def test_ipw_numpy_data():
w, t, y = generate_wty_linear_multi_w_data(n=N,
wdim=5,
binary_treatment=True,
delta=ATE,
data_format='numpy')
ipw = IPWEstimator()
ipw.fit(w, t, y)
assert ipw.estimate_ate() == approx(ATE, rel=.2)
def test_ipw_matches_causallib(linear_data_pandas):
w, t, y = linear_data_pandas
causallib_ipw = IPW(learner=LogisticRegression())
causallib_ipw.fit(w, t)
potential_outcomes = causallib_ipw.estimate_population_outcome(
w, t, y, treatment_values=[0, 1])
causallib_effect = causallib_ipw.estimate_effect(potential_outcomes[1],
potential_outcomes[0])[0]
ipw = IPWEstimator()
ipw.fit(w, t, y)
our_effect = ipw.estimate_ate()
assert our_effect == causallib_effect
ATE = 5
N = 50
@pytest.fixture('module')
def linear_gen_model():
w, t, y = generate_wty_linear_multi_w_data(n=N,
wdim=5,
binary_treatment=True,
delta=ATE,
data_format='numpy')
return LinearGenModel(w, t, y, binary_treatment=True)
@pytest.fixture('module',
params=[IPWEstimator(),
StandardizationEstimator()],
ids=class_name)
def estimator(request):
return request.param
def test_ate_metrics(linear_gen_model, estimator):
metrics = calculate_metrics(gen_model=linear_gen_model,
estimator=estimator,
n_seeds=10,
conf_ints=False)
assert metrics['ate_squared_bias'] + metrics['ate_variance'] == approx(
metrics['ate_mse'])
assert metrics['ate_bias'] == approx(0, abs=1)
def ipw_estimator(request, linear_data):
w, t, y = linear_data
prop_score_model = request.param
ipw = IPWEstimator(prop_score_model=prop_score_model)
ipw.fit(w, t, y)
return ipw
def test_ipw_weight_trimming_and_stabilized_weights(linear_data):
w, t, y = linear_data
ipw = IPWEstimator(trim_weights=True, stabilized=True)
ipw.fit(w, t, y)
assert ipw.estimate_ate() == approx(ATE, rel=.2)
def test_ipw_trim_eps(linear_data):
w, t, y = linear_data
ipw = IPWEstimator(trim_eps=.05)
ipw.fit(w, t, y)
assert ipw.estimate_ate() == approx(ATE, rel=.2)
model_type='outcome',
gen_models=GEN_MODELS)
print('STRATIFIED STANDARDIZATION')
strat_df = run_experiments_for_estimator(
lambda model: StratifiedStandardizationEstimator(outcome_models=model),
model_grid=OUTCOME_MODEL_GRID,
exclude=[('lalonde_cps', 'KernelRidge')],
save_location=RESULTS_DIR / 'psid_cps_twins_strat_standard.csv',
meta_est_name='stratified_standardization',
model_type='outcome',
gen_models=GEN_MODELS)
print('IPW')
ps_df = run_experiments_for_estimator(
lambda model: IPWEstimator(prop_score_model=model),
model_grid=PROP_SCORE_MODEL_GRID,
# exclude=[('lalonde_psid', 'SVM_rbf')],
exclude=['SVM_rbf'],
save_location=RESULTS_DIR / 'psid_cps_twins_ipw.csv',
meta_est_name='ipw',
model_type='prop_score',
gen_models=GEN_MODELS)
print('IPW TRIM EPS 0.01')
ps_trim_df = run_experiments_for_estimator(
lambda model: IPWEstimator(prop_score_model=model, trim_eps=0.01),
model_grid=PROP_SCORE_MODEL_GRID,
# exclude=[('lalonde_psid', 'SVM_rbf')],
exclude=['SVM_rbf'],
save_location=RESULTS_DIR / 'psid_cps_twins_ipw_trim_01.csv',