def build_sampling_algorithm(
algo, proposal_distribution: Optional[str],
proposal_distribution_sigma: Optional[numpy_types],
proposal_listeners: List[proposal_listener_types]):
if algo != "metropolis":
if proposal_distribution is not None:
raise TypeError(
"Only Metropolis Hastings supports the proposal_distribution parameter"
)
if len(proposal_listeners) > 0:
raise TypeError(
"Only Metropolis Hastings supports the proposal_listeners parameter"
)
if (proposal_distribution is None and len(proposal_listeners) > 0):
raise TypeError(
"If you pass in proposal_listeners you must also specify proposal_distribution"
)
builder: JavaObject = algorithms[algo].builder()
if proposal_distribution is not None:
proposal_distribution_object = ProposalDistribution(
type_=proposal_distribution,
sigma=proposal_distribution_sigma,
listeners=proposal_listeners)
builder = builder.proposalDistribution(
proposal_distribution_object.unwrap())
sampling_algorithm: JavaObject = builder.build()
return sampling_algorithm
def __init__(
self,
proposal_distribution: str,
latents: Iterable[Vertex],
proposal_listeners=[],
proposal_distribution_sigma: Union[tensor_arg_types,
List[tensor_arg_types]] = None):
if (proposal_distribution is None and len(proposal_listeners) > 0):
raise TypeError(
"If you pass in proposal_listeners you must also specify proposal_distribution"
)
builder: JavaObject = k.jvm_view().MetropolisHastings.builder()
latents = list(latents)
proposal_distribution_object = ProposalDistribution(
type_=proposal_distribution,
sigma=proposal_distribution_sigma,
latents=latents,
listeners=proposal_listeners)
rejection_strategy = k.jvm_view().RollBackToCachedValuesOnRejection()
builder = builder.proposalDistribution(
proposal_distribution_object.unwrap())
builder = builder.rejectionStrategy(rejection_strategy)
super().__init__(builder.build())
def test_it_throws_if_it_doesnt_recognise_the_type() -> None:
with pytest.raises(KeyError, match=r"'foo'"):
ProposalDistribution("foo")
def test_it_throws_if_you_specify_sigma_but_the_type_isnt_gaussian() -> None:
with pytest.raises(TypeError, match=r'Parameter sigma is not valid unless type is "gaussian"'):
ProposalDistribution("prior", sigma=1.)
def test_it_throws_if_you_specify_gaussian_with_empty_list_of_latents(net: BayesNet) -> None:
with pytest.raises(TypeError, match=r"Gaussian Proposal Distribution requires values for latents"):
ProposalDistribution("gaussian", latents=[], sigma=[])
def test_it_throws_if_you_specify_gaussian_without_values_for_latents() -> None:
with pytest.raises(TypeError, match=r"Gaussian Proposal Distribution requires values for latents"):
ProposalDistribution("gaussian", sigma=1.)
def test_it_throws_if_you_specify_gaussian_with_not_enough_sigmas_for_each_latent(net: BayesNet) -> None:
with pytest.raises(
TypeError, match=r"Gaussian Proposal Distribution requires a sigma or a list of sigmas for each latent"):
ProposalDistribution("gaussian", latents=list(net.iter_latent_vertices()), sigma=[1.])
def test_you_can_create_a_gaussian_proposal_distribution(sigma: tensor_arg_types, net: BayesNet) -> None:
ProposalDistribution("gaussian", latents=list(net.iter_latent_vertices()), sigma=sigma)
def test_you_can_create_a_prior_proposal_distribution(net) -> None:
ProposalDistribution("prior", latents=list(net.iter_latent_vertices()))
def test_you_can_create_a_prior_proposal_distribution() -> None:
ProposalDistribution("prior")
def test_it_throws_if_it_doesnt_recognise_the_type() -> None:
with pytest.raises(KeyError) as excinfo:
ProposalDistribution("foo")
assert str(excinfo.value) == "'foo'"
def test_it_throws_if_you_specify_sigma_but_the_type_isnt_gaussian() -> None:
with pytest.raises(TypeError) as excinfo:
ProposalDistribution("prior", sigma=np.array(1.))
assert str(excinfo.value
) == 'Parameter sigma is not valid unless type is "gaussian"'
def test_it_throws_if_you_specify_gaussian_without_a_value_for_sigma() -> None:
with pytest.raises(TypeError) as excinfo:
ProposalDistribution("gaussian")
assert str(excinfo.value
) == "Gaussian Proposal Distribution requires a value for sigma"
def test_you_can_create_a_gaussian_proposal_distribution() -> None:
ProposalDistribution("gaussian", sigma=np.array(1.))