def test_classification_error_cuda(self):
if not torch.cuda.is_available():
return
with least_used_cuda_device():
train_x, train_y = train_data(cuda=True)
likelihood = BernoulliLikelihood().cuda()
model = GPClassificationModel(train_x).cuda()
mll = gpytorch.mlls.VariationalMarginalLogLikelihood(likelihood, model, num_data=len(train_y))
# Find optimal model hyperparameters
model.train()
optimizer = optim.Adam(model.parameters(), lr=0.1)
optimizer.n_iter = 0
for _ in range(75):
optimizer.zero_grad()
output = model(train_x)
loss = -mll(output, train_y)
loss.backward()
optimizer.n_iter += 1
optimizer.step()
for param in model.parameters():
self.assertTrue(param.grad is not None)
self.assertGreater(param.grad.norm().item(), 0)
for param in likelihood.parameters():
self.assertTrue(param.grad is not None)
self.assertGreater(param.grad.norm().item(), 0)
optimizer.step()
# Set back to eval mode
model.eval()
test_preds = likelihood(model(train_x)).mean.round()
mean_abs_error = torch.mean(torch.abs(train_y - test_preds) / 2)
self.assertLess(mean_abs_error.item(), 1e-5)
def test_posterior_with_exact_computations_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
with gpytorch.settings.fast_computations(
covar_root_decomposition=False, log_prob=False):
self.test_posterior_latent_gp_and_likelihood_with_optimization(
cuda=True)
def test_regression_error_cuda(self):
if not torch.cuda.is_available():
return
with least_used_cuda_device():
train_x, train_y = train_data(cuda=True)
likelihood = GaussianLikelihood().cuda()
inducing_points = torch.linspace(0, 1,
25).unsqueeze(-1).repeat(2, 1, 1)
model = SVGPRegressionModel(inducing_points).cuda()
mll = gpytorch.mlls.VariationalELBO(likelihood,
model,
num_data=train_y.size(-1))
# Find optimal model hyperparameters
model.train()
likelihood.train()
optimizer = optim.Adam([{
"params": model.parameters()
}, {
"params": likelihood.parameters()
}],
lr=0.01)
for _ in range(150):
optimizer.zero_grad()
output = model(train_x)
loss = -mll(output, train_y)
loss = loss.sum()
loss.backward()
optimizer.step()
for param in model.parameters():
self.assertTrue(param.grad is not None)
self.assertGreater(param.grad.norm().item(), 0)
for param in likelihood.parameters():
self.assertTrue(param.grad is not None)
self.assertGreater(param.grad.norm().item(), 0)
# Set back to eval mode
model.eval()
likelihood.eval()
test_preds = likelihood(model(train_x)).mean.squeeze()
mean_abs_error = torch.mean(
torch.abs(train_y[0, :] - test_preds[0, :]) / 2)
mean_abs_error2 = torch.mean(
torch.abs(train_y[1, :] - test_preds[1, :]) / 2)
self.assertLess(mean_abs_error.item(), 1e-1)
self.assertLess(mean_abs_error2.item(), 1e-1)
def test_kissgp_gp_mean_abs_error_cuda(self):
if not torch.cuda.is_available():
return
with least_used_cuda_device():
train_x, train_y, test_x, test_y = make_data(cuda=True)
likelihood = FixedNoiseGaussianLikelihood(torch.ones(100) *
0.001).cuda()
gp_model = GPRegressionModel(train_x, train_y, likelihood).cuda()
mll = gpytorch.mlls.ExactMarginalLogLikelihood(
likelihood, gp_model)
# Optimize the model
gp_model.train()
likelihood.train()
optimizer = optim.Adam(list(gp_model.parameters()) +
list(likelihood.parameters()),
lr=0.1)
optimizer.n_iter = 0
with gpytorch.settings.debug(False):
for _ in range(25):
optimizer.zero_grad()
output = gp_model(train_x)
loss = -mll(output, train_y)
loss.backward()
optimizer.n_iter += 1
optimizer.step()
for param in gp_model.parameters():
self.assertTrue(param.grad is not None)
self.assertGreater(param.grad.norm().item(), 0)
for param in likelihood.parameters():
self.assertTrue(param.grad is not None)
self.assertGreater(param.grad.norm().item(), 0)
# Test the model
gp_model.eval()
likelihood.eval()
test_preds = likelihood(gp_model(test_x)).mean
mean_abs_error = torch.mean(torch.abs(test_y - test_preds))
