1 // Copyright 2012 The Rust Project Developers. See the COPYRIGHT
2 // file at the top-level directory of this distribution and at
3 // http://rust-lang.org/COPYRIGHT.
5 // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
6 // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
7 // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
8 // option. This file may not be copied, modified, or distributed
9 // except according to those terms.
11 #![allow(missing_docs)]
13 use std::cmp::Ordering::{mod, Less, Greater, Equal};
14 use std::collections::hash_map::Entry::{Occupied, Vacant};
15 use std::collections::hash_map;
20 use std::num::{Float, FloatMath, FromPrimitive};
22 fn local_cmp<T:Float>(x: T, y: T) -> Ordering {
23 // arbitrarily decide that NaNs are larger than everything.
26 } else if x.is_nan() {
37 fn local_sort<T: Float>(v: &mut [T]) {
38 v.sort_by(|x: &T, y: &T| local_cmp(*x, *y));
41 /// Trait that provides simple descriptive statistics on a univariate set of numeric samples.
42 pub trait Stats <T: FloatMath + FromPrimitive> for Sized? {
44 /// Sum of the samples.
46 /// Note: this method sacrifices performance at the altar of accuracy
47 /// Depends on IEEE-754 arithmetic guarantees. See proof of correctness at:
48 /// ["Adaptive Precision Floating-Point Arithmetic and Fast Robust Geometric Predicates"]
49 /// (http://www.cs.cmu.edu/~quake-papers/robust-arithmetic.ps)
50 /// *Discrete & Computational Geometry 18*, 3 (Oct 1997), 305-363, Shewchuk J.R.
53 /// Minimum value of the samples.
56 /// Maximum value of the samples.
59 /// Arithmetic mean (average) of the samples: sum divided by sample-count.
61 /// See: https://en.wikipedia.org/wiki/Arithmetic_mean
64 /// Median of the samples: value separating the lower half of the samples from the higher half.
65 /// Equal to `self.percentile(50.0)`.
67 /// See: https://en.wikipedia.org/wiki/Median
68 fn median(&self) -> T;
70 /// Variance of the samples: bias-corrected mean of the squares of the differences of each
71 /// sample from the sample mean. Note that this calculates the _sample variance_ rather than the
72 /// population variance, which is assumed to be unknown. It therefore corrects the `(n-1)/n`
73 /// bias that would appear if we calculated a population variance, by dividing by `(n-1)` rather
76 /// See: https://en.wikipedia.org/wiki/Variance
79 /// Standard deviation: the square root of the sample variance.
81 /// Note: this is not a robust statistic for non-normal distributions. Prefer the
82 /// `median_abs_dev` for unknown distributions.
84 /// See: https://en.wikipedia.org/wiki/Standard_deviation
85 fn std_dev(&self) -> T;
87 /// Standard deviation as a percent of the mean value. See `std_dev` and `mean`.
89 /// Note: this is not a robust statistic for non-normal distributions. Prefer the
90 /// `median_abs_dev_pct` for unknown distributions.
91 fn std_dev_pct(&self) -> T;
93 /// Scaled median of the absolute deviations of each sample from the sample median. This is a
94 /// robust (distribution-agnostic) estimator of sample variability. Use this in preference to
95 /// `std_dev` if you cannot assume your sample is normally distributed. Note that this is scaled
96 /// by the constant `1.4826` to allow its use as a consistent estimator for the standard
99 /// See: http://en.wikipedia.org/wiki/Median_absolute_deviation
100 fn median_abs_dev(&self) -> T;
102 /// Median absolute deviation as a percent of the median. See `median_abs_dev` and `median`.
103 fn median_abs_dev_pct(&self) -> T;
105 /// Percentile: the value below which `pct` percent of the values in `self` fall. For example,
106 /// percentile(95.0) will return the value `v` such that 95% of the samples `s` in `self`
107 /// satisfy `s <= v`.
109 /// Calculated by linear interpolation between closest ranks.
111 /// See: http://en.wikipedia.org/wiki/Percentile
112 fn percentile(&self, pct: T) -> T;
114 /// Quartiles of the sample: three values that divide the sample into four equal groups, each
115 /// with 1/4 of the data. The middle value is the median. See `median` and `percentile`. This
116 /// function may calculate the 3 quartiles more efficiently than 3 calls to `percentile`, but
117 /// is otherwise equivalent.
119 /// See also: https://en.wikipedia.org/wiki/Quartile
120 fn quartiles(&self) -> (T,T,T);
122 /// Inter-quartile range: the difference between the 25th percentile (1st quartile) and the 75th
123 /// percentile (3rd quartile). See `quartiles`.
