| 1 | //! Univariate analysis |
| 2 | |
| 3 | mod bootstrap; |
| 4 | mod percentiles; |
| 5 | mod resamples; |
| 6 | mod sample; |
| 7 | |
| 8 | pub mod kde; |
| 9 | pub mod mixed; |
| 10 | pub mod outliers; |
| 11 | |
| 12 | use crate::stats::float::Float; |
| 13 | use crate::stats::tuple::{Tuple, TupledDistributionsBuilder}; |
| 14 | #[cfg (feature = "rayon" )] |
| 15 | use rayon::prelude::*; |
| 16 | use std::cmp; |
| 17 | |
| 18 | use self::resamples::Resamples; |
| 19 | |
| 20 | pub use self::percentiles::Percentiles; |
| 21 | pub use self::sample::Sample; |
| 22 | |
| 23 | /// Performs a two-sample bootstrap |
| 24 | /// |
| 25 | /// - Multithreaded |
| 26 | /// - Time: `O(nresamples)` |
| 27 | /// - Memory: `O(nresamples)` |
| 28 | #[cfg_attr (feature = "cargo-clippy" , allow(clippy::cast_lossless))] |
| 29 | pub fn bootstrap<A, B, T, S>( |
| 30 | a: &Sample<A>, |
| 31 | b: &Sample<B>, |
| 32 | nresamples: usize, |
| 33 | statistic: S, |
| 34 | ) -> T::Distributions |
| 35 | where |
| 36 | A: Float, |
| 37 | B: Float, |
| 38 | S: Fn(&Sample<A>, &Sample<B>) -> T + Sync, |
| 39 | T: Tuple + Send, |
| 40 | T::Distributions: Send, |
| 41 | T::Builder: Send, |
| 42 | { |
| 43 | let nresamples_sqrt = (nresamples as f64).sqrt().ceil() as usize; |
| 44 | let per_chunk = (nresamples + nresamples_sqrt - 1) / nresamples_sqrt; |
| 45 | |
| 46 | #[cfg (feature = "rayon" )] |
| 47 | { |
| 48 | (0..nresamples_sqrt) |
| 49 | .into_par_iter() |
| 50 | .map_init( |
| 51 | || (Resamples::new(a), Resamples::new(b)), |
| 52 | |(a_resamples, b_resamples), i| { |
| 53 | let start = i * per_chunk; |
| 54 | let end = cmp::min((i + 1) * per_chunk, nresamples); |
| 55 | let a_resample = a_resamples.next(); |
| 56 | |
| 57 | let mut sub_distributions: T::Builder = |
| 58 | TupledDistributionsBuilder::new(end - start); |
| 59 | |
| 60 | for _ in start..end { |
| 61 | let b_resample = b_resamples.next(); |
| 62 | sub_distributions.push(statistic(a_resample, b_resample)); |
| 63 | } |
| 64 | sub_distributions |
| 65 | }, |
| 66 | ) |
| 67 | .reduce( |
| 68 | || T::Builder::new(0), |
| 69 | |mut a, mut b| { |
| 70 | a.extend(&mut b); |
| 71 | a |
| 72 | }, |
| 73 | ) |
| 74 | .complete() |
| 75 | } |
| 76 | #[cfg (not(feature = "rayon" ))] |
| 77 | { |
| 78 | let mut a_resamples = Resamples::new(a); |
| 79 | let mut b_resamples = Resamples::new(b); |
| 80 | (0..nresamples_sqrt) |
| 81 | .map(|i| { |
| 82 | let start = i * per_chunk; |
| 83 | let end = cmp::min((i + 1) * per_chunk, nresamples); |
| 84 | let a_resample = a_resamples.next(); |
| 85 | |
| 86 | let mut sub_distributions: T::Builder = |
| 87 | TupledDistributionsBuilder::new(end - start); |
| 88 | |
| 89 | for _ in start..end { |
| 90 | let b_resample = b_resamples.next(); |
| 91 | sub_distributions.push(statistic(a_resample, b_resample)); |
| 92 | } |
| 93 | sub_distributions |
| 94 | }) |
| 95 | .fold(T::Builder::new(0), |mut a, mut b| { |
| 96 | a.extend(&mut b); |
| 97 | a |
| 98 | }) |
| 99 | .complete() |
| 100 | } |
| 101 | } |
| 102 | |