from typing import List
import torch
from evox.utils import lexsort
[docs]
def tournament_selection_multifit(n_round: int, fitnesses: List[torch.Tensor], tournament_size: int = 2) -> torch.Tensor:
"""
Perform tournament selection based on multiple fitness values.
:param n_round: Number of rounds of selection (how many solutions to select).
:param fitnesses: A list of 1D tensors, each representing the fitness values of candidates for different objectives.
:param tournament_size: Number of candidates in each tournament. Defaults to 2.
:return: Indices of the selected solutions after tournament selection.
This function performs tournament selection by randomly selecting a group of candidates for each round,
and selecting the best one from each group based on their fitness values across multiple objectives.
"""
fitness_tensor = torch.stack(fitnesses, dim=1)
num_candidates = fitness_tensor.size(0)
parents = torch.randint(0, num_candidates, (n_round, tournament_size), device=fitnesses[0].device)
candidates_fitness = fitness_tensor[parents]
candidates_fitness = lexsort(candidates_fitness.unbind(-1))
selected_parents = torch.gather(parents, 1, candidates_fitness[:, 0].unsqueeze(1)).squeeze()
return selected_parents
[docs]
def tournament_selection(n_round: int, fitness: torch.Tensor, tournament_size: int = 2) -> torch.Tensor:
"""
Perform tournament selection based on single fitness values.
:param n_round: Number of rounds of selection (how many solutions to select).
:param fitness: A 1D tensor representing the fitness values of candidates.
:param tournament_size: Number of candidates in each tournament. Defaults to 2.
:return: Indices of the selected solutions after tournament selection.
This function performs tournament selection by randomly selecting a group of candidates for each round,
and selecting the best one from each group based on their fitness values.
"""
num_candidates = fitness.size(0)
parents = torch.randint(0, num_candidates, (n_round, tournament_size), device=fitness.device)
candidates_fitness = fitness[parents]
winner_indices = torch.argmin(candidates_fitness, dim=1)
selected_parents = torch.gather(parents, 1, winner_indices.unsqueeze(1)).squeeze()
return selected_parents
EvoX