def test_lorenz() -> None:
error = math.inf
window = 0
prior_mu: Union[Tuple[float, float, float],
Tuple[numpy_types, numpy_types,
numpy_types]] = (3., 3., 3.)
model = LorenzModel(sigma, beta, rho, time_step)
observed = list(model.run_model(window_size * max_windows))
while error > converged_error and window < max_windows:
xt0 = Gaussian(prior_mu[0], 1.0)
yt0 = Gaussian(prior_mu[1], 1.0)
zt0 = Gaussian(prior_mu[2], 1.0)
graph_time_steps = list(build_graph((xt0, yt0, zt0)))
xt0.set_and_cascade(prior_mu[0])
yt0.set_and_cascade(prior_mu[1])
zt0.set_and_cascade(prior_mu[2])
apply_observations(graph_time_steps, window, observed)
optimizer = GradientOptimizer(xt0)
optimizer.max_a_posteriori()
posterior = get_time_slice_values(graph_time_steps, window_size - 1)
post_t = (window + 1) * (window_size - 1)
actual_at_post_t = observed[post_t]
error = math.sqrt((actual_at_post_t.x - posterior[0])**2 +
(actual_at_post_t.y - posterior[1])**2 +
(actual_at_post_t.z - posterior[2])**2)
prior_mu = (posterior[0], posterior[1], posterior[2])
window += 1
assert error <= converged_error
def thermometers_max_likelihood_gradient(model: Model, algorithm) -> None:
net = BayesNet(model.temperature.iter_connected_graph())
gradient_optimizer = GradientOptimizer(net, algorithm)
result = gradient_optimizer.max_likelihood()
assert result.fitness() < 0.
temperature = result.value_for(model.temperature)
assert 20.99 < temperature < 21.01
def test_thermometers_max_likelihood_gradient(model: Model) -> None:
net = BayesNet(model.temperature.get_connected_graph())
gradient_optimizer = GradientOptimizer(net)
logProb = gradient_optimizer.max_likelihood()
assert logProb < 0.
temperature = model.temperature.get_value()
assert 20.995 < temperature < 21.005
def gradient_optimzer_example():
model = build_model()
model.first_thermometer.observe(25)
bayes_net = model.to_bayes_net()
# %%SNIPPET_START%% PythonGradientOptimizer
optimizer = GradientOptimizer(bayes_net,
max_evaluations=5000,
relative_threshold=1e-8,
absolute_threshold=1e-8)
optimizer.max_a_posteriori()
calculated_temperature = model.temperature.get_value()
def test_thermometers_example():
# %%SNIPPET_START%% PythonTwoThermometers
with Model() as m:
m.temperature = Uniform(20., 30.)
m.first_thermometer = Gaussian(m.temperature, 2.5)
m.second_thermometer = Gaussian(m.temperature, 5.)
m.first_thermometer.observe(25.)
m.second_thermometer.observe(30.)
bayes_net = m.to_bayes_net()
optimizer = GradientOptimizer(bayes_net)
optimizer.max_a_posteriori()
calculated_temperature = m.temperature.get_value()
def test_gradient_can_set_max_eval_builder_properties(model: Model) -> None:
gradient_optimizer = GradientOptimizer(model.temperature, max_evaluations=5)
with pytest.raises(Py4JJavaError):
#This throws a Gradient Optimizer: "Reached Max Evaluations" error
logProb = gradient_optimizer.max_a_posteriori()
def test_gradient_op_throws_with_invalid_net_param() -> None:
with pytest.raises(TypeError) as excinfo:
GradientOptimizer(500) # type: ignore # this is expected to fail mypy
assert str(excinfo.value) == "net must be a Vertex or a BayesNet. Was given {}".format(int)
def test_gradient_op_vertex(model: Model) -> None:
gradient_optimizer = GradientOptimizer(model.temperature)
assert len(list(gradient_optimizer.net.get_latent_vertices())) == 1
def test_gradient_op_bayes_net(model: Model) -> None:
net = BayesNet(model.temperature.get_connected_graph())
gradient_optimizer = GradientOptimizer(net)
assert gradient_optimizer.net is net
def test_gradient_can_set_max_eval_builder_properties_for_conjugate_gradient(model: Model) -> None:
gradient_optimizer = GradientOptimizer(model.temperature, ConjugateGradient(max_evaluations=5))
with pytest.raises(Py4JJavaError, match=r"An error occurred while calling o[\d]*.maxAPosteriori."):
# This throws a Gradient Optimizer: "Reached Max Evaluations" error
logProb = gradient_optimizer.max_a_posteriori()
error = math.inf
window = 0
prior_mu = (3., 3., 3.)
model = LorenzModel(sigma, beta, rho, time_step)
observed = list(model.run_model(window_size * max_windows))
while error > converged_error and window < max_windows:
xt0 = Gaussian(prior_mu[0], 1.0)
yt0 = Gaussian(prior_mu[1], 1.0)
zt0 = Gaussian(prior_mu[2], 1.0)
graph_time_steps = list(build_graph((xt0, yt0, zt0)))
xt0.set_and_cascade(prior_mu[0])
yt0.set_and_cascade(prior_mu[1])
zt0.set_and_cascade(prior_mu[2])
apply_observations(graph_time_steps, window, observed)
optimizer = GradientOptimizer(xt0)
optimizer.max_a_posteriori()
posterior = get_time_slice_values(graph_time_steps, window_size - 1)
post_t = (window + 1) * (window_size - 1)
actual_at_post_t = observed[post_t]
error = math.sqrt((actual_at_post_t.x - posterior[0])**2 +
(actual_at_post_t.y - posterior[1])**2 +
(actual_at_post_t.z - posterior[2])**2)
prior_mu = (posterior[0], posterior[1], posterior[2])
window += 1
# %%SNIPPET_END%% LorenzPythonFull