simplers_optimization 0.5.0

A Rust implementation of the Simple(x) black-box optimization algorithm.
Documentation 
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

use crate::point::*;
use ordered_float::OrderedFloat;
use num_traits::{Float, float::FloatCore};

/// encapsulate a function and its domain of definition
pub struct SearchSpace<F, CoordFloat, ValueFloat>
    where F: FnMut(&[CoordFloat]) -> ValueFloat,
          CoordFloat: Float + FloatCore,
          ValueFloat: Float
{
    f: F,
    hypercube: Vec<(CoordFloat, CoordFloat)>,
    pub minimize: bool,
    pub dimension: usize,
    /// stores all evaluated points in evaluation order as (coordinates_in_input_space, value)
    pub history: Vec<(Coordinates<CoordFloat>, ValueFloat)>
}

impl<F, CoordFloat, ValueFloat> SearchSpace<F, CoordFloat, ValueFloat>
    where F: FnMut(&[CoordFloat]) -> ValueFloat,
          CoordFloat: Float + FloatCore,
          ValueFloat: Float
{
    /// builds a new search space that encapsulate both the function to evaluate and its domain of definition
    pub fn new(f: F, hypercube: &[(CoordFloat, CoordFloat)], minimize: bool) -> Self
    {
        let dimension = hypercube.len();
        let history = Vec::new();
        SearchSpace { f, hypercube: hypercube.to_vec(), minimize, dimension, history }
    }

    /// Converts coordinates from the hypercube to the unit simplex
    /// This function is useful when one wants to suggest a point to the algorithm
    /// for the formula used, see: https://math.stackexchange.com/a/385071/495073
    pub fn to_simplex(&self, c: &[CoordFloat]) -> Coordinates<CoordFloat>
    {
        // goes to the unit hypercube
        let c: Coordinates<CoordFloat> =
            c.iter().zip(self.hypercube.iter()).map(|(&x, &(inf, sup))| (x - inf) / (sup - inf)).collect();
        // goes to the unit simplex
        let sum = c.iter().copied().fold(CoordFloat::zero(), ::std::ops::Add::add); // sum
        let max = c.iter()
                   .copied()
                   .max_by_key(|&c| OrderedFloat(c))
                   .expect("You should have at least one coordinate.");
        let ratio = if sum.is_zero() { CoordFloat::zero() } else { max / sum };
        c.iter().map(|&x| x * ratio).collect()
    }

    /// converts coordinates from the unit simplex to the hypercube
    /// formula deduced from: https://math.stackexchange.com/a/385071/495073
    pub fn to_hypercube(&self, c: &[CoordFloat]) -> Coordinates<CoordFloat>
    {
        // gets the ratio to go from the unit hypercube to the unit simplex
        let sum = c.iter().copied().fold(CoordFloat::zero(), ::std::ops::Add::add); // sum
        let max = c.iter()
                   .copied()
                   .max_by_key(|&c| OrderedFloat(c))
                   .expect("You should have at least one coordinate.");
        let ratio = if max.is_zero() { CoordFloat::zero() } else { sum / max };
        // goes from the simplex to the target hypercube
        c.iter().zip(self.hypercube.iter()).map(|(&x, &(inf, sup))| inf + x * ratio * (sup - inf)).collect()
    }

    /// takes coordinates in the unit simplex and evaluate them with the function
    pub fn evaluate(&mut self, c: &Coordinates<CoordFloat>) -> ValueFloat
    {
        let c_hypercube = self.to_hypercube(c);
        let evaluation = (self.f)(&c_hypercube);
        self.history.push((c_hypercube, evaluation));
        if self.minimize { -evaluation } else { evaluation }
    }
}