aprender-core 0.29.3

Next-generation machine learning library in pure Rust
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

//! Differential Evolution (DE) optimizer.
//!
//! A population-based metaheuristic for continuous optimization.
//!
//! # Algorithm
//!
//! DE evolves a population of candidate solutions through mutation based on
//! difference vectors between population members:
//!
//! ```text
//! For each target vector xᵢ:
//!   1. Select 3 distinct random vectors xₐ, xᵦ, xᵧ
//!   2. Mutant: v = xₐ + F·(xᵦ - xᵧ)
//!   3. Crossover: uⱼ = vⱼ if rand() < CR else xᵢⱼ
//!   4. Selection: xᵢ' = u if f(u) < f(xᵢ) else xᵢ
//! ```
//!
//! # References
//!
//! - Storn & Price (1997): "Differential Evolution - A Simple and Efficient
//!   Heuristic for Global Optimization over Continuous Spaces"
//! - Zhang & Sanderson (2009): "JADE: Adaptive Differential Evolution"
//! - Tanabe & Fukunaga (2013): "SHADE: Success-History Based Adaptation"

use rand::prelude::*;
use serde::{Deserialize, Serialize};

use super::{
    budget::ConvergenceTracker, Budget, OptimizationResult, PerturbativeMetaheuristic, SearchSpace,
};
use crate::metaheuristics::traits::TerminationReason;

/// Differential Evolution optimizer.
///
/// # Example
///
/// ```
/// use aprender::metaheuristics::{DifferentialEvolution, SearchSpace, Budget, PerturbativeMetaheuristic};
///
/// // Sphere function: f(x) = Σxᵢ²
/// let objective = |x: &[f64]| x.iter().map(|xi| xi * xi).sum();
///
/// let mut de = DifferentialEvolution::default();
/// let space = SearchSpace::continuous(5, -5.0, 5.0);  // 5D is easier
/// let result = de.optimize(&objective, &space, Budget::Evaluations(10_000));
///
/// assert!(result.objective_value < 1.0);  // Converges to near-zero
/// ```
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct DifferentialEvolution {
    /// Population size (default: 10 * dimension)
    pub population_size: usize,

    /// Mutation factor F ∈ [0, 2] (default: 0.8)
    pub mutation_factor: f64,

    /// Crossover rate CR ∈ [0, 1] (default: 0.9)
    pub crossover_rate: f64,

    /// Mutation strategy
    pub strategy: DEStrategy,

    /// Parameter adaptation strategy
    pub adaptation: AdaptationStrategy,

    /// Random seed for reproducibility
    #[serde(default)]
    seed: Option<u64>,

    // Internal state (not serialized)
    #[serde(skip)]
    population: Vec<Vec<f64>>,
    #[serde(skip)]
    fitness: Vec<f64>,
    #[serde(skip)]
    best_idx: usize,
    #[serde(skip)]
    history: Vec<f64>,
}

/// DE mutation strategy.
///
/// Different strategies offer trade-offs between exploration and exploitation.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Default)]
pub enum DEStrategy {
    /// DE/rand/1/bin: v = xₐ + F·(xᵦ - xᵧ)
    ///
    /// Good exploration, slower convergence. Default choice.
    #[default]
    Rand1Bin,

    /// DE/best/1/bin: v = `x_best` + F·(xₐ - xᵦ)
    ///
    /// Fast convergence, may premature converge.
    Best1Bin,

    /// DE/rand/2/bin: v = xₐ + F·(xᵦ - xᵧ) + F·(xδ - xε)
    ///
    /// More exploration, uses 5 random vectors.
    Rand2Bin,

    /// DE/current-to-best/1/bin: v = xᵢ + `F·(x_best` - xᵢ) + F·(xₐ - xᵦ)
    ///
    /// Balance between rand and best.
    CurrentToBest1Bin,
}

/// Adaptation strategy for self-adaptive DE variants.
#[derive(Debug, Clone, Serialize, Deserialize, Default)]
pub enum AdaptationStrategy {
    /// Fixed F and CR (original DE)
    #[default]
    None,

    /// JADE: Adaptive F and CR with external archive
    ///
    /// Reference: Zhang & Sanderson (2009)
    JADE {
        /// External archive of inferior solutions
        archive: Vec<Vec<f64>>,
        /// Maximum archive size (typically = `population_size`)
        archive_size: usize,
        /// Location parameter for F (Cauchy distribution)
        mu_f: f64,
        /// Location parameter for CR (Normal distribution)
        mu_cr: f64,
        /// Learning rate for adaptation (typically 0.1)
        c: f64,
    },

    /// SHADE: Success-History Based Adaptation
    ///
    /// Reference: Tanabe & Fukunaga (2013)
    SHADE {
        /// Historical successful F values
        memory_f: Vec<f64>,
        /// Historical successful CR values
        memory_cr: Vec<f64>,
        /// Memory size H (typically 5-10)
        memory_size: usize,
        /// Current position in circular buffer
        memory_index: usize,
    },
}

include!("de_impl.rs");
include!("de_optimize.rs");

#[cfg(test)]
#[path = "tests_de_contract.rs"]
mod tests_de_contract;