def test_kl_divergence(self):
model1 = beer.NormalGammaPrior(self.mean, self.precision,
self.prior_count)
model2 = beer.NormalGammaPrior(self.mean, self.precision,
self.prior_count)
div = beer.kl_div(model1, model2)
self.assertAlmostEqual(div, 0.)
def create_modelset_diag(ncomps, dim, type_t):
posts = [
beer.NormalGammaPrior(
torch.zeros(dim).type(type_t),
torch.ones(dim).type(type_t), 1.) for _ in range(ncomps)
]
modelset = beer.NormalDiagonalCovarianceSet(
beer.NormalGammaPrior(
torch.zeros(dim).type(type_t),
torch.ones(dim).type(type_t), 1.), posts)
return modelset
def create_normalgamma(t_type):
dim = int(1 + torch.randint(100, (1, 1)).item())
mean = torch.randn(dim).type(t_type)
scale = (1 + torch.randn(dim)**2).type(t_type)
shape = (1 + torch.randn(dim)**2).type(t_type)
rate = (1 + torch.randn(dim)**2).type(t_type)
return beer.NormalGammaPrior(mean, scale, shape, rate)
def test_log_norm(self):
model = beer.NormalGammaPrior(self.mean, self.precision,
self.prior_count)
model_log_norm = model.log_norm.numpy()
natural_params = model.natural_params.numpy()
log_norm = normalgamma_log_norm(natural_params)
self.assertAlmostEqual(model_log_norm, log_norm)
def setUp(self):
self.dim = int(1 + torch.randint(100, (1, 1)).item())
self.mean = torch.randn(self.dim).type(self.type)
self.scale = (1 + torch.randn(self.dim)**2).type(self.type)
self.shape = (1 + torch.randn(self.dim)**2).type(self.type)
self.rate = (1 + torch.randn(self.dim)**2).type(self.type)
self.model = beer.NormalGammaPrior(self.mean, self.scale,
self.shape, self.rate)
def test_exp_sufficient_statistics(self):
model = beer.NormalGammaPrior(self.mean, self.precision,
self.prior_count)
model_s_stats = model.expected_sufficient_statistics.numpy()
natural_params = model.natural_params.numpy()
s_stats = normalgamma_grad_log_norm(natural_params)
self.assertTrue(np.allclose(model_s_stats, s_stats, rtol=TOL,
atol=TOL))
def test_create(self):
model = beer.NormalGammaPrior(self.mean, self.precision,
self.prior_count)
self.assertTrue(isinstance(model, beer.ExpFamilyDensity))