def test_simulatedAnnealing():
carDomain = Struct({
'position': Box([-10,10], [-10,10], [0,1]),
'heading': Box([0, math.pi]),
})
space = FeatureSpace({
'cars': Feature(Array(carDomain, [2]))
})
def f(sample):
return sample.cars[0].heading[0] - 0.75
def proposal_func(sample, iteration, decay_rate, num_variables):
decayed_width = decay_rate**iteration*1
lower_bound = [(sample[i] - decayed_width/2) for i in range(num_variables)]
upper_bound = [(sample[i] + decayed_width/2) for i in range(num_variables)]
for index in range(num_variables): # check whether updated range for each parameter is within original ranges
if lower_bound[index] < 0:
lower_bound[index] = 0
if upper_bound[index] > 1:
upper_bound[index] = 1
return np.random.uniform(lower_bound, upper_bound)
sa_params = DotMap()
sa_params.T = 10
sa_params.decay_rate = 0.9
sa_params.iterations = 3
sa_params.reset_temp = 5
sa_params.num_epoch = 4
sa_params.loss_f = f
sa_params.temp_f = lambda t: 0.91*t
sa_params.iter_f = lambda length: int(math.ceil(0.5*length))
sa_params.proposal_f = proposal_func
sampler = FeatureSampler.simulatedAnnealingSamplerFor(space, sa_params=sa_params)
for i in range(10):
print(f'Sample #{i}:')
print(sampler.nextSample())
upper_bound = [(sample[i] + decayed_width / 2)
for i in range(num_variables)]
# check whether updated range for each parameter is within original ranges
for index in range(num_variables):
if lower_bound[index] < 0:
lower_bound[index] = 0
if upper_bound[index] > 1:
upper_bound[index] = 1
return np.random.uniform(lower_bound, upper_bound)
# Basic set of parameters
sa_params = DotMap()
sa_params.T = 10
sa_params.decay_rate = 0.9
sa_params.iterations = 3
sa_params.reset_temp = 5
sa_params.num_epoch = 10
sa_params.temp_f = lambda t: 0.91 * t
sa_params.iter_f = lambda length: int(math.ceil(0.5 * length))
sa_params.proposal_f = proposal_func
## Tests
def test_simulatedAnnealing():
carDomain = Struct({
'position': Box([-10, 10], [-10, 10], [0, 1]),
'heading': Box([0, math.pi]),
