import numpy as np import ioh import math # ------------------------ (1+1)EA ------------------------ class OnePlusLambdaVarEA: lam: int = 10 budget_factor = 5 seed = 42 r : float = 2.0 F : float = 0.98 def __call__(self, f, seed) -> tuple[float, np.ndarray]: """ Run the (1+λ)EA on a single IOH problem object `f`. Returns ------- (fx_best, x_best) : tuple[float, np.ndarray] Best fitness found and corresponding solution. """ self.seed = seed rng = np.random.default_rng(self.seed) n = f.meta_data.n_variables budget = self.budget_factor * n * n # Initialize x within bounds (PBO: binary {0,1}) x = rng.integers(0,2,n) fx = f(x) c = 0 while f.state.evaluations < budget: # Generate λ offspring best_y = None best_fy = None mean = self.r std = math.sqrt(pow(self.F,c) * self.r * (1-self.r/n)) for _ in range(self.lam): ell = min(math.floor(rng.normal(mean,std)),n / 2) while ell < 1: ell = min(math.floor(rng.normal(mean,std)),n / 2) idx = rng.choice(n, size=ell, replace=False) y = x.copy() # bit flip for {0,1}: new = lb + ub - old (works for 0/1 bounds) y[idx] = 1 - y[idx] fy = f(y) if best_y is None: best_y, best_fy, win_r = y, fy, ell else: if fy > best_fy: best_y, best_fy, win_r = y, fy, ell if best_fy >= fx: x, fx = best_y, best_fy if best_fy >= f.optimum.y: break if win_r == self.r: c += 1 else: self.r = win_r c = 0 return fx, x