WeightedIndex` distribution returns the index of a randomly
/// selected element from the iterator used when the `WeightedIndex` was
/// created. The chance of a given element being picked is proportional to the
/// weight of the element. The weights can use any type `X` for which an
/// implementation of [`Uniform<X>`] exists. The implementation guarantees that
/// elements with zero weight are never picked, even when the weights are
/// floating point numbers.
///
/// # Performance
///
/// Time complexity of sampling from `WeightedIndex` is `O(log N)` where
/// `N` is the number of weights. As an alternative,
/// [`rand_distr::weighted_alias`](https://docs.rs/rand_distr/*/rand_distr/weighted_alias/index.html)
/// supports `O(1)` sampling, but with much higher initialisation cost.
///
/// A `WeightedIndex<X>` contains a `Vec<X>` and a [`Uniform<X>`] and so its
/// size is the sum of the size of those objects, possibly plus some alignment.
///
/// Creating a `WeightedIndex<X>` will allocate enough space to hold `N - 1`
/// weights of type `X`, where `N` is the number of weights. However, since
/// `Vec` doesn't guarantee a particular growth strategy, additional memory
/// might be allocated but not used. Since the `WeightedIndex` object also
/// contains an instance of `X::Sampler`, this might cause additional allocations,
/// though for primitive types, [`Uniform<X>`] doesn't allocate any memory.
///
/// Sampling from `WeightedIndex` will result in a single call to
/// `Uniform<X>::sample` (method of the [`Distribution`] trait), which typically
/// will request a single value from the underlying [`RngCore`], though the
/// exact number depends on the implementation of `Uniform<X>::sample`.
///
/// # Example
///
/// ```
/// use rand::prelude::*;
/// use rand::distributions::WeightedIndex;
///
/// let choices = ['a', 'b', 'c'];
/// let weights = [2,   1,   1];
/// let dist = WeightedIndex::new(&weights).unwrap();
/// let mut rng = thread_rng();
/// for _ in 0..100 {
///     // 50% chance to print 'a', 25% chance to print 'b', 25% chance to print 'c'
///     println!("{}", choices[dist.sample(&mut rng)]);
/// }
///
/// let items = [('a', 0.0), ('b', 3.0), ('c', 7.0)];
/// let dist2 = WeightedIndex::new(items.iter().map(|item| item.1)).unwrap();
/// for _ in 0..100 {
///     // 0% chance to print 'a', 30% chance to print 'b', 70% chance to print 'c'
///     println!("{}", items[dist2.sample(&mut rng)].0);
/// }
/// ```
///
/// [`Uniform<X>`]: crate::distributions::Uniform
/// [`RngCore`]: crate::RngCore
#[derive(Debug, Clone, PartialEq)]
#[cfg_attr(feature = "serde1", derive(Serialize, Deserialize))]
#[cfg_attr(doc_cfg, doc(cfg(feature = "alloc")))]
pub struct WeightedIndex<X: SampleUniform + PartialOrd> {
    cumulative_weights: Vec<X>,
    total_weight: X,
    weight_distribution: X::Sampler,
}

impl<X: SampleUniform + PartialOrd> WeightedIndex<X> {
    /// Creates a new a `WeightedIndex` [`Distribution`] using the values
    /// in `weights`. The weights can use any type `X` for which an
    /// implementation of [`Uniform<X>`] exists.
    ///
    /// Returns an error if the iterator is empty, if any weight is `< 0`, or
    /// if its total value is 0.
    ///
    /// [`Uniform<X>`]: crate::distributions::uniform::Uniform
    pub fn new<I>(weights: I) -> Result<WeightedIndex<X>, WeightedError>
    where
        I: IntoIterator,
        I::Item: SampleBorrow<X>,
        X: for<'a> ::core::ops::AddAssign<&'a X> + Clone + Default,
    {
        let mut iter = weights.into_iter();
        let mut total_weight: X = iter.next().ok_or(WeightedError::NoItem)?.borrow().clone();