125 /// See also: https://en.wikipedia.org/wiki/Interquartile_range
129 /// Extracted collection of all the summary statistics of a sample set.
130 #[deriving(Clone, PartialEq)]
131 #[allow(missing_docs)]
132 pub struct Summary<T> {
141 pub median_abs_dev: T,
142 pub median_abs_dev_pct: T,
143 pub quartiles: (T,T,T),
147 impl<T: FloatMath + FromPrimitive> Summary<T> {
148 /// Construct a new summary of a sample set.
149 pub fn new(samples: &[T]) -> Summary<T> {
154 mean: samples.mean(),
155 median: samples.median(),
157 std_dev: samples.std_dev(),
158 std_dev_pct: samples.std_dev_pct(),
159 median_abs_dev: samples.median_abs_dev(),
160 median_abs_dev_pct: samples.median_abs_dev_pct(),
161 quartiles: samples.quartiles(),
167 impl<T: FloatMath + FromPrimitive> Stats<T> for [T] {
168 // FIXME #11059 handle NaN, inf and overflow
170 let mut partials = vec![];
172 for &mut x in self.iter() {
174 // This inner loop applies `hi`/`lo` summation to each
175 // partial so that the list of partial sums remains exact.
176 for i in range(0, partials.len()) {
177 let mut y: T = partials[i];
178 if x.abs() < y.abs() {
179 mem::swap(&mut x, &mut y);
181 // Rounded `x+y` is stored in `hi` with round-off stored in
182 // `lo`. Together `hi+lo` are exactly equal to `x+y`.
184 let lo = y - (hi - x);
185 if lo != Float::zero() {
191 if j >= partials.len() {
195 partials.truncate(j+1);
198 let zero: T = Float::zero();
199 partials.iter().fold(zero, |p, q| p + *q)
203 assert!(self.len() != 0);
204 self.iter().fold(self[0], |p, q| p.min(*q))
208 assert!(self.len() != 0);
209 self.iter().fold(self[0], |p, q| p.max(*q))
212 fn mean(&self) -> T {
213 assert!(self.len() != 0);
214 self.sum() / FromPrimitive::from_uint(self.len()).unwrap()
217 fn median(&self) -> T {
218 self.percentile(FromPrimitive::from_uint(50).unwrap())
225 let mean = self.mean();
226 let mut v: T = Float::zero();
227 for s in self.iter() {
231 // NB: this is _supposed to be_ len-1, not len. If you
232 // change it back to len, you will be calculating a
233 // population variance, not a sample variance.
234 let denom = FromPrimitive::from_uint(self.len()-1).unwrap();
239 fn std_dev(&self) -> T {
243 fn std_dev_pct(&self) -> T {
244 let hundred = FromPrimitive::from_uint(100).unwrap();
245 (self.std_dev() / self.mean()) * hundred
248 fn median_abs_dev(&self) -> T {
249 let med = self.median();
250 let abs_devs: Vec<T> = self.iter().map(|&v| (med - v).abs()).collect();
251 // This constant is derived by smarter statistics brains than me, but it is
252 // consistent with how R and other packages treat the MAD.
253 let number = FromPrimitive::from_f64(1.4826).unwrap();
254 abs_devs.median() * number
257 fn median_abs_dev_pct(&self) -> T {
258 let hundred = FromPrimitive::from_uint(100).unwrap();
259 (self.median_abs_dev() / self.median()) * hundred
262 fn percentile(&self, pct: T) -> T {
263 let mut tmp = self.to_vec();
264 local_sort(tmp.as_mut_slice());
265 percentile_of_sorted(tmp.as_slice(), pct)
268 fn quartiles(&self) -> (T,T,T) {
269 let mut tmp = self.to_vec();
270 local_sort(tmp.as_mut_slice());
271 let first = FromPrimitive::from_uint(25).unwrap();
272 let a = percentile_of_sorted(tmp.as_slice(), first);
273 let secound = FromPrimitive::from_uint(50).unwrap();
274 let b = percentile_of_sorted(tmp.as_slice(), secound);
275 let third = FromPrimitive::from_uint(75).unwrap();
276 let c = percentile_of_sorted(tmp.as_slice(), third);
281 let (a,_,c) = self.quartiles();
287 // Helper function: extract a value representing the `pct` percentile of a sorted sample-set, using
288 // linear interpolation. If samples are not sorted, return nonsensical value.
