def GenKnapsack(n=1000,
v=10,
r=5,
type_wp='uc',
type_c='rk',
addr="problems/knapsack"):
assert type_wp in ['uc', 'wc', 'sc'], 'type_wp is not valid'
assert type_c in ['rk', 'ak'], 'type_wp is not valid'
# type_wp = 'uc'; strong or weakly or un-correlated
# type_c = 'rk'; average or restrictive knapsack --- ALWAYS we choose average
w = (1 + np.round(np.random.rand(n) * (v - 1)))
if type_wp == 'uc':
p = 1 + np.round(np.random.rand(n) * (v - 1))
elif type_wp == 'wc':
p = w + np.round(r - 2 * r * np.random.rand(n))
p[p <= 0] = w[p <= 0]
elif type_wp == 'sc':
p = w + r
if type_c == 'rk':
cap = int(2 * v)
elif type_c == 'ak':
cap = int(0.5 * np.sum(w))
# print(w, p, cap)
th_best, _ = knapsack(w, p, cap)
KP_uc_rk = {}
KP_uc_rk['w'] = w
KP_uc_rk['p'] = p
KP_uc_rk['cap'] = cap
KP_uc_rk['opt'] = th_best
Tools.save_to_file(os.path.join(addr, 'KP_{}_{}'.format(type_wp, type_c)),
KP_uc_rk)
def BGA(problem,
dims,
th_best,
pop=200,
gen=1000,
addr="problems/toy_problem"):
"""[bestSol, fitness_hist] = BGA(problem,dims,th_best): simple binary GA with
uniform crossover and bit-flip mutation.
INPUT:
problem: problem type, 'onemax', 'onemin', or 'trap5'
dims: problem dimensionality
th_best: global best fitness value (used for early stop to build
probabilistic models)
OUTPUT:
bestSol: best solution
fitness: history of best fitness for each generation"""
bestSol = None
all_models = Tools.load_from_file(os.path.join(addr, 'all_models'))
fitness_hist = np.zeros(gen)
population = np.round(np.random.rand(pop, dims))
fitness = fitness_eval(population, problem, dims)
buildmodel = True
fitness_hist[0] = max(fitness)
print('Generation 0 best fitness = ', str(fitness_hist[0]))
for i in range(1, gen):
parent1 = population[np.random.permutation(pop), :]
parent2 = population[np.random.permutation(pop), :]
tmp = np.random.rand(pop, dims)
offspring = np.zeros((pop, dims))
index = tmp >= 0.5
offspring[index] = parent1[index]
index = tmp < 0.5
offspring[index] = parent2[index]
tmp = np.random.rand(pop, dims)
index = tmp < (1 / dims)
offspring[index] = np.abs(1 - offspring[index])
cfitness = fitness_eval(offspring, problem, dims)
interpop = np.append(population, offspring, 0)
interfitness = np.append(fitness, cfitness)
index = np.argsort((-interfitness))
interfitness = interfitness[index]
fitness = interfitness[:pop]
interpop = interpop[index, :]
population = interpop[:pop, :]
fitness_hist[i] = fitness[0]
print('Generation ', str(i), ' best fitness = ', str(fitness_hist[i]))
if (fitness[1] >= th_best or i == gen) and buildmodel:
print('Building probablistic model...')
model = ProbabilityModel('umd')
model.buildmodel(population)
all_models.append(model)
Tools.save_to_file(os.path.join(addr, 'all_models'), all_models)
print('Complete!')
fitness_hist[i + 1:] = fitness[0]
bestSol = population[0, :]
break
return bestSol, fitness_hist
def KP_BGA(problem,
dims,
premSTOP,
pop=100,
gen=100,
addr="problems/toy_problem"):
"""[bestSol, fitness_hist] = BGA(problem,dims,th_best): simple binary GA with
uniform crossover and bit-flip mutation.
INPUT:
problem: problem generated by GenKnapsack
dims: problem dimensionality
premSTOP: used for early stop if no improvement is made for 20
consecutive generations
OUTPUT:
bestSol: best solution
fitness: history of best fitness for each ge neration"""
bestSol = None
all_models = Tools.load_from_file(os.path.join(addr, 'all_models'))
fitness_hist = np.zeros(gen)
population = np.round(np.random.rand(pop, dims))
fitness = knapsack_fitness_eval(population, problem, dims, pop)
fitness_hist[0] = bestfitness = max(fitness)
print('Generation 0 best fitness = ', str(fitness_hist[0]))
counter = 0
for i in range(1, gen):
parent1 = population[np.random.permutation(pop), :]
parent2 = population[np.random.permutation(pop), :]
tmp = np.random.rand(pop, dims)
offspring = np.zeros((pop, dims))
index = tmp >= 0.5
offspring[index] = parent1[index]
index = tmp < 0.5
offspring[index] = parent2[index]
tmp = np.random.rand(pop, dims)
index = tmp < (1 / dims)
offspring[index] = np.abs(1 - offspring[index])
cfitness = knapsack_fitness_eval(population, problem, dims, pop)
interpop = np.append(population, offspring, 0)
interfitness = np.append(fitness, cfitness)
index = np.argsort((-interfitness))
interfitness = interfitness[index]
fitness = interfitness[:pop]
interpop = interpop[index, :]
population = interpop[:pop, :]
fitness_hist[i] = fitness[0]
print('Generation ', str(i), ' best fitness = ', str(fitness_hist[i]))
if fitness[0] > bestfitness:
bestfitness = fitness[0]
counter = 0
else:
counter += 1
if counter == 20 and premSTOP:
fitness_hist[i:] = fitness_hist[i]
break
bestSol = population[0, :]
model = ProbabilityModel('umd')
model.buildmodel(population)
all_models.append(model)
Tools.save_to_file(os.path.join(addr, 'all_models'), all_models)
return bestSol, fitness_hist
