def test_sample_prior_predictive_var_names(model_fixture):
model, observed = model_fixture
prior = forward_sampling.sample_prior_predictive(model(),
var_names=["model/sd"],
sample_shape=())
assert set(prior) == set(["model/sd"])
prior = forward_sampling.sample_prior_predictive(
model(), var_names=["model/x", "model/y"], sample_shape=())
assert set(prior) == set(["model/x", "model/y"])
# Assert we can get the values of observeds if we ask for them explicitly
# even if sample_from_observed_fixture is False
prior = forward_sampling.sample_prior_predictive(
model(),
var_names=["model/x", "model/y"],
sample_shape=(),
sample_from_observed=False)
assert set(prior) == set(["model/x", "model/y"])
assert np.all(prior["model/x"] == observed)
# Assert an exception is raised if we pass wrong names
model_func = model()
expected_message = "Some of the supplied var_names are not defined in the supplied model {}.\nList of unknown var_names: {}".format(
model_func, ["X"])
with pytest.raises(ValueError, match=re.escape(expected_message)):
prior = forward_sampling.sample_prior_predictive(
model_func, var_names=["X", "model/y"], sample_shape=())
def test_sample_prior_predictive_int_sample_shape(model_fixture,
n_draws_fixture):
model, observed = model_fixture
prior_int = forward_sampling.sample_prior_predictive(
model(), sample_shape=n_draws_fixture)
prior_tuple = forward_sampling.sample_prior_predictive(
model(), sample_shape=(n_draws_fixture, ))
assert all((prior_int[k].shape == v.shape for k, v in prior_tuple.items()))
def test_sample_prior_predictive_on_glm(glm_model_fixture,
use_auto_batching_fixture,
sample_shape_fixture):
model, is_vectorized_model, core_shapes = glm_model_fixture
if not use_auto_batching_fixture and not is_vectorized_model and len(
sample_shape_fixture) > 0:
with pytest.raises(AssertionError):
prior = forward_sampling.sample_prior_predictive(
model(),
sample_shape=sample_shape_fixture,
use_auto_batching=use_auto_batching_fixture,
).prior_predictive
for k, v in core_shapes.items():
# The (1,) comes from trace_to_arviz imposed chain axis
assert prior[k].shape == (1, ) + sample_shape_fixture + v
else:
prior = forward_sampling.sample_prior_predictive(
model(),
sample_shape=sample_shape_fixture,
use_auto_batching=use_auto_batching_fixture).prior_predictive
for k, v in core_shapes.items():
# The (1,) comes from trace_to_arviz imposed chain axis
assert prior[k].shape == (1, ) + sample_shape_fixture + v
def test_sample_prior_predictive(model_fixture, sample_shape_fixture,
sample_from_observed_fixture):
model, observed = model_fixture
prior = forward_sampling.sample_prior_predictive(
model(), sample_shape_fixture, sample_from_observed_fixture)
if sample_from_observed_fixture:
assert set(["model/sd", "model/x", "model/y", "model/mu",
"model/dy"]) == set(prior)
assert all(
(value.shape == sample_shape_fixture for value in prior.values()))
else:
assert set(["model/sd", "model/y", "model/mu",
"model/dy"]) == set(prior)
assert all((value.shape == sample_shape_fixture
for name, value in prior.items()
if name in {"model/sd", "model/mu"}))
assert all((value.shape == sample_shape_fixture + observed.shape
for name, value in prior.items()
if name in {"model/x", "model/y", "model/dy"}))
assert np.allclose(prior["model/y"], observed, rtol=1e-5)
assert np.allclose(prior["model/y"] * 2, prior["model/dy"])