The SelectionAlgorithm module provides different strategies for selecting agents for reproduction.

Overview

Selection algorithms determine which agents in a population survive and reproduce. These algorithms form the backbone of the evolutionary process by implementing the principle of "survival of the fittest."

Selection Strategies

Bardo implements several selection strategies, each with different characteristics:

Hall of Fame Competition

• Maintains an elite subset of the best-performing agents across generations
• New agents must compete against the hall of fame to be selected
• Provides protection against evolutionary forgetting and cycling

Tournament Selection

• Randomly selects subgroups of agents and chooses the best from each group
• Selection pressure can be tuned by adjusting tournament size
• Balances exploration and exploitation effectively

Truncation Selection

• Simply selects the top N performers from the population
• Provides high selection pressure for rapid convergence
• May lead to premature convergence on suboptimal solutions

Rank-Based Selection

• Selection probability is based on rank rather than absolute fitness
• Maintains selection pressure even when fitness differences are small
• Helps prevent premature convergence in some scenarios

Fitness Proportionate Selection

• Also known as "roulette wheel" selection
• Selection probability is directly proportional to fitness
• Simple but can lead to early domination by slightly superior agents

Implementation Notes

• Selection algorithms operate within species to preserve diversity
• Parameters can adjust selection pressure to balance exploration vs. exploitation
• Custom selection algorithms can be implemented by adding new functions to this module