Implement genetic algorithm for optimizing continuous functions

genetic_algorithm/genetic_algorithm_optimization.py

@@ -0,0 +1,222 @@
+import random
+import numpy as np
+from concurrent.futures import ThreadPoolExecutor
+
+
+# Parameters
+N_POPULATION = 100  # Population size
+N_GENERATIONS = 500  # Maximum number of generations
+N_SELECTED = 50  # Number of parents selected for the next generation
+MUTATION_PROBABILITY = 0.1  # Mutation probability
+CROSSOVER_RATE = 0.8  # Probability of crossover
+SEARCH_SPACE = (-10, 10)  # Search space for the variables
+
+
+# Random number generator
+rng = np.random.default_rng()
+
+
+class GeneticAlgorithm:
+    def __init__(
+        self,
+        function: callable,
+        bounds: list[tuple[float, float]],
+        population_size: int,
+        generations: int,
+        mutation_prob: float,
+        crossover_rate: float,
+        maximize: bool = True
+    ) -> None:
+        self.function = function  # Target function to optimize
+        self.bounds = bounds  # Search space bounds (for each variable)
+        self.population_size = population_size
+        self.generations = generations
+        self.mutation_prob = mutation_prob
+        self.crossover_rate = crossover_rate
+        self.maximize = maximize
+        self.dim = len(bounds)  # Dimensionality of the function (number of variables)
+
+        # Initialize population
+        self.population = self.initialize_population()
+
+    def initialize_population(self) -> list[np.ndarray]:
+        """
+        Initialize the population with random individuals within the search space.
+
+        Returns:
+            list[np.ndarray]: A list of individuals represented as numpy arrays.
+
+        Example:
+        >>> ga = GeneticAlgorithm(lambda x, y: x**2 + y**2, [(-10, 10), (-10, 10)], 10, 100, 0.1, 0.8, False)
+        >>> len(ga.initialize_population()) == ga.population_size
+        True
+        """
+        return [
+            rng.uniform(low=self.bounds[i][0], high=self.bounds[i][1], size=self.dim)
+            for i in range(self.population_size)
+        ]
+
+    def fitness(self, individual: np.ndarray) -> float:
+        """
+        Calculate the fitness value (function value) for an individual.
+
+        Args:
+            individual (np.ndarray): The individual to evaluate.
+
+        Returns:
+            float: The fitness value of the individual.
+
+        Example:
+        >>> ga = GeneticAlgorithm(lambda x, y: -(x**2 + y**2), [(-10, 10), (-10, 10)], 10, 100, 0.1, 0.8, True)
+        >>> ind = np.array([1, 2])
+        >>> ga.fitness(ind)
+        -5.0
+        """
+        value = self.function(*individual)
+        return value if self.maximize else -value  # If minimizing, invert the fitness
+
+    def select_parents(self, population_score: list[tuple[np.ndarray, float]]) -> list[np.ndarray]:
+        """
+        Select top N_SELECTED parents based on fitness.
+
+        Args:
+            population_score (list[tuple[np.ndarray, float]]): The population with their respective fitness scores.
+
+        Returns:
+            list[np.ndarray]: The selected parents for the next generation.
+
+        Example:
+        >>> ga = GeneticAlgorithm(lambda x, y: -(x**2 + y**2), [(-10, 10), (-10, 10)], 10, 100, 0.1, 0.8, True)
+        >>> pop_score = [(np.array([1, 2]), -5), (np.array([3, 4]), -25)]
+        >>> len(ga.select_parents(pop_score)) == N_SELECTED
+        True
+        """
+        population_score.sort(key=lambda x: x[1], reverse=True)
+        return [ind for ind, _ in population_score[:N_SELECTED]]
+
+    def crossover(self, parent1: np.ndarray, parent2: np.ndarray) -> tuple[np.ndarray, np.ndarray]:
+        """
+        Perform uniform crossover between two parents to generate offspring.
+
+        Args:
+            parent1 (np.ndarray): The first parent.
+            parent2 (np.ndarray): The second parent.
