def setUp(self):
mean = np.array([-1, 1, 0])
covariance = np.array([[6, 0.7, 0.2], [0.7, 3, 0.9], [0.2, 0.9, 1]])
self.test_model = JGD(["x", "y", "z"], mean, covariance)
self.nuts_sampler = NUTSda(
model=self.test_model, grad_log_pdf=GradLogPDFGaussian
)
class TestNUTSInference(unittest.TestCase):
def setUp(self):
mean = np.array([-1, 1, 0])
covariance = np.array([[6, 0.7, 0.2], [0.7, 3, 0.9], [0.2, 0.9, 1]])
self.test_model = JGD(['x', 'y', 'z'], mean, covariance)
self.nuts_sampler = NUTSda(model=self.test_model,
grad_log_pdf=GradLogPDFGaussian)
def test_errors(self):
with self.assertRaises(TypeError):
NUTS(model=self.test_model, grad_log_pdf=JGD)
with self.assertRaises(TypeError):
NUTS(model=self.test_model,
grad_log_pdf=None,
simulate_dynamics=GradLogPDFGaussian)
with self.assertRaises(ValueError):
NUTSda(model=self.test_model, delta=-0.2, grad_log_pdf=None)
with self.assertRaises(ValueError):
NUTSda(model=self.test_model,
delta=1.1,
grad_log_pdf=GradLogPDFGaussian)
with self.assertRaises(TypeError):
NUTS(self.test_model,
GradLogPDFGaussian).sample(initial_pos={1, 1, 1},
num_samples=1)
with self.assertRaises(ValueError):
NUTS(self.test_model,
GradLogPDFGaussian).sample(initial_pos=[1, 1], num_samples=1)
with self.assertRaises(TypeError):
NUTSda(self.test_model, GradLogPDFGaussian).sample(initial_pos=1,
num_samples=1,
num_adapt=1)
with self.assertRaises(ValueError):
NUTSda(self.test_model,
GradLogPDFGaussian).sample(initial_pos=[1, 1, 1, 1],
num_samples=1,
num_adapt=1)
with self.assertRaises(TypeError):
NUTS(self.test_model,
GradLogPDFGaussian).generate_sample(initial_pos=0.1,
num_samples=1).send(None)
with self.assertRaises(ValueError):
NUTS(self.test_model,
GradLogPDFGaussian).generate_sample(initial_pos=(0, 1, 1, 1),
num_samples=1).send(None)
with self.assertRaises(TypeError):
NUTSda(self.test_model,
GradLogPDFGaussian).generate_sample(initial_pos=[[1, 2, 3]],
num_samples=1,
num_adapt=1).send(None)
with self.assertRaises(ValueError):
NUTSda(self.test_model,
GradLogPDFGaussian).generate_sample(initial_pos=[1],
num_samples=1,
num_adapt=1).send(None)
def test_sampling(self):
np.random.seed(1010101)
samples = self.nuts_sampler.sample(initial_pos=[-0.4, 1, 3.6],
num_adapt=0,
num_samples=10000,
return_type='recarray')
sample_array = np.array(
[samples[var_name] for var_name in self.test_model.variables])
sample_covariance = np.cov(sample_array)
self.assertTrue(
np.linalg.norm(sample_covariance - self.test_model.covariance) < 3)
np.random.seed(1210161)
samples = self.nuts_sampler.generate_sample(initial_pos=[-0.4, 1, 3.6],
num_adapt=0,
num_samples=10000)
samples_array = np.array([sample for sample in samples])
sample_covariance = np.cov(samples_array.T)
self.assertTrue(
np.linalg.norm(sample_covariance - self.test_model.covariance) < 3)
np.random.seed(12313131)
samples = self.nuts_sampler.sample(initial_pos=[0.2, 0.4, 2.2],
num_adapt=10000,
num_samples=10000)
sample_covariance = np.cov(samples.values.T)
self.assertTrue(
np.linalg.norm(sample_covariance -
self.test_model.covariance) < 0.4)
np.random.seed(921312312)
samples = self.nuts_sampler.generate_sample(
initial_pos=[0.2, 0.4, 2.2], num_adapt=10000, num_samples=10000)
samples_array = np.array([sample for sample in samples])
sample_covariance = np.cov(samples_array.T)
self.assertTrue(
np.linalg.norm(sample_covariance -
self.test_model.covariance) < 0.4)
def tearDown(self):
del self.test_model
del self.nuts_sampler
