def get_re_stats(self, method="QP", alpha=0.05):
"""Get random-effect statistics."""
if method == "QP":
n_iters = np.atleast_2d(self.tau2).shape[1]
if n_iters > 10:
warn("Method 'QP' is not parallelized; it may take a while to "
"compute CIs for {} parallel tau^2 values.".format(
n_iters))
# Make sure we have an estimate of v if it wasn't observed
v = self.estimator.get_v(self.dataset)
cis = []
for i in range(n_iters):
args = {
"y": self.dataset.y[:, i],
"v": v[:, i],
"X": self.dataset.X,
"alpha": alpha,
}
try:
q_cis = q_profile(**args)
except Exception:
q_cis = {"ci_l": np.nan, "ci_u": np.nan}
cis.append(q_cis)
else:
raise ValueError(
"Invalid CI method '{}'; currently only 'QP' is available.".
format(method))
return {
"tau^2": self.tau2,
"ci_l": np.array([ci["ci_l"] for ci in cis]),
"ci_u": np.array([ci["ci_u"] for ci in cis]),
}
def test_q_profile(vars_with_intercept):
bounds = q_profile(*vars_with_intercept, 0.05)
assert set(bounds.keys()) == {"ci_l", "ci_u"}
assert round(bounds["ci_l"], 4) == 3.8076
assert round(bounds["ci_u"], 2) == 59.61
def test_q_profile(vars_with_intercept):
bounds = q_profile(*vars_with_intercept, 0.05)
assert set(bounds.keys()) == {'ci_l', 'ci_u'}
assert round(bounds['ci_l'], 4) == 3.8076
assert round(bounds['ci_u'], 4) == 59.6144