289 fn percentile_of_sorted<T: Float + FromPrimitive>(sorted_samples: &[T],
291 assert!(sorted_samples.len() != 0);
292 if sorted_samples.len() == 1 {
293 return sorted_samples[0];
295 let zero: T = Float::zero();
296 assert!(zero <= pct);
297 let hundred = FromPrimitive::from_uint(100).unwrap();
298 assert!(pct <= hundred);
300 return sorted_samples[sorted_samples.len() - 1];
302 let length = FromPrimitive::from_uint(sorted_samples.len() - 1).unwrap();
303 let rank = (pct / hundred) * length;
304 let lrank = rank.floor();
305 let d = rank - lrank;
306 let n = lrank.to_uint().unwrap();
307 let lo = sorted_samples[n];
308 let hi = sorted_samples[n+1];
313 /// Winsorize a set of samples, replacing values above the `100-pct` percentile and below the `pct`
314 /// percentile with those percentiles themselves. This is a way of minimizing the effect of
315 /// outliers, at the cost of biasing the sample. It differs from trimming in that it does not
316 /// change the number of samples, just changes the values of those that are outliers.
318 /// See: http://en.wikipedia.org/wiki/Winsorising
319 pub fn winsorize<T: Float + FromPrimitive>(samples: &mut [T], pct: T) {
320 let mut tmp = samples.to_vec();
321 local_sort(tmp.as_mut_slice());
322 let lo = percentile_of_sorted(tmp.as_slice(), pct);
323 let hundred: T = FromPrimitive::from_uint(100).unwrap();
324 let hi = percentile_of_sorted(tmp.as_slice(), hundred-pct);
325 for samp in samples.iter_mut() {
328 } else if *samp < lo {
334 /// Render writes the min, max and quartiles of the provided `Summary` to the provided `Writer`.
335 pub fn write_5_number_summary<W: Writer, T: Float + Show>(w: &mut W,
336 s: &Summary<T>) -> io::IoResult<()> {
337 let (q1,q2,q3) = s.quartiles;
338 write!(w, "(min={}, q1={}, med={}, q3={}, max={})",
346 /// Render a boxplot to the provided writer. The boxplot shows the min, max and quartiles of the
347 /// provided `Summary` (thus includes the mean) and is scaled to display within the range of the
348 /// nearest multiple-of-a-power-of-ten above and below the min and max of possible values, and
349 /// target `width_hint` characters of display (though it will be wider if necessary).
351 /// As an example, the summary with 5-number-summary `(min=15, q1=17, med=20, q3=24, max=31)` might
355 /// 10 | [--****#******----------] | 40
357 pub fn write_boxplot<W: Writer, T: Float + Show + FromPrimitive>(
361 -> io::IoResult<()> {
363 let (q1,q2,q3) = s.quartiles;
365 // the .abs() handles the case where numbers are negative
366 let ten: T = FromPrimitive::from_uint(10).unwrap();
367 let lomag = ten.powf(s.min.abs().log10().floor());
368 let himag = ten.powf(s.max.abs().log10().floor());
370 // need to consider when the limit is zero
371 let zero: T = Float::zero();
372 let lo = if lomag == Float::zero() {
375 (s.min / lomag).floor() * lomag
378 let hi = if himag == Float::zero() {
381 (s.max / himag).ceil() * himag
386 let lostr = lo.to_string();
387 let histr = hi.to_string();
389 let overhead_width = lostr.len() + histr.len() + 4;
390 let range_width = width_hint - overhead_width;
391 let range_float = FromPrimitive::from_uint(range_width).unwrap();
392 let char_step = range / range_float;
394 try!(write!(w, "{} |", lostr));
399 while c < range_width && v < s.min {
400 try!(write!(w, " "));
404 try!(write!(w, "["));
406 while c < range_width && v < q1 {
407 try!(write!(w, "-"));
411 while c < range_width && v < q2 {
412 try!(write!(w, "*"));
416 try!(write!(w, "#"));
418 while c < range_width && v < q3 {
419 try!(write!(w, "*"));
423 while c < range_width && v < s.max {
424 try!(write!(w, "-"));
428 try!(write!(w, "]"));
429 while c < range_width {
430 try!(write!(w, " "));
435 try!(write!(w, "| {}", histr));
439 /// Returns a HashMap with the number of occurrences of every element in the
440 /// sequence that the iterator exposes.
441 pub fn freq_count<T: Iterator<Item=U>, U: Eq+Hash>(mut iter: T) -> hash_map::HashMap<U, uint> {
442 let mut map: hash_map::HashMap<U,uint> = hash_map::HashMap::new();
444 match map.entry(elem) {
445 Occupied(mut entry) => { *entry.get_mut() += 1; },
446 Vacant(entry) => { entry.set(1); },
452 // Test vectors generated from R, using the script src/etc/stat-test-vectors.r.
