A weighted random sampling crate using Walker's Alias Method.
Walker's Alias Method (WAM) is one method for performing weighted random sampling.
WAM weights each index of a array by giving two pieces of information: an alias to a different index and a probability to decide whether to jump to that index.
WAM is a very fast algorithm, and its computational complexity of the search is O(1).
The difference in complexity between WAM and the Cumulative Sum Method is as follows.
| Algorithm | Building table | Search |
|---|---|---|
| Walker's Alias Method | O(N) | O(1) |
| Cumulative Sum Method | O(N) | O(log N) |
The API documentation is here.
Add this to your Cargo.toml:
[dependencies]
weighted_rand = "0.4"use weighted_rand::builder::WalkerTableBuilder;
fn main() {
let fruit = ["Apple", "Banana", "Orange", "Peach"];
// Define the weights for each index corresponding
// to the above list.
// In the following case, the ratio of each weight
// is "2 : 1 : 7 : 0", and the output probabilities
// for each index are 0.2, 0.1, 0.7 and 0.
let index_weights = [2, 1, 7, 0];
let builder = WalkerTableBuilder::new(&index_weights);
let wa_table = builder.build();
for i in (0..10).map(|_| wa_table.next()) {
println!("{}", fruit[i]);
}
}Also, index_weiaghts supports &[f32], like:
use rand;
use weighted_rand::builder::*;
fn main() {
// Coin with a 5% higher probability of heads than tails
let cheating_coin = ["Heads!", "Tails!"];
let index_weights = [0.55, 0.45];
let builder = WalkerTableBuilder::new(&index_weights);
let wa_table = builder.build();
// If you want to process something in a large number of
// loops, we recommend using the next_rng method with an
// external ThreadRng instance.
let mut result = [""; 10000];
let mut rng = rand::thread_rng();
for r in &mut result {
let j = wa_table.next_rng(&mut rng);
*r = cheating_coin[j];
}
// println!("{:?}", result);
}Licensed under either of
- Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)
at your option.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.