def setUp(self):
mean = np.array([-1, 1, 0])
covariance = np.array([[6, 0.7, 0.2], [0.7, 3, 0.9], [0.2, 0.9, 1]])
self.test_model = JGD(['x', 'y', 'z'], mean, covariance)
self.nuts_sampler = NUTSda(model=self.test_model, grad_log_pdf=GradLogPDFGaussian)
class TestNUTSInference(unittest.TestCase):
def setUp(self):
mean = np.array([-1, 1, 0])
covariance = np.array([[6, 0.7, 0.2], [0.7, 3, 0.9], [0.2, 0.9, 1]])
self.test_model = JGD(['x', 'y', 'z'], mean, covariance)
self.nuts_sampler = NUTSda(model=self.test_model, grad_log_pdf=GradLogPDFGaussian)
def test_errors(self):
with self.assertRaises(TypeError):
NUTS(model=self.test_model, grad_log_pdf=JGD)
with self.assertRaises(TypeError):
NUTS(model=self.test_model, grad_log_pdf=None, simulate_dynamics=GradLogPDFGaussian)
with self.assertRaises(ValueError):
NUTSda(model=self.test_model, delta=-0.2, grad_log_pdf=None)
with self.assertRaises(ValueError):
NUTSda(model=self.test_model, delta=1.1, grad_log_pdf=GradLogPDFGaussian)
with self.assertRaises(TypeError):
NUTS(self.test_model, GradLogPDFGaussian).sample(initial_pos={1, 1, 1}, num_samples=1)
with self.assertRaises(ValueError):
NUTS(self.test_model, GradLogPDFGaussian).sample(initial_pos=[1, 1], num_samples=1)
with self.assertRaises(TypeError):
NUTSda(self.test_model, GradLogPDFGaussian).sample(initial_pos=1, num_samples=1, num_adapt=1)
with self.assertRaises(ValueError):
NUTSda(self.test_model, GradLogPDFGaussian).sample(initial_pos=[1, 1, 1, 1], num_samples=1, num_adapt=1)
with self.assertRaises(TypeError):
NUTS(self.test_model, GradLogPDFGaussian).generate_sample(initial_pos=0.1, num_samples=1).send(None)
with self.assertRaises(ValueError):
NUTS(self.test_model, GradLogPDFGaussian).generate_sample(initial_pos=(0, 1, 1, 1),
num_samples=1).send(None)
with self.assertRaises(TypeError):
NUTSda(self.test_model, GradLogPDFGaussian).generate_sample(initial_pos=[[1, 2, 3]], num_samples=1,
num_adapt=1).send(None)
with self.assertRaises(ValueError):
NUTSda(self.test_model, GradLogPDFGaussian).generate_sample(initial_pos=[1], num_samples=1,
num_adapt=1).send(None)
def test_sampling(self):
np.random.seed(1010101)
samples = self.nuts_sampler.sample(initial_pos=[-0.4, 1, 3.6], num_adapt=0, num_samples=10000,
return_type='recarray')
sample_array = np.array([samples[var_name] for var_name in self.test_model.variables])
sample_covariance = np.cov(sample_array)
self.assertTrue(np.linalg.norm(sample_covariance - self.test_model.covariance) < 3)
np.random.seed(1210161)
samples = self.nuts_sampler.generate_sample(initial_pos=[-0.4, 1, 3.6], num_adapt=0, num_samples=10000)
samples_array = np.array([sample for sample in samples])
sample_covariance = np.cov(samples_array.T)
self.assertTrue(np.linalg.norm(sample_covariance - self.test_model.covariance) < 3)
np.random.seed(12313131)
samples = self.nuts_sampler.sample(initial_pos=[0.2, 0.4, 2.2], num_adapt=10000, num_samples=10000)
sample_covariance = np.cov(samples.values.T)
self.assertTrue(np.linalg.norm(sample_covariance - self.test_model.covariance) < 0.4)
np.random.seed(921312312)
samples = self.nuts_sampler.generate_sample(initial_pos=[0.2, 0.4, 2.2], num_adapt=10000, num_samples=10000)
samples_array = np.array([sample for sample in samples])
sample_covariance = np.cov(samples_array.T)
self.assertTrue(np.linalg.norm(sample_covariance - self.test_model.covariance) < 0.4)
def tearDown(self):
del self.test_model
del self.nuts_sampler