+
+        Returns:
+            tuple[np.ndarray, np.ndarray]: The two offspring generated by crossover.
+
+        Example:
+        >>> ga = GeneticAlgorithm(lambda x, y: -(x**2 + y**2), [(-10, 10), (-10, 10)], 10, 100, 0.1, 0.8, True)
+        >>> parent1, parent2 = np.array([1, 2]), np.array([3, 4])
+        >>> len(ga.crossover(parent1, parent2)) == 2
+        True
+        """
+        if random.random() < self.crossover_rate:
+            cross_point = random.randint(1, self.dim - 1)
+            child1 = np.concatenate((parent1[:cross_point], parent2[cross_point:]))
+            child2 = np.concatenate((parent2[:cross_point], parent1[cross_point:]))
+            return child1, child2
+        return parent1, parent2
+
+    def mutate(self, individual: np.ndarray) -> np.ndarray:
+        """
+        Apply mutation to an individual.
+
+        Args:
+            individual (np.ndarray): The individual to mutate.
+
+        Returns:
+            np.ndarray: The mutated individual.
+        """
+        for i in range(self.dim):
+            if random.random() < self.mutation_prob:
+                individual[i] = rng.uniform(self.bounds[i][0], self.bounds[i][1])
+        return individual
+
+    def evaluate_population(self) -> list[tuple[np.ndarray, float]]:
+        """
+        Evaluate the fitness of the entire population in parallel.
+
+        Returns:
+            list[tuple[np.ndarray, float]]: The population with their respective fitness values.
+        """
+        with ThreadPoolExecutor() as executor:
+            return list(
+                executor.map(lambda ind: (ind, self.fitness(ind)), self.population)
+            )
+
+    def evolve(self) -> np.ndarray:
+        """
+        Evolve the population over the generations to find the best solution.
+
+        Returns:
+            np.ndarray: The best individual found during the evolution process.
+        """
+        for generation in range(self.generations):
+            # Evaluate population fitness (multithreaded)
+            population_score = self.evaluate_population()
+
+            # Check the best individual
+            best_individual = max(population_score, key=lambda x: x[1])[0]
+            best_fitness = self.fitness(best_individual)
+
+            # Select parents for next generation
+            parents = self.select_parents(population_score)
+            next_generation = []
+
+            # Generate offspring using crossover and mutation
+            for i in range(0, len(parents), 2):
+                parent1, parent2 = parents[i], parents[(i + 1) % len(parents)]
+                child1, child2 = self.crossover(parent1, parent2)
+                next_generation.append(self.mutate(child1))
+                next_generation.append(self.mutate(child2))
+
+            # Ensure population size remains the same
+            self.population = next_generation[: self.population_size]
+
+            if generation % 10 == 0:
+                print(f"Generation {generation}: Best Fitness = {best_fitness}")
+
+        return best_individual
+
+
+# Example target function for optimization
+def target_function(var_x: float, var_y: float) -> float:
+    """
+    Example target function (parabola) for optimization.
+
+    Args:
+        var_x (float): The x-coordinate.
+        var_y (float): The y-coordinate.
+
+    Returns:
+        float: The value of the function at (var_x, var_y).
+
+    Example:
+    >>> target_function(0, 0)
+    0
+    >>> target_function(1, 1)
+    2
+    """
+    return var_x**2 + var_y**2  # Simple parabolic surface (minimization)
+
+
+# Set bounds for the variables (var_x, var_y)
+bounds = [(-10, 10), (-10, 10)]  # Both var_x and var_y range from -10 to 10
+
+
+# Instantiate and run the genetic algorithm
+ga = GeneticAlgorithm(
+    function=target_function,
+    bounds=bounds,
+    population_size=N_POPULATION,
+    generations=N_GENERATIONS,
+    mutation_prob=MUTATION_PROBABILITY,
+    crossover_rate=CROSSOVER_RATE,
+    maximize=False,  # Minimize the function
+)
+
+
+best_solution = ga.evolve()
+print(f"Best solution found: {best_solution}")
+print(f"Best fitness (minimum value of function): {target_function(*best_solution)}")