Implement Estimation of Distribution Algorithm (EDA/UMDA)

[eprifti]

Priority: Medium. Learn probability distribution P(feature_i=1) from best solutions each generation. Sample new models from distribution. Like ACO pheromone but more principled — pure statistical learning, no crossover/mutation.