simulation.py

import numpy as np
from scipy.optimize import minimize

def run_stress_test(noise_ratio, num_vars=100, num_clauses=400):
    # Αυστηρή αρχικοποίηση για στατιστική ορθότητα
    np.random.seed(42)
    hidden_truth = np.random.randint(0, 2, num_vars)
    clauses = []
    for _ in range(num_clauses):
        vars_c = np.random.choice(range(1, num_vars + 1), 3, replace=False)
        if np.random.rand() < noise_ratio:
            clause = [v if np.random.rand() > 0.5 else -v for v in vars_c]
        else:
            clause = [v if hidden_truth[v-1] == 1 else -v for v in vars_c]
        clauses.append(clause)

    def objective(x):
        # Soft-constraint penalty function
        x = np.clip(x, 1e-12, 1 - 1e-12)
        loss = 0.0
        for clause in clauses:
            p_false = 1.0
            for l in clause:
                val = x[abs(l)-1]
                p_false *= (1 - val) if l > 0 else val
            loss += p_false
        return loss

    # Ανώτατη ακρίβεια σύγκλισης
    res = minimize(objective, np.full(num_vars, 0.5), method='L-BFGS-B', 
                   bounds=[(0, 1)]*num_vars, 
                   options={'ftol': 1e-25, 'gtol': 1e-25, 'maxiter': 50000})
    
    match = np.mean((res.x > 0.5).astype(int) == hidden_truth)
    return match

# Σάρωση 0.0 έως 1.0 με βήμα 0.05
noise_levels = np.linspace(0, 1, 21)
print(f"{'Noise':>8} | {'Accuracy':>10}")
print("-" * 22)
for n in noise_levels:
    acc = run_stress_test(n)
    print(f"{n*100:6.0f}% | {acc*100:9.2f}%")