def test_direct_times(self):
""" Tests the running time of the package with and without file writing. """
self.report('DiRect running times')
log_file_names = ['', 'test_log']
for prob in self.problems:
clock_times = []
real_times = []
for log_file_name in log_file_names:
start_clock = time.clock()
start_real_time = time.time()
_, _, _ = oper_utils.direct_ft_maximise(prob.obj, prob.bounds, self.max_evals,
log_file_name=log_file_name)
clock_time = time.clock() - start_clock
real_time = time.time() - start_real_time
clock_times.append(clock_time)
real_times.append(real_time)
if log_file_name:
try:
os.remove(log_file_name)
except OSError:
pass
# Print results out
result_str = ', '.join(['file: \'%s\': clk=%0.4f, real=%0.4f, #evals=%d'%(
log_file_names[i], clock_times[i], real_times[i], self.max_evals) \
for i in range(len(log_file_names))])
self.report('%s:: %s'%(prob.descr, result_str), 'test_result')
def test_direct(self):
""" Test direct optmisation."""
self.report('DiRect minimise and maximise:')
num_min_successes = 0
num_max_successes = 0
for prob in self.problems:
# First the minimimum
min_val_soln, _, _ = oper_utils.direct_ft_minimise(
prob.obj, prob.lb, prob.ub, self.max_evals)
diff = abs(prob.min_val - min_val_soln)
self.report(prob.descr +
'(min):: True: %0.4f, Soln: %0.4f, diff: %0.4f.' %
(prob.min_val, min_val_soln, diff))
min_is_successful = 1
num_min_successes += min_is_successful
# Now the maximum
max_val_soln, _, _ = oper_utils.direct_ft_maximise(
prob.obj, prob.lb, prob.ub, self.max_evals)
diff = abs(prob.max_val - max_val_soln)
self.report(
prob.descr + '(max):: True: %0.4f, Soln: %0.4f, diff: %0.4f' %
(prob.max_val, max_val_soln, diff), 'test_result')
max_is_successful = 1
num_max_successes += max_is_successful
# Check if successful
assert num_min_successes == len(self.problems)
assert num_max_successes == len(self.problems)
def test_direct_with_history(self):
""" Tests direct with history. """
self.report('DiRect with history.')
for prob in self.problems:
min_val_soln, _, history = oper_utils.direct_ft_maximise(prob.obj, prob.bounds, \
self.max_evals, return_history=True)
if history is None:
assert is_nondecreasing(history.curr_opt_vals)
assert np.abs(min_val_soln - history.curr_opt_vals[-1]) < 1e-4
def _direct_maximise_obj(self, obj, num_evals):
""" Maximise with direct. """
if num_evals is None:
lead_const = 10 * min(5, self.dim)**2
num_evals = lambda t: np.clip(lead_const * np.sqrt(min(t, 1000)),
2000, 3e4)
lb = self.bounds[:, 0]
ub = self.bounds[:, 1]
opt_val, opt_pt, _ = direct_ft_maximise(obj, lb, ub, num_evals)
return opt_val, opt_pt
def _direct_wrap(*args): """ A wrapper so as to only return the optimal point. """ _, opt_pt, _ = direct_ft_maximise(*args) return opt_pt