self.assertLess(mean_abs_error.squeeze().item(), 0.02)
def test_gp_posterior_mean_skip_variances_slow_cuda(self):
if not torch.cuda.is_available():
return
with least_used_cuda_device():
train_x, test_x, train_y, _ = self._get_data(cuda=True)
likelihood = GaussianLikelihood()
gp_model = ExactGPModel(train_x, train_y, likelihood)
gp_model.cuda()
likelihood.cuda()
# Compute posterior distribution
gp_model.eval()
likelihood.eval()
with gpytorch.settings.fast_pred_var(False):
with gpytorch.settings.skip_posterior_variances(True):
mean_skip_var = gp_model(test_x).mean
mean = gp_model(test_x).mean
likelihood_mean = likelihood(gp_model(test_x)).mean
self.assertTrue(torch.allclose(mean_skip_var, mean))
self.assertTrue(torch.allclose(mean_skip_var, likelihood_mean))
def test_gauss_hermite_quadrature_1D_normal_batch_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_gauss_hermite_quadrature_1D_normal_nonbatch(
cuda=True)
def test_multitask_gp_mean_abs_error_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_multitask_gp_mean_abs_error(cuda=True)
def test_simple_model_list_gp_regression_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_simple_model_list_gp_regression(cuda=True)
def test_regression_error_skip_logdet_forward_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_regression_error(skip_logdet_forward=True, cuda=True)
def test_lkj_covariance_prior_batch_log_prob_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_lkj_covariance_prior_batch_log_prob(cuda=True)
def test_posterior_latent_gp_and_likelihood_with_optimization_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_posterior_latent_gp_and_likelihood_with_optimization(
cuda=True)
def test_kl_divergence_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_kl_divergence(cuda=True)
def test_multivariate_normal_batch_non_lazy_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_multivariate_normal_batch_non_lazy(cuda=True)
def test_gamma_prior_log_prob_log_transform_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
return self.test_gamma_prior_log_prob_log_transform(cuda=True)
def test_regression_error_cuda(self):
if not torch.cuda.is_available():
return
with least_used_cuda_device():
return self.test_regression_error(cuda=True)
def test_psd_safe_cholesky_psd_cuda(self, cuda=False):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_psd_safe_cholesky_psd(cuda=True)
def test_smoothed_box_prior_log_prob_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
return self.test_smoothed_box_prior_log_prob(cuda=True)
def test_multivariate_normal_batch_correlated_sampels_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_multivariate_normal_batch_correlated_sampels(
cuda=True)
def test_normal_prior_batch_log_prob_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
return self.test_normal_prior_batch_log_prob(cuda=True)
def test_posterior_latent_gp_and_likelihood_fast_pred_var_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_posterior_latent_gp_and_likelihood_fast_pred_var(
cuda=True)
def test_fantasy_updates_batch_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_fantasy_updates_batch(cuda=True)
def test_grid_gp_mean_abs_error_2d_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_grid_gp_mean_abs_error(cuda=True, num_dim=2)
def test_from_independent_mvns_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_from_independent_mvns(cuda=True)
def test_lkj_cholesky_factor_prior_log_prob_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_lkj_cholesky_factor_prior_log_prob(cuda=True)
def test_prior_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_prior(cuda=True)
def test_multitask_multivariate_normal_cuda(self):
if torch.cuda.is_available():
with least_used_cuda_device():
self.test_multitask_multivariate_normal(cuda=True)