        let zero = <X as Default>::default();
        if !(total_weight >= zero) {
            return Err(WeightedError::InvalidWeight);
        }

        let mut weights = Vec::<X>::with_capacity(iter.size_hint().0);
        for w in iter {
            // Note that `!(w >= x)` is not equivalent to `w < x` for partially
            // ordered types due to NaNs which are equal to nothing.
            if !(w.borrow() >= &zero) {
                return Err(WeightedError::InvalidWeight);
            }
            weights.push(total_weight.clone());
            total_weight += w.borrow();
        }

        if total_weight == zero {
            return Err(WeightedError::AllWeightsZero);
        }
        let distr = X::Sampler::new(zero, total_weight.clone());

        Ok(WeightedIndex {
            cumulative_weights: weights,
            total_weight,
            weight_distribution: distr,
        })
    }

    /// Update a subset of weights, without changing the number of weights.
    ///
    /// `new_weights` must be sorted by the index.
    ///
    /// Using this method instead of `new` might be more efficient if only a small number of
    /// weights is modified. No allocations are performed, unless the weight type `X` uses
    /// allocation internally.
    ///
    /// In case of error, `self` is not modified.
    pub fn update_weights(&mut self, new_weights: &[(usize, &X)]) -> Result<(), WeightedError>
    where X: for<'a> ::core::ops::AddAssign<&'a X>
            + for<'a> ::core::ops::SubAssign<&'a X>
            + Clone
            + Default {
        if new_weights.is_empty() {
            return Ok(());
        }

        let zero = <X as Default>::default();

        let mut total_weight = self.total_weight.clone();

        // Check for errors first, so we don't modify `self` in case something
        // goes wrong.
        let mut prev_i = None;
        for &(i, w) in new_weights {
            if let Some(old_i) = prev_i {
                if old_i >= i {
                    return Err(WeightedError::InvalidWeight);
                }
            }
            if !(*w >= zero) {
                return Err(WeightedError::InvalidWeight);
            }
            if i > self.cumulative_weights.len() {
                return Err(WeightedError::TooMany);
            }

            let mut old_w = if i < self.cumulative_weights.len() {
                self.cumulative_weights[i].clone()
            } else {
                self.total_weight.clone()
            };
            if i > 0 {
                old_w -= &self.cumulative_weights[i - 1];
            }

            total_weight -= &old_w;
            total_weight += w;
            prev_i = Some(i);
        }
        if total_weight <= zero {
            return Err(WeightedError::AllWeightsZero);
        }

        // Update the weights. Because we checked all the preconditions in the
        // previous loop, this should never panic.
        let mut iter = new_weights.iter();

        let mut prev_weight = zero.clone();
        let mut next_new_weight = iter.next();
        let &(first_new_index, _) = next_new_weight.unwrap();
        let mut cumulative_weight = if first_new_index > 0 {
            self.cumulative_weights[first_new_index - 1].clone()
        } else {
            zero.clone()
        };
        for i in first_new_index..self.cumulative_weights.len() {
            match next_new_weight {
                Some(&(j, w)) if i == j => {
                    cumulative_weight += w;
                    next_new_weight = iter.next();
                }
                _ => {
                    let mut tmp = self.cumulative_weights[i].clone();
                    tmp -= &prev_weight; // We know this is positive.
                    cumulative_weight += &tmp;
                }
            }
            prev_weight = cumulative_weight.clone();
            core::mem::swap(&mut prev_weight, &mut self.cumulative_weights[i]);
        }

        self.total_weight = total_weight;
        self.weight_distribution = X::Sampler::new(zero, self.total_weight.clone());

        Ok(())
    }
}

impl<X> Distribution<usize> for WeightedIndex<X>
where X: SampleUniform + PartialOrd
{
    fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> usize {
        use ::core::cmp::Ordering;
        let chosen_weight = self.weight_distribution.sample(rng);
        // Find the first item which has a weight *higher* than the chosen weight.
        self.cumulative_weights
            .binary_search_by(|w| {
                if *w <= chosen_weight {
                    Ordering::Less
                } else {
                    Ordering::Greater
                }
            })
            .unwrap_err()
    }
}

