def testRandomImputation(self):
rs = numpy.random.RandomState(1)
for i in range(0, 150, 15):
# First random imputation sanity check
num_samples = max(1, i * 10)
num_feat = max(1, i)
num_censored = int(num_samples * 0.1)
X = rs.rand(num_samples, num_feat)
y = numpy.sin(X[:, 0:1])
cutoff = max(y) * 0.9
y[y > cutoff] = cutoff
# We have some cen data
cen_X = X[:num_censored, :]
cen_y = y[:num_censored]
uncen_X = X[num_censored:, :]
uncen_y = y[num_censored:]
cen_y /= 2
cs = ConfigurationSpace()
for i in range(num_feat):
cs.add_hyperparameter(
UniformFloatHyperparameter(name="a_%d" % i,
lower=0,
upper=1,
default_value=0.5))
types, bounds = get_types(cs, None)
print(types)
print(bounds)
print('#' * 120)
print(cen_X)
print(uncen_X)
print('~' * 120)
self.model = RandomForestWithInstances(types=types,
bounds=bounds,
instance_features=None,
seed=1234567980)
imputor = rfr_imputator.RFRImputator(rng=rs,
cutoff=cutoff,
threshold=cutoff * 10,
change_threshold=0.01,
max_iter=5,
model=self.model)
imp_y = imputor.impute(censored_X=cen_X,
censored_y=cen_y,
uncensored_X=uncen_X,
uncensored_y=uncen_y)
if imp_y is None:
continue
for idx in range(cen_y.shape[0]):
self.assertGreater(imp_y[idx], cen_y[idx])
self.assertTrue(numpy.isfinite(imp_y).all())
def testRealImputation(self):
rs = numpy.random.RandomState(1)
imputor = rfr_imputator.RFRImputator(rng=rs,
cutoff=self.scen.cutoff,
threshold=self.scen.cutoff*10,
change_threshold=0.01, max_iter=10,
model=self.model)
r2e = runhistory2epm.RunHistory2EPM4LogCost(
scenario=self.scen, num_params=3,
success_states=[StatusType.SUCCESS, ],
impute_censored_data=True, impute_state=[StatusType.TIMEOUT],
imputor=imputor, rng=rs)
print("%s" % str(r2e.transform(self.rh)[0]))
def testRealImputation(self):
# Without instance features
rs = numpy.random.RandomState(1)
cs, rh = self.get_runhistory(num_success=5,
num_timeout=1,
num_capped=2)
scen = self.get_scenario()
model = self.get_model(cs)
imputor = rfr_imputator.RFRImputator(rng=rs,
cutoff=scen.cutoff,
threshold=scen.cutoff * 10,
change_threshold=0.01,
max_iter=10,
model=model)
r2e = runhistory2epm.RunHistory2EPM4LogCost(
scenario=scen,
num_params=3,
success_states=[
StatusType.SUCCESS,
],
impute_censored_data=True,
impute_state=[StatusType.TIMEOUT],
imputor=imputor,
rng=rs,
)
self.assertEqual(r2e.transform(rh)[1].shape, (8, 1))
self.assertEqual(r2e.transform(rh)[1].shape, (8, 1))
# Now with instance features
instance_features = {
run_key.instance_id: numpy.random.rand(10)
for run_key in rh.data
}
scen = self.get_scenario(instance_features)
model = self.get_model(cs, instance_features)
with unittest.mock.patch.object(model,
attribute='train',
wraps=model.train) as train_wrapper:
imputor = rfr_imputator.RFRImputator(rng=rs,
cutoff=scen.cutoff,
threshold=scen.cutoff * 10,
change_threshold=0.01,
max_iter=10,
model=model)
r2e = runhistory2epm.RunHistory2EPM4LogCost(
scenario=scen,
num_params=3,
success_states=[
StatusType.SUCCESS,
],
impute_censored_data=True,
impute_state=[StatusType.TIMEOUT],
imputor=imputor,
rng=rs,
)
X, y = r2e.transform(rh)
self.assertEqual(X.shape, (8, 13))
self.assertEqual(y.shape, (8, 1))
num_calls = len(train_wrapper.call_args_list)
self.assertGreater(num_calls, 1)
self.assertEqual(train_wrapper.call_args_list[0][0][0].shape,
(5, 13))
self.assertEqual(train_wrapper.call_args_list[1][0][0].shape,
(8, 13))
X, y = r2e.transform(rh)
self.assertEqual(X.shape, (8, 13))
self.assertEqual(y.shape, (8, 1))
self.assertGreater(len(train_wrapper.call_args_list),
num_calls + 1)
self.assertEqual(
train_wrapper.call_args_list[num_calls][0][0].shape, (5, 13))
self.assertEqual(
train_wrapper.call_args_list[num_calls + 1][0][0].shape,
(8, 13))
