def test_add_obseration():
np.random.seed(100691)
param_vects = np.random.uniform(low=0, high=1, size=(3, 2))
values_vects = np.random.uniform(low=0, high=1, size=(3, 1))
params = []
values = []
for param_vect, value_vect in zip(param_vects, values_vects):
params.append(
ParameterVector().from_dict({"p0": param_vect[0], "p1": param_vect[1]})
)
values.append(ParameterVector().from_dict({"obj0": value_vect[0]}))
obs = Observations()
obs.add_observation(params, values)
assert (
np.linalg.norm(
np.array(obs.params) - np.array([param.to_array() for param in params])
)
< 1e-7
)
assert (
np.linalg.norm(
np.array(obs.values) - np.array([value.to_array() for value in values])
)
< 1e-7
)
def test_planner_ask_tell(two_param_space, show_progressbar):
planner = Hyperopt(show_progressbar=show_progressbar)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(two_param_space, goal, rng):
planner = Smac(goal=goal, rng=rng)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(two_param_space, disp, eps, ftol, gtol, maxcor, maxfun, maxiter, maxls):
planner = DifferentialEvolution()
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(two_param_space, disp, eps, ftol, maxiter):
planner = Slsqp(disp=disp, eps=eps, ftol=ftol, maxiter=maxiter)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(two_param_space, learning_rate, dx, random_seed, init_guess):
planner = SteepestDescent(learning_rate=1e-3, dx=1e-5, random_seed=None, init_guess=None)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(two_param_space, stddev):
planner = Cma(stddev=stddev)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(two_param_space, disp, maxiter, maxfev, initial_simplex, xatol, fatol, adaptive):
planner = Simplex(disp=disp, maxiter=maxiter, maxfev=maxfev, initial_simplex=initial_simplex, xatol=xatol,
fatol=fatol, adaptive=adaptive)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_generating_new_offsprings(two_param_space):
planner = Genetic(pop_size=4)
planner.set_param_space(param_space=two_param_space)
obs = Observations()
for i in range(10):
param = planner.recommend(observations=obs)
obj = np.sum(param.to_array()**2)
value = ParameterVector(dict={'objective': obj})
obs.add_observation(param, value)
def test_planner_ask_tell(two_param_space, pop_size, cx_prob, mut_prob, mate_args, mutate_args, select_args):
planner = Genetic(pop_size=pop_size, cx_prob=cx_prob, mut_prob=mut_prob, mate_args=mate_args, mutate_args=mutate_args,
select_args=select_args)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(two_param_space, batch_size, exact_eval, model_type, acquisition_type):
planner = Gpyopt(batch_size=batch_size, exact_eval=exact_eval, model_type=model_type,
acquisition_type=acquisition_type)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def _test_planner_ask_tell(two_param_space, max_iters, options, particles):
planner = ParticleSwarms(max_iters=max_iters,
options=options,
particles=particles)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(
two_param_space
): #, disp, eps, ftol, gtol, maxcor, maxfun, maxiter, maxls):
# planner = Lbfgs(disp=disp, eps=eps, ftol=ftol, gtol=gtol, maxcor=maxcor, maxfun=maxfun, maxiter=maxiter, maxls=maxls)
planner = Sobol()
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planners_ask_tell(two_param_space, planner_kind):
"""Tests the function Planner
"""
planner = Planner(kind=planner_kind)
planner.set_param_space(param_space=two_param_space)
obs = Observations()
planner.tell(observations=obs)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_custom_planner_class(two_param_space):
"""Tests using a custom planner that is built by inheriting from AbstractPlanner.