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

    #[cfg(feature = "serde1")]
    #[test]
    fn test_weightedindex_serde1() {
        let weighted_index = WeightedIndex::new(&[1, 2, 3, 4, 5, 6, 7, 8, 9, 10]).unwrap();

        let ser_weighted_index = bincode::serialize(&weighted_index).unwrap();
        let de_weighted_index: WeightedIndex<i32> =
            bincode::deserialize(&ser_weighted_index).unwrap();

        assert_eq!(
            de_weighted_index.cumulative_weights,
            weighted_index.cumulative_weights
        );
        assert_eq!(de_weighted_index.total_weight, weighted_index.total_weight);
    }

    #[test]
    fn test_accepting_nan(){
        assert_eq!(
            WeightedIndex::new(&[core::f32::NAN, 0.5]).unwrap_err(),
            WeightedError::InvalidWeight,
        );
        assert_eq!(
            WeightedIndex::new(&[core::f32::NAN]).unwrap_err(),
            WeightedError::InvalidWeight,
        );
        assert_eq!(
            WeightedIndex::new(&[0.5, core::f32::NAN]).unwrap_err(),
            WeightedError::InvalidWeight,
        );

        assert_eq!(
            WeightedIndex::new(&[0.5, 7.0])
                .unwrap()
                .update_weights(&[(0, &core::f32::NAN)])
                .unwrap_err(),
            WeightedError::InvalidWeight,
        )
    }


    #[test]
    #[cfg_attr(miri, ignore)] // Miri is too slow
    fn test_weightedindex() {
        let mut r = crate::test::rng(700);
        const N_REPS: u32 = 5000;
        let weights = [1u32, 2, 3, 0, 5, 6, 7, 1, 2, 3, 4, 5, 6, 7];
        let total_weight = weights.iter().sum::<u32>() as f32;

        let verify = |result: [i32; 14]| {
            for (i, count) in result.iter().enumerate() {
                let exp = (weights[i] * N_REPS) as f32 / total_weight;
                let mut err = (*count as f32 - exp).abs();
                if err != 0.0 {
                    err /= exp;
                }
                assert!(err <= 0.25);
            }
        };

        // WeightedIndex from vec
        let mut chosen = [0i32; 14];
        let distr = WeightedIndex::new(weights.to_vec()).unwrap();
        for _ in 0..N_REPS {
            chosen[distr.sample(&mut r)] += 1;
        }
        verify(chosen);

        // WeightedIndex from slice
        chosen = [0i32; 14];
        let distr = WeightedIndex::new(&weights[..]).unwrap();
        for _ in 0..N_REPS {
            chosen[distr.sample(&mut r)] += 1;
        }
        verify(chosen);

        // WeightedIndex from iterator
        chosen = [0i32; 14];
        let distr = WeightedIndex::new(weights.iter()).unwrap();
        for _ in 0..N_REPS {
            chosen[distr.sample(&mut r)] += 1;
        }
        verify(chosen);

        for _ in 0..5 {
            assert_eq!(WeightedIndex::new(&[0, 1]).unwrap().sample(&mut r), 1);
            assert_eq!(WeightedIndex::new(&[1, 0]).unwrap().sample(&mut r), 0);
            assert_eq!(
                WeightedIndex::new(&[0, 0, 0, 0, 10, 0])
                    .unwrap()
                    .sample(&mut r),
                4
            );
        }

        assert_eq!(
            WeightedIndex::new(&[10][0..0]).unwrap_err(),
            WeightedError::NoItem
        );
        assert_eq!(
            WeightedIndex::new(&[0]).unwrap_err(),
            WeightedError::AllWeightsZero
        );
        assert_eq!(
            WeightedIndex::new(&[10, 20, -1, 30]).unwrap_err(),
            WeightedError::InvalidWeight
        );
        assert_eq!(
            WeightedIndex::new(&[-10, 20, 1, 30]).unwrap_err(),
            WeightedError::InvalidWeight
        );
        assert_eq!(
            WeightedIndex::new(&[-10]).unwrap_err(),
            WeightedError::InvalidWeight
        );
    }

