def AMT_BGA(problem,
dims,
reps,
trans,
fitness_func='toy_problem',
addr="problems/toy_problem"):
"""[bestSol, fitness_hist, alpha] = TSBGA(problem, dims, reps, trans): Adaptive
Model-based Transfer Binary GA. The crossover and mutation for this simple
binary GA are uniform crossover and bit-flip mutation.
INPUT:
problem: problem type, 'onemax', 'onemin', or 'trap5'
dims: problem dimensionality
reps: number of repeated trial runs
trans: trans.transfer: binary variable
trans.TrInt: transfer interval for AMT
OUTPUT:
bestSol: best solution for each repetiion
fitness: history of best fitness for each generation
alpha: transfer coefficient
"""
pop = 50
gen = 100
transfer = trans['transfer']
if transfer:
TrInt = trans['TrInt']
all_models = Tools.load_from_file(os.path.join(addr, 'all_models'))
fitness_hist = np.zeros((reps, gen))
bestSol = np.zeros((reps, dims))
alpha = [None] * (reps)
for rep in range(reps):
alpha_rep = []
population = np.round(np.random.rand(pop, dims))
if fitness_func == 'toy_problem':
fitness = fitness_eval(population, problem, dims)
else:
fitness = knapsack_fitness_eval(population, problem, dims, pop)
ind = np.argmax(fitness)
best_fit = fitness[ind]
print('Generation 0 best fitness = ', str(best_fit))
fitness_hist[rep, 0] = best_fit
for i in range(1, gen):
# As we consider all the population as parents, we don't samplt P^{s}
if transfer and i % TrInt == 0:
mmodel = MixtureModel(all_models)
mmodel.createtable(population, True, 'umd')
mmodel.EMstacking()
# Recombination of probability models
mmodel.mutate()
# Mutation of stacked probability model
offspring = mmodel.sample(pop)
alpha_rep.append(mmodel.alpha)
print('Transfer coefficient at generation ', str(i), ': ',
str(mmodel.alpha))
else:
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])
if fitness_func == 'toy_problem':
cfitness = fitness_eval(population, problem, dims)
else:
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, :]
print('Generation ', str(i), ' best fitness = ',
str(np.max(fitness_hist)))
fitness_hist[rep, i] = fitness[0]
alpha[rep] = alpha_rep
bestSol[rep, :] = population[ind, :]
return bestSol, fitness_hist, alpha
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
