def test_error_numerator_with_unstabilized(self):
df = load_sample_data(False)
ipm = IPMW(df, missing_variable='dead', stabilized=False)
with pytest.raises(ValueError):
ipm.regression_models(
model_denominator='male + age0 + dvl0 + cd40',
model_numerator='male')
def test_nonmonotone_detection(self):
df = pd.DataFrame()
df['a'] = [0, 0, np.nan]
df['b'] = [np.nan, 0, 1]
ipm = IPMW(df, missing_variable=['a', 'b'])
with pytest.raises(ValueError):
ipm.regression_models(model_denominator=['b', 'a'])
def test_check_uniform(self):
df = pd.DataFrame()
df['a'] = [0, 0, 1, np.nan]
df['b'] = [1, 0, 0, np.nan]
df['c'] = [1, np.nan, np.nan, np.nan]
ipm = IPMW(df, missing_variable=['a', 'b', 'c'])
# Should be uniform
assert ipm._check_uniform(df, miss1='a', miss2='b')
# Not uniform
assert not ipm._check_uniform(df, miss1='a', miss2='c')
def test_check_overall_uniform(self):
df = pd.DataFrame()
df['a'] = [0, 0, 1, np.nan]
df['b'] = [1, 0, 0, np.nan]
df['c'] = [1, 0, 0, np.nan]
df['d'] = [1, np.nan, np.nan, np.nan]
ipm = IPMW(df, missing_variable=['a', 'b', 'c'])
# Should be uniform
assert ipm._check_overall_uniform(df, miss_vars=['a', 'b', 'c'])[1]
# Not uniform
assert not ipm._check_overall_uniform(df, miss_vars=['a', 'b', 'c', 'd'])[1]
def test_stabilized_weights(self):
df = load_sample_data(False)
ipm = IPMW(df, missing_variable='dead', stabilized=True)
ipm.regression_models(model_denominator='male + age0 + dvl0 + cd40')
ipm.fit()
npt.assert_allclose(np.mean(ipm.Weight), 0.99993685627)
npt.assert_allclose(np.std(ipm.Weight, ddof=1), 0.019866910369)
def test_unstabilized_weights(self):
df = load_sample_data(False)
ipm = IPMW(df, missing_variable='dead', stabilized=False)
ipm.regression_models(model_denominator='male + age0 + dvl0 + cd40')
ipm.fit()
npt.assert_allclose(np.mean(ipm.Weight), 1.0579602715)
npt.assert_allclose(np.std(ipm.Weight, ddof=1), 0.021019729152)
def test_weighted_model(self):
df = load_sample_data(False).drop(columns=['dead'])
df['age_sq'] = df['age0'] ** 2
df['age_cu'] = df['age0'] ** 3
df['cd4_sq'] = df['cd40'] ** 2
df['cd4_cu'] = df['cd40'] ** 3
ipmw = IPMW(df, missing_variable='cd4_wk45')
ipmw.regression_models('art + male + age0 + age_sq + age_cu + cd40 + cd4_sq + cd4_cu + dvl0',
print_results=False)
ipmw.fit()
df['ipcw'] = ipmw.Weight
snm = GEstimationSNM(df.dropna(), exposure='art', outcome='cd4_wk45', weights='ipcw')
snm.exposure_model('male + age0 + age_sq + age_cu + cd40 + cd4_sq + cd4_cu + dvl0', print_results=False)
snm.structural_nested_model('art')
snm.fit()
npt.assert_allclose(snm.psi, [244.379181], atol=1e-5)
snm = GEstimationSNM(df.dropna(), exposure='art', outcome='cd4_wk45', weights='ipcw')
snm.exposure_model('male + age0 + age_sq + age_cu + cd40 + cd4_sq + cd4_cu + dvl0', print_results=False)
snm.structural_nested_model('art')
snm.fit(solver='search')
npt.assert_allclose(snm.psi, [244.379181], atol=1e-5)
def test_single_model_does_not_break(self):
df = load_monotone_missing_data()
ipm = IPMW(df,
missing_variable=['B', 'C'],
stabilized=False,
monotone=True)
ipm.regression_models(model_denominator='L')
ipm.fit()
x = ipm.Weight
def test_monotone_example(self):
# TODO find R or SAS to test against
df = load_monotone_missing_data()
ipm = IPMW(df, missing_variable=['B', 'C'], stabilized=False, monotone=True)
ipm.regression_models(model_denominator=['L + A', 'L + B'])
ipm.fit()
df['w'] = ipm.Weight
dfs = df.dropna(subset=['w'])
npt.assert_allclose(np.average(dfs['B'], weights=dfs['w']), 0.41877344861340654)
npt.assert_allclose(np.average(dfs['C'], weights=dfs['w']), 0.5637116735464095)
def test_error_for_non_nan2(self):
df = pd.DataFrame()
df['a'] = [0, 0, 1]
df['b'] = [0, 0, np.nan]
with pytest.raises(ValueError):
IPMW(df, missing_variable=['a', 'b'], stabilized=True)
def test_error_too_many_model(self):
df = load_sample_data(False)
ipm = IPMW(df, missing_variable='dead')
with pytest.raises(ValueError):
ipm.regression_models(
model_denominator=['male + age0', 'male + age0 + dvl0'])
def test_missing_count2(self):
df = load_sample_data(False)
ipm = IPMW(df, missing_variable='dead', stabilized=True)
ipm.regression_models(model_denominator='art')
assert 517 == np.sum(ipm.df['_observed_indicator_'])
assert 30 == np.sum(1 - ipm.df['_observed_indicator_'])
def test_missing_count(self, mdata):
ipm = IPMW(mdata, missing_variable='M', stabilized=True)
ipm.regression_models(model_denominator='A')
assert 6 == np.sum(ipm.df['_observed_indicator_'])
assert 4 == np.sum(1 - ipm.df['_observed_indicator_'])
def test_error_for_non_nan(self, mdata):
with pytest.raises(ValueError):
IPMW(mdata, missing_variable='L', stabilized=True)
def test_missing_count(self, mdata):
ipm = IPMW(mdata, missing_variable='M', stabilized=True)
assert 6 == np.sum(ipm.df['_observed_indicator_'])
assert 4 == np.sum(1 - ipm.df['_observed_indicator_'])
