def test_bootstrap_run(df, n):
np.random.seed(99)
b_run = BinneyRun(col_success='success',
col_total='total',
covariates=['x1'],
df=df,
solver_method='ipopt',
data_type='binomial')
b_run.fit()
b_run.make_uncertainty(n_boots=15)
assert np.vstack(b_run.bootstrap.parameters).shape == (15, 2)
np.testing.assert_array_almost_equal(b_run.bootstrap.parameters[0],
np.array([0.99940296, 1.99394325]))
np.testing.assert_array_almost_equal(b_run.bootstrap.parameters[5],
np.array([1.01366631, 1.99569274]))
np.testing.assert_array_almost_equal(np.vstack(
b_run.bootstrap.parameters).mean(axis=0),
b_run.params_opt,
decimal=1)
uis = np.quantile(np.vstack(b_run.bootstrap.parameters),
q=[0.025, 0.975],
axis=0)
assert all(b_run.params_opt > uis[0, :])
assert all(b_run.params_opt < uis[1, :])
draws = b_run.predict_draws(df=df)
assert draws.shape == (15, n)
assert all(draws.var(axis=1) > 0)
def test_hierarchical_bootstrap(n, group_data):
np.random.seed(99)
n_boots = 30
b_run = BinneyRun(col_success='success',
col_total='total',
covariates=['x1'],
df=group_data,
col_group='g',
solver_method='scipy',
data_type='bernoulli')
b_run.fit()
b_run.make_uncertainty(n_boots=n_boots)
assert len(b_run.bootstrap.parameters) == n_boots
for param in b_run.bootstrap.parameters:
assert isinstance(param, Dict)
for group, item in param.items():
assert len(param[group]) == 2
draws = b_run.predict_draws(df=group_data)
assert draws.shape == (n_boots, n)
assert all(draws.var(axis=1) > 0)
assert (group_data['p'].values > np.quantile(draws, q=0.025, axis=0)).all()
assert (group_data['p'].values < np.quantile(draws, q=0.975, axis=0)).all()