458 use stats::write_5_number_summary;
459 use stats::write_boxplot;
463 macro_rules! assert_approx_eq {
464 ($a:expr, $b:expr) => ({
466 let (a, b) = (&$a, &$b);
467 assert!((*a - *b).abs() < 1.0e-6,
468 "{} is not approximately equal to {}", *a, *b);
472 fn check(samples: &[f64], summ: &Summary<f64>) {
474 let summ2 = Summary::new(samples);
476 let mut w = io::stdout();
478 (write!(w, "\n")).unwrap();
479 write_5_number_summary(w, &summ2).unwrap();
480 (write!(w, "\n")).unwrap();
481 write_boxplot(w, &summ2, 50).unwrap();
482 (write!(w, "\n")).unwrap();
484 assert_eq!(summ.sum, summ2.sum);
485 assert_eq!(summ.min, summ2.min);
486 assert_eq!(summ.max, summ2.max);
487 assert_eq!(summ.mean, summ2.mean);
488 assert_eq!(summ.median, summ2.median);
490 // We needed a few more digits to get exact equality on these
491 // but they're within float epsilon, which is 1.0e-6.
492 assert_approx_eq!(summ.var, summ2.var);
493 assert_approx_eq!(summ.std_dev, summ2.std_dev);
494 assert_approx_eq!(summ.std_dev_pct, summ2.std_dev_pct);
495 assert_approx_eq!(summ.median_abs_dev, summ2.median_abs_dev);
496 assert_approx_eq!(summ.median_abs_dev_pct, summ2.median_abs_dev_pct);
498 assert_eq!(summ.quartiles, summ2.quartiles);
499 assert_eq!(summ.iqr, summ2.iqr);
503 fn test_min_max_nan() {
504 let xs = &[1.0, 2.0, f64::NAN, 3.0, 4.0];
505 let summary = Summary::new(xs);
506 assert_eq!(summary.min, 1.0);
507 assert_eq!(summary.max, 4.0);
516 let summ = &Summary {
517 sum: 1882.0000000000,
520 mean: 941.0000000000,
521 median: 941.0000000000,
523 std_dev: 24.0416305603,
524 std_dev_pct: 2.5549022912,
525 median_abs_dev: 25.2042000000,
526 median_abs_dev_pct: 2.6784484591,
527 quartiles: (932.5000000000,941.0000000000,949.5000000000),
533 fn test_norm10narrow() {
546 let summ = &Summary {
547 sum: 9996.0000000000,
549 max: 1217.0000000000,
550 mean: 999.6000000000,
551 median: 970.5000000000,
552 var: 16050.7111111111,
553 std_dev: 126.6914010938,
554 std_dev_pct: 12.6742097933,
555 median_abs_dev: 102.2994000000,
556 median_abs_dev_pct: 10.5408964451,
557 quartiles: (956.7500000000,970.5000000000,1078.7500000000),
563 fn test_norm10medium() {
576 let summ = &Summary {
577 sum: 8653.0000000000,
579 max: 1084.0000000000,
580 mean: 865.3000000000,
581 median: 911.5000000000,
582 var: 48628.4555555556,
583 std_dev: 220.5186059170,
584 std_dev_pct: 25.4846418487,
585 median_abs_dev: 195.7032000000,
586 median_abs_dev_pct: 21.4704552935,
587 quartiles: (771.0000000000,911.5000000000,1017.2500000000),
593 fn test_norm10wide() {
606 let summ = &Summary {
607 sum: 9349.0000000000,
609 max: 1591.0000000000,
610 mean: 934.9000000000,
611 median: 913.5000000000,
612 var: 239208.9888888889,
613 std_dev: 489.0899599142,
614 std_dev_pct: 52.3146817750,
615 median_abs_dev: 611.5725000000,
616 median_abs_dev_pct: 66.9482758621,
617 quartiles: (567.2500000000,913.5000000000,1331.2500000000),
623 fn test_norm25verynarrow() {
651 let summ = &Summary {
652 sum: 24926.0000000000,
654 max: 1040.0000000000,
655 mean: 997.0400000000,
656 median: 998.0000000000,
658 std_dev: 19.8294393937,
659 std_dev_pct: 1.9888308788,
660 median_abs_dev: 22.2390000000,
661 median_abs_dev_pct: 2.2283567134,
662 quartiles: (983.0000000000,998.0000000000,1013.0000000000),
681 let summ = &Summary {
686 median: 11.5000000000,