    #[test]
    fn test_update_weights() {
        let data = [
            (
                &[10u32, 2, 3, 4][..],
                &[(1, &100), (2, &4)][..], // positive change
                &[10, 100, 4, 4][..],
            ),
            (
                &[1u32, 2, 3, 0, 5, 6, 7, 1, 2, 3, 4, 5, 6, 7][..],
                &[(2, &1), (5, &1), (13, &100)][..], // negative change and last element
                &[1u32, 2, 1, 0, 5, 1, 7, 1, 2, 3, 4, 5, 6, 100][..],
            ),
        ];

        for (weights, update, expected_weights) in data.iter() {
            let total_weight = weights.iter().sum::<u32>();
            let mut distr = WeightedIndex::new(weights.to_vec()).unwrap();
            assert_eq!(distr.total_weight, total_weight);

            distr.update_weights(update).unwrap();
            let expected_total_weight = expected_weights.iter().sum::<u32>();
            let expected_distr = WeightedIndex::new(expected_weights.to_vec()).unwrap();
            assert_eq!(distr.total_weight, expected_total_weight);
            assert_eq!(distr.total_weight, expected_distr.total_weight);
            assert_eq!(distr.cumulative_weights, expected_distr.cumulative_weights);
        }
    }

    #[test]
    fn value_stability() {
        fn test_samples<X: SampleUniform + PartialOrd, I>(
            weights: I, buf: &mut [usize], expected: &[usize],
        ) where
            I: IntoIterator,
            I::Item: SampleBorrow<X>,
            X: for<'a> ::core::ops::AddAssign<&'a X> + Clone + Default,
        {
            assert_eq!(buf.len(), expected.len());
            let distr = WeightedIndex::new(weights).unwrap();
            let mut rng = crate::test::rng(701);
            for r in buf.iter_mut() {
                *r = rng.sample(&distr);
            }
            assert_eq!(buf, expected);
        }

        let mut buf = [0; 10];
        test_samples(&[1i32, 1, 1, 1, 1, 1, 1, 1, 1], &mut buf, &[
            0, 6, 2, 6, 3, 4, 7, 8, 2, 5,
        ]);
        test_samples(&[0.7f32, 0.1, 0.1, 0.1], &mut buf, &[
            0, 0, 0, 1, 0, 0, 2, 3, 0, 0,
        ]);
        test_samples(&[1.0f64, 0.999, 0.998, 0.997], &mut buf, &[
            2, 2, 1, 3, 2, 1, 3, 3, 2, 1,
        ]);
    }

    #[test]
    fn weighted_index_distributions_can_be_compared() {
        assert_eq!(WeightedIndex::new(&[1, 2]), WeightedIndex::new(&[1, 2]));
    }
}

/// Error type returned from `WeightedIndex::new`.
#[cfg_attr(doc_cfg, doc(cfg(feature = "alloc")))]
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum WeightedError {
    /// The provided weight collection contains no items.
    NoItem,

    /// A weight is either less than zero, greater than the supported maximum,
    /// NaN, or otherwise invalid.
    InvalidWeight,

    /// All items in the provided weight collection are zero.
    AllWeightsZero,

    /// Too many weights are provided (length greater than `u32::MAX`)
    TooMany,
}

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

impl fmt::Display for WeightedError {
    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
        f.write_str(match *self {
            WeightedError::NoItem => "No weights provided in distribution",
            WeightedError::InvalidWeight => "A weight is invalid in distribution",
            WeightedError::AllWeightsZero => "All weights are zero in distribution",
            WeightedError::TooMany => "Too many weights (hit u32::MAX) in distribution",
        })
    }
}