def fit(self, kernel=None, n_restarts_optimizer=4):
start = current_milli_time()
if kernel is None: kernel = self.kernel
# self.model = GaussianProcessRegressor(kernel=RationalQuadratic(length_scale_bounds=(0.08, 100)) + WhiteKernel(noise_level_bounds=(1e-5, 1e-2)), n_restarts_optimizer=10)
self.model = GaussianProcessRegressor(
kernel=kernel,
n_restarts_optimizer=n_restarts_optimizer,
copy_X_train=True,
random_state=self.seed)
self.model.fit(self.first_xys + self.fixed_xys,
self.first_zs + probe(self.f, self.fixed_xys))
self.model_time += current_milli_time() - start
def calculate_tour(self):
"""ps. Keep the old tour if it's still within the budget and the number of cities remains compatible with the tour length."""
start = current_milli_time()
xys = [self.depot] + self.fixed_xys + self.xys
n = len(self.tour)
tmptour = self.tour + [0]
self.cost, self.feasible = n, True
for i in list(range(n)):
a, b = xys[tmptour[i]]
c, d = xys[tmptour[i + 1]]
self.cost += dist(a, b, c, d)
if self.budget - self.cost >= 1:
self.possibly_no_more_room_left = False
if self.cost > self.budget or self.tour == [] or n != len(xys):
self.tour, self.feasible, self.cost, self.possibly_no_more_room_left = plan_tour(
[self.depot] + self.fixed_xys + self.xys, self.budget, True,
self.fixed_tour)
self.tour_time += current_milli_time() - start
def select_kernel(self):
start = current_milli_time()
self.kernel = kernel_selector(
self.first_xys + self.fixed_xys,
self.first_zs + probe(self.f, self.fixed_xys))
self.model_time += current_milli_time() - start
def predict(self, xys):
start = current_milli_time()
preds = self.model.predict(xys, return_std=False)
self.pred_time += current_milli_time() - start
return preds
def predict_stds(self, xys):
start = current_milli_time()
_, stds = self.model.predict(xys, return_std=True)
self.pred_time += current_milli_time() - start
return stds
trip = Trip(f, depot, first_xys, first_zs, budget, plotter)
print('# selecting kernel...')
trip.select_kernel() # TOD
print('# fitting...')
trip.fit()
# Main loop. Report stddev and error...: first iteration -> ...using only known points; (conta == 0)
# second iteration -> ...after orienteering; (conta == 1)
# third iteration -> ...after first distortion; (conta == 2)
# next iterations -> ...after probing [only for online mode] and next distortions; (conta > 2)
conta, acctime = 0, 0
trip_var = sum(trip.stds_simulated(TSxy))
if argv[1] == 'plotvar': trip.plotvar = True
while acctime < time_limit * 3600000:
trip.tour_time, trip.model_time, trip.pred_time = 0, 0, 0
start = current_milli_time()
if conta > 0: trip.add_while_possible(trip.add_maxvar_point(TSxy))
if conta == 1:
trip_var = sum(trip.stds_simulated(TSxy))
elif conta > 1:
if online and conta > 2: trip.probe_next()
if len(trip.xys) > 0:
if alg == '1c':
trip_var = custom_distortion(trip, TSxy, nb, random_distortion)
if alg == 'sw':
trip_var = swarm_distortion(
trip, TSxy, time_limit * 3600000 - acctime -
(current_milli_time() - start))
if alg == 'ga': trip_var = ga_distortion(trip, TSxy)
if alg == 'a4':
def py_outf(it, leader, fx, x):
"""Function called at every iteration. It is for logging purposes, but also to stop when a criterion is matched."""
elapsed = current_milli_time() - start
# print(available_time - elapsed )
return available_time - elapsed # negative number = stop
def swarm_distortion(trip,
testset_xy,
available_time,
maxf=4000,
maxit=4000,
size=100):
"""Optimize points in trip via PSO, evaluating over a test set. The available time is almost never respected."""
def py_outf(it, leader, fx, x):
"""Function called at every iteration. It is for logging purposes, but also to stop when a criterion is matched."""
elapsed = current_milli_time() - start
# print(available_time - elapsed )
return available_time - elapsed # negative number = stop
def py_objf(xs):
"""Function called at every iteration. It is for swarm fitness evaluation."""
def var(x):
"""Evaluates fitness of a given particle of the swarm."""
xys = tuplefy(
x
) # Ps.: According to my tests with oldtrip.count(), trip methods don't need to be thread-safe here.
v, _ = trip.fitness(xys, testset_xy)
return v
return [var(x) for x in xs]
# Initial PSO settings.
start = current_milli_time()
x0 = flat(trip.xys)
variabs = len(x0)
problem = {
'Variables': variabs,
'objf': py_objf,
'lb': zeros(variabs),
'ub': ones(variabs),
'x0': x0
}
# , 'A': [[-1.0 / sqrt(3), 1], [-1.0, sqrt(3)], [1.0, sqrt(3)]], 'b': [0, 0, 6] # <- Defaults
options = {
'maxf': maxf,
'maxit': maxit,
'social': 0.5,
'cognitial': 0.5,
'fweight': 0.4,
'iweight': 0.9,
'size': size,
'iprint': 10,
'tol': 1E-5,
'ddelta': 0.5,
'idelta': 2.0,
'outputfcn': py_outf,
'vectorized': 1
}
# PSO running.
result = pswarm(problem, options)
# PSO results.
if result['ret'] == 0: # zero means successful
new_xys = tuplefy(result['x'])
trip.xys = new_xys.copy()
return result['f']