def test_param(self):
"""Testing the completion of the fit and effect estimation of an automated Parametric DML"""
Y, T, X, _ = ihdp_surface_B()
est = AutomatedLinearDML(model_y=automl_model_reg(),
model_t=GradientBoostingClassifier(),
featurizer=None,
discrete_treatment=True)
est.fit(Y, T, X=X)
_ = est.effect(X)
def test_SLearner(self):
"""Testing the completion of the fit and effect estimation of an AutomatedSLearner"""
# Test constant treatment effect with multi output Y
# Test heterogeneous treatment effect
# Need interactions between T and features
Y, T, X, _ = ihdp_surface_B()
est = AutomatedSLearner(overall_model=automl_model_reg())
est.fit(Y, T, X=X)
_ = est.effect(X)
def test_TLearner(self):
"""Testing the completion of the fit and effect estimation of an AutomatedTLearner"""
# TLearner test
# Instantiate TLearner
Y, T, X, _ = ihdp_surface_B()
est = AutomatedTLearner(models=automl_model_reg())
# Test constant and heterogeneous treatment effect, single and multi output y
est.fit(Y, T, X=X)
_ = est.effect(X)
def test_nonparam(self):
"""Testing the completion of the fit and effect estimation of an automated Nonparametic DML"""
Y, T, X, _ = ihdp_surface_B()
est = AutomatedNonParamDML(
model_y=automl_model_reg(),
model_t=automl_model_clf(),
model_final=automl_model_sample_weight_reg(),
featurizer=None,
discrete_treatment=True)
est.fit(Y, T, X=X)
_ = est.effect(X)
def simulate2(trial, std0=0, std1=0, scenario='A'):
rct_res = []
adpt0_res = []
adpt1_res = []
true_res = []
for t in range(trial):
if scenario == 'A':
Y, T, X, true_ITE = dgps.ihdp_surface_A()
elif scenario == 'B':
Y, T, X, true_ITE = dgps.ihdp_surface_B()
perm = np.random.permutation(len(Y))
Y = Y[perm]
T = T[perm]
X = X[perm]
Y1 = Y.copy()
Y1[T==0] += true_ITE[T==0]
Y1 += np.random.normal(0, std1, size=len(Y1))
Y0 = Y.copy()
Y0[T==1] -= true_ITE[T==1]
Y0 += np.random.normal(0, std0, size=len(Y0))
rct = RCT()
adpt0 = Adapt(pretraining=10)
adpt1 = Adapt(pretraining=100)
opt0 = OPT()
opt1 = OPT()
rct_temp = []
adpt0_temp = []
adpt1_temp = []
true_temp = []
for period_t in range(len(Y)):
rct(period_t, X[period_t], Y0[period_t], Y1[period_t])
adpt0(period_t, X[period_t], Y0[period_t], Y1[period_t])
adpt1(period_t, X[period_t], Y0[period_t], Y1[period_t])
if period_t > 2:
rct_temp.append(rct.effect())
adpt0_temp.append(adpt0.effect())
adpt1_temp.append(adpt1.effect())
true_temp.append(np.mean(true_ITE))
rct_res.append(rct_temp)
adpt0_res.append(adpt0_temp)
adpt1_res.append(adpt1_temp)
true_res.append(true_temp)
return rct_res, adpt0_res, adpt1_res, true_res
def test_forest_dml(self):
"""Testing the completion of the fit and effect estimation of an AutomatedForestDML"""
Y, T, X, _ = ihdp_surface_B()
est = AutomatedForestDML(model_y=automl_model_reg(),
model_t=GradientBoostingClassifier(),
discrete_treatment=True,
n_estimators=1000,
subsample_fr=.8,
min_samples_leaf=10,
min_impurity_decrease=0.001,
verbose=0,
min_weight_fraction_leaf=.01)
est.fit(Y, T, X=X)
_ = est.effect(X)
def automl_model_clf():
return copy.deepcopy(AUTOML_CONFIG_CLF)
# Linear models are required for parametric dml
def automl_model_linear_reg():
return copy.deepcopy(AUTOML_CONFIG_LINEAR_REG)
# sample weighting models are required for nonparametric dml
def automl_model_sample_weight_reg():
return copy.deepcopy(AUTOML_CONFIG_SAMPLE_WEIGHT_REG)
# Test values
Y, T, X, _ = ihdp_surface_B()
except ImportError:
pass # automl not installed
@pytest.mark.automl
class TestAutomatedDML(unittest.TestCase):
@classmethod
def setUpClass(cls):
subscription_id = os.getenv("SUBSCRIPTION_ID")
resource_group = os.getenv("RESOURCE_GROUP")
workspace_name = os.getenv("WORKSPACE_NAME")
auth = AzureCliAuthentication()
setAutomatedMLWorkspace(auth=auth,