688 std_dev: 16.9643416875,
689 std_dev_pct: 101.5828843560,
690 median_abs_dev: 13.3434000000,
691 median_abs_dev_pct: 116.0295652174,
692 quartiles: (4.2500000000,11.5000000000,22.5000000000),
711 let summ = &Summary {
716 median: 24.5000000000,
718 std_dev: 19.5848580967,
719 std_dev_pct: 74.4671410520,
720 median_abs_dev: 22.9803000000,
721 median_abs_dev_pct: 93.7971428571,
722 quartiles: (9.5000000000,24.5000000000,36.5000000000),
741 let summ = &Summary {
746 median: 22.0000000000,
748 std_dev: 21.4050876611,
749 std_dev_pct: 88.4507754589,
750 median_abs_dev: 21.4977000000,
751 median_abs_dev_pct: 97.7168181818,
752 quartiles: (7.7500000000,22.0000000000,35.0000000000),
786 let summ = &Summary {
791 median: 19.0000000000,
793 std_dev: 24.5161851301,
794 std_dev_pct: 103.3565983562,
795 median_abs_dev: 19.2738000000,
796 median_abs_dev_pct: 101.4410526316,
797 quartiles: (6.0000000000,19.0000000000,31.0000000000),
831 let summ = &Summary {
836 median: 20.0000000000,
838 std_dev: 4.5650848842,
839 std_dev_pct: 22.2037202539,
840 median_abs_dev: 5.9304000000,
841 median_abs_dev_pct: 29.6520000000,
842 quartiles: (17.0000000000,20.0000000000,24.0000000000),
848 fn test_pois25lambda30() {
876 let summ = &Summary {
881 median: 32.0000000000,
883 std_dev: 5.1568724372,
884 std_dev_pct: 16.3814245145,
885 median_abs_dev: 5.9304000000,
886 median_abs_dev_pct: 18.5325000000,
887 quartiles: (28.0000000000,32.0000000000,34.0000000000),
893 fn test_pois25lambda40() {
921 let summ = &Summary {
922 sum: 1019.0000000000,
926 median: 42.0000000000,
928 std_dev: 5.8685603004,
929 std_dev_pct: 14.3978417577,
930 median_abs_dev: 5.9304000000,
931 median_abs_dev_pct: 14.1200000000,
932 quartiles: (37.0000000000,42.0000000000,45.0000000000),
938 fn test_pois25lambda50() {
966 let summ = &Summary {
967 sum: 1235.0000000000,
971 median: 50.0000000000,
973 std_dev: 5.6273143387,
974 std_dev_pct: 11.3913245723,
975 median_abs_dev: 4.4478000000,
976 median_abs_dev_pct: 8.8956000000,
977 quartiles: (44.0000000000,50.0000000000,52.0000000000),
1011 let summ = &Summary {
1012 sum: 1242.0000000000,
1015 mean: 49.6800000000,
1016 median: 45.0000000000,
1017 var: 1015.6433333333,
1018 std_dev: 31.8691595957,
1019 std_dev_pct: 64.1488719719,
1020 median_abs_dev: 45.9606000000,
1021 median_abs_dev_pct: 102.1346666667,
1022 quartiles: (29.0000000000,45.0000000000,79.0000000000),
1029 fn test_boxplot_nonpositive() {
1030 fn t(s: &Summary<f64>, expected: String) {
1031 let mut m = Vec::new();
1032 write_boxplot(&mut m, s, 30).unwrap();
1033 let out = String::from_utf8(m).unwrap();
1034 assert_eq!(out, expected);
1037 t(&Summary::new(&[-2.0f64, -1.0f64]),
1038 "-2 |[------******#*****---]| -1".to_string());
1039 t(&Summary::new(&[0.0f64, 2.0f64]),
1040 "0 |[-------*****#*******---]| 2".to_string());
1041 t(&Summary::new(&[-2.0f64, 0.0f64]),
1042 "-2 |[------******#******---]| 0".to_string());
1046 fn test_sum_f64s() {
1047 assert_eq!([0.5f64, 3.2321f64, 1.5678f64].sum(), 5.2999);
1050 fn test_sum_f64_between_ints_that_sum_to_0() {
1051 assert_eq!([1e30f64, 1.2f64, -1e30f64].sum(), 1.2);
1061 pub fn sum_three_items(b: &mut Bencher) {
1063 [1e20f64, 1.5f64, -1e20f64].sum();
1067 pub fn sum_many_f64(b: &mut Bencher) {
1068 let nums = [-1e30f64, 1e60, 1e30, 1.0, -1e60];
1069 let v = Vec::from_fn(500, |i| nums[i%5]);