if torch.cuda.is_available():
del x_u, y_u, x_f, y_f, u, u_x, u_xx, k, pred_u, pred_k, pred_f
torch.cuda.empty_cache()
return ll, output
# sampling
params_hmc = util.sample_model_bpinns(nets,
data,
model_loss=model_loss,
num_samples=num_samples,
num_steps_per_sample=L,
step_size=step_size,
burn=burn,
tau_priors=tau_priors,
tau_likes=tau_likes,
device=device,
n_params_single=n_params_single,
pde=pde,
pinns=pinns,
epochs=epochs)
pred_list, log_prob_list = util.predict_model_bpinns(
nets,
params_hmc,
data_val,
model_loss=model_loss,
tau_priors=tau_priors,
tau_likes=tau_likes,
n_params_single=n_params_single,
elif method == 'BO':
del MSE_BO
torch.cuda.empty_cache()
return ll, output
# regression
params_init_val = None
params_hmc = util.sample_model_bpinns(nets,
data,
model_loss=model_loss,
step_size=step_size,
device=device,
pde=True,
pinns=True,
epochs=epochs,
params_init_val=params_init_val)
pred_list, log_prob_list = util.predict_model_bpinns(nets,
params_hmc,
data_val,
model_loss=model_loss,
pde=True)
print('\nExpected validation log probability: {:.3f}'.format(
torch.stack(log_prob_list).mean()))
pred_list_u = pred_list[0].cpu().numpy()