def test_restart_with_seed_gives_same_predictions(temp_dir, inter_max_samples):
wrapper = gpyopt.GPyOptObjectiveWrapper(_polynomial)
wrapper.set_variable_parameter('x', 'continuous', (-1., 1.))
wrapper.set_fixed_parameter('d', 2)
wrapper.set_fixed_parameter('shift', 1)
hist_path = os.path.join(temp_dir, 'hist.csv')
params = gpyopt.BayesianOptimizationParams(
max_samples=inter_max_samples,
model_type='gp',
seed=1,
)
gpyopt.bayesopt(wrapper, params, hist_path)
params.max_samples = 8
best_1 = gpyopt.bayesopt(wrapper, params, hist_path)
best_2 = gpyopt.bayesopt(wrapper, params)
assert abs(best_1['x'] - best_2['x']) < 1e-5
def test_simple_objective():
wrapper = gpyopt.GPyOptObjectiveWrapper(_squared)
wrapper.set_fixed_parameter('shift', 1.)
wrapper.set_variable_parameter('x', 'continuous', (-1., 1.))
params = gpyopt.BayesianOptimizationParams(
model_type='gp',
kernel_type='rbf',
seed=1,
max_samples=10,
noise_var=0,
)
best_1 = gpyopt.bayesopt(wrapper, params)
best_2 = gpyopt.bayesopt(wrapper, params)
assert best_1['x'] == best_2['x']
params.log10_min_diff = -3
params.max_samples = None
best_3 = gpyopt.bayesopt(wrapper, params)
# note that log10_min_diff is not the same as the absolute
# value between best results
assert abs(_squared(best_3['x'], 1.) - 1.) < 1e-2
def test_strings_are_provided():
def _obj(mystring):
assert isinstance(mystring, str)
return int(mystring[0])
wrapper = gpyopt.GPyOptObjectiveWrapper(_obj)
wrapper.set_variable_parameter('mystring', 'categorical', '1234')
params = gpyopt.BayesianOptimizationParams(
initial_design_numdata=2,
max_samples=10,
)
best = gpyopt.bayesopt(wrapper, params)
assert best['mystring'] == '1'
def test_can_serialize_lots_of_stuff(temp_dir):
def _obj(**kwargs):
return 0.
wrapper = gpyopt.GPyOptObjectiveWrapper(_obj)
wrapper.set_fixed_parameter('a', object())
wrapper.set_variable_parameter('b', 'continuous', (0., 1.))
wrapper.set_variable_parameter('c', 'discrete', (0, 1))
d_tup = (1, None, 'None', 0., object())
wrapper.set_variable_parameter('d', 'categorical', d_tup)
hist_path = os.path.join(temp_dir, 'hist.csv')
params = gpyopt.BayesianOptimizationParams(
initial_design_samples=4,
max_samples=5,
seed=5,
)
best_1 = gpyopt.bayesopt(wrapper, params, hist_path)
assert best_1['d'] in d_tup
best_2 = gpyopt.bayesopt(wrapper, params, hist_path)
assert best_1['c'] == best_2['c']
assert abs(best_1['b'] - best_2['b']) < 1e-5
assert best_1['d'] == best_2['d']
def test_multiple_variable_types():
wrapper = gpyopt.GPyOptObjectiveWrapper(_polynomial)
wrapper.set_variable_parameter('shift', 'categorical', (10, 0, 1))
wrapper.set_variable_parameter('x', 'continuous', (.1, .9))
wrapper.set_variable_parameter('d', 'discrete', tuple(range(-1, 5)))
params = gpyopt.BayesianOptimizationParams(
model_type='gp',
seed=1,
log10_min_diff=-2,
initial_design_samples=10,
noise_var=0,
)
best = gpyopt.bayesopt(wrapper, params)
assert not best['shift']
assert best['d'] == 5
def test_constraints():
wrapper = gpyopt.GPyOptObjectiveWrapper(_enforcer_objective)
wrapper.set_variable_parameter('int_', 'discrete', tuple(range(100)))
wrapper.set_variable_parameter('day', 'categorical', ('monday', 'tuesday'))
params = gpyopt.BayesianOptimizationParams(
initial_design_samples=4,
max_iterations=5,
seed=2,
)
best = gpyopt.bayesopt(wrapper,
params,
constraints=[
_day_cannot_be_tuesday,
_int_not_odd,
])
assert best['int_'] in range(0, 100, 2)
assert best['day'] == 'monday'