"""
planner = Planner(kind=MyPlannerAlg)
planner.set_param_space(param_space=two_param_space)
obs = Observations()
planner.tell(observations=obs)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(two_param_space, batches, boosted, parallel,
sampling_strategies):
planner = Phoenics(batches=batches,
boosted=boosted,
parallel=parallel,
sampling_strategies=sampling_strategies)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(two_param_space, disp, maxiter, gtol, norm, eps):
planner = ConjugateGradient(disp=disp,
maxiter=maxiter,
gtol=gtol,
norm=norm,
eps=eps)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector().from_dict({'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_planner_ask_tell(two_param_space, levels, budget, shuffle,
random_seed):
planner = Grid(goal='minimize',
levels=levels,
budget=budget,
shuffle=shuffle,
random_seed=random_seed)
planner.set_param_space(param_space=two_param_space)
param = planner.ask()
value = ParameterVector(dict={'objective': 0.})
obs = Observations()
obs.add_observation(param, value)
planner.tell(observations=obs)
def test_get_values():
np.random.seed(100691)
param_vects = np.random.uniform(low=0, high=1, size=(3, 2))
values_vects = np.random.uniform(low=0, high=1, size=(3, 1))
params = []
values = []
for param_vect, value_vect in zip(param_vects, values_vects):
params.append(ParameterVector().from_dict({
'p0': param_vect[0],
'p1': param_vect[1]
}))
values.append(ParameterVector().from_dict({'obj0': value_vect[0]}))
obs = Observations()
obs.add_observation(params, values)
obs_vals = obs.get_values(as_array=True, opposite=False)
assert [_.to_array() for _ in values] == list(obs_vals)
obs_vals = obs.get_values(as_array=True, opposite=True)
assert [-1 * _.to_array() for _ in values] == list(obs_vals)
obs_vals = obs.get_values(as_array=False, opposite=False)
assert [_.to_dict() for _ in values] == list(obs_vals)
obs_vals = obs.get_values(as_array=False, opposite=True)
obs_dicts = [_.to_dict() for _ in values]
for _, obs_dict in enumerate(obs_dicts):
for key, val in obs_dict.items():
assert val == -1 * obs_vals[_][key]
def test_planners_recommend(two_param_space, planner_kind):
"""Tests the recommend method in all planners
"""
planner = Planner(kind=planner_kind)
planner.set_param_space(param_space=two_param_space)
obs = Observations()
# special case for LatinHypercube: we need to set the number of iterations in advance!
num_iter = 3
if planner_kind == 'LatinHypercube':
planner.budget = num_iter
for i in range(num_iter):
param = planner.recommend(observations=obs)
value = ParameterVector().from_dict({'objective': 0.})
obs.add_observation(param, value)
def test_add_obseration():
np.random.seed(100691)
param_vects = np.random.uniform(low=0, high=1, size=(3, 2))
values_vects = np.random.uniform(low=0, high=1, size=(3, 1))
params = []
values = []
for param_vect, value_vect in zip(param_vects, values_vects):
params.append(ParameterVector().from_dict({
'p0': param_vect[0],
'p1': param_vect[1]
}))
values.append(ParameterVector().from_dict({'obj0': value_vect[0]}))
obs = Observations()
obs.add_observation(params, values)
assert obs.params == params
assert obs.values == values
def test_resetting_planner(two_param_space):
planner = Genetic(pop_size=3)
planner.set_param_space(param_space=two_param_space)
# run once
obs = Observations()
for i in range(5):
param = planner.recommend(observations=obs)
obj = np.sum(param.to_array() ** 2)
value = ParameterVector(dict={'objective': obj})
obs.add_observation(param, value)
# run again from scratch
planner.reset()
obs = Observations()
for i in range(5):
param = planner.recommend(observations=obs)
obj = np.sum(param.to_array() ** 2)
value = ParameterVector(dict={'objective': obj})
obs.add_observation(param, value)
def test_declaration():
np.random.seed(100691)
param_vects = np.random.uniform(low=0, high=1, size=(3, 2))
values_vects = np.random.uniform(low=0, high=1, size=(3, 1))
params = []
values = []
for param_vect, value_vect in zip(param_vects, values_vects):
params.append(
ParameterVector().from_dict({"p0": param_vect[0], "p1": param_vect[1]})
)
values.append(ParameterVector().from_dict({"obj0": value_vect[0]}))
observations = Observations()
observations.add_observation(params, values)
test_params = [param.to_array() for param in params]
test_values = [value.to_array() for value in values]
assert np.linalg.norm(test_params - observations.get_params()) < 1e-7
assert np.linalg.norm(test_values - observations.get_values()) < 1e-7