scirs2_optimize/reinforcement_learning/
evolutionary_strategies.rs

1//! Evolutionary Strategies for RL Optimization
2//!
3//! Population-based reinforcement learning optimization methods.
4
5use crate::error::OptimizeResult;
6use crate::result::OptimizeResults;
7use scirs2_core::ndarray::{Array1, ArrayView1};
8use scirs2_core::random::{rng, Rng};
9// Unused import
10// use scirs2_core::error::CoreResult;
11
12/// Evolutionary strategy optimizer
13#[derive(Debug, Clone)]
14pub struct EvolutionaryStrategy {
15    /// Population size
16    pub population_size: usize,
17    /// Current population
18    pub population: Vec<Array1<f64>>,
19    /// Population fitness
20    pub fitness: Vec<f64>,
21    /// Mutation strength
22    pub sigma: f64,
23}
24
25impl EvolutionaryStrategy {
26    /// Create new evolutionary strategy
27    pub fn new(population_size: usize, dimensions: usize, sigma: f64) -> Self {
28        let mut population = Vec::with_capacity(population_size);
29        for _ in 0..population_size {
30            let individual = Array1::from_shape_fn(dimensions, |_| {
31                scirs2_core::random::rng().random::<f64>() - 0.5
32            });
33            population.push(individual);
34        }
35
36        Self {
37            population_size,
38            population,
39            fitness: vec![f64::INFINITY; population_size],
40            sigma,
41        }
42    }
43
44    /// Evaluate population
45    pub fn evaluate<F>(&mut self, objective: &F)
46    where
47        F: Fn(&ArrayView1<f64>) -> f64,
48    {
49        for (i, individual) in self.population.iter().enumerate() {
50            self.fitness[i] = objective(&individual.view());
51        }
52    }
53
54    /// Evolve population
55    pub fn evolve(&mut self) {
56        // Select best half
57        let mut indices: Vec<usize> = (0..self.population_size).collect();
58        indices.sort_by(|&a, &b| self.fitness[a].partial_cmp(&self.fitness[b]).unwrap());
59
60        let elite_size = self.population_size / 2;
61
62        // Generate new population
63        for i in elite_size..self.population_size {
64            let parent_idx = indices[scirs2_core::random::rng().random_range(0..elite_size)];
65            let parent = &self.population[parent_idx];
66
67            // Mutate
68            let mut offspring = parent.clone();
69            for j in 0..offspring.len() {
70                offspring[j] += self.sigma * (scirs2_core::random::rng().random_range(-0.5..0.5));
71            }
72
73            self.population[i] = offspring;
74        }
75    }
76
77    /// Get best individual
78    pub fn get_best(&self) -> (Array1<f64>, f64) {
79        let mut best_idx = 0;
80        let mut best_fitness = self.fitness[0];
81
82        for (i, &fitness) in self.fitness.iter().enumerate() {
83            if fitness < best_fitness {
84                best_fitness = fitness;
85                best_idx = i;
86            }
87        }
88
89        (self.population[best_idx].clone(), best_fitness)
90    }
91}
92
93/// Evolutionary strategy optimization
94#[allow(dead_code)]
95pub fn evolutionary_optimize<F>(
96    objective: F,
97    initial_params: &ArrayView1<f64>,
98    num_generations: usize,
99) -> OptimizeResult<OptimizeResults<f64>>
100where
101    F: Fn(&ArrayView1<f64>) -> f64,
102{
103    let mut es = EvolutionaryStrategy::new(50, initial_params.len(), 0.1);
104
105    // Initialize with initial _params
106    es.population[0] = initial_params.to_owned();
107
108    for _generation in 0..num_generations {
109        es.evaluate(&objective);
110        es.evolve();
111    }
112
113    let (best_params, best_fitness) = es.get_best();
114
115    Ok(OptimizeResults::<f64> {
116        x: best_params,
117        fun: best_fitness,
118        success: true,
119        nit: num_generations,
120        message: "Evolutionary strategy completed".to_string(),
121        jac: None,
122        hess: None,
123        constr: None,
124        nfev: num_generations * 50, // Population size * _generations
125        njev: 0,
126        nhev: 0,
127        maxcv: 0,
128        status: 0,
129    })
130}
131
132#[allow(dead_code)]
133pub fn placeholder() {}
scirs2_optimize/reinforcement_learning/evolutionary_strategies.rs

scirs2_optimize/reinforcement_learning/
evolutionary_strategies.rs
