9708a6eaeed14443c9ccf26594155767fcbf83d6,gpytorch/models/exact_gp.py,ExactGP,__call__,#ExactGP#,57

Before Change

                if not isinstance(test_test_covar, LazyVariable):
                    test_test_covar = NonLazyVariable(test_test_covar)
                covar_correction = test_train_covar.matmul(self.chol)
                predictive_covar = test_test_covar + CholLazyVariable(covar_correction).mul(-1)
            else:
                if isinstance(train_test_covar, LazyVariable):
                    train_test_covar = train_test_covar.evaluate()
                if isinstance(test_train_covar, LazyVariable):

After Change

                init_vector /= torch.norm(init_vector, 2, 0)
                q_mat, t_mat = lq_object.lanczos_batch(train_train_matmul, init_vector)
                self.low_rank_left = Variable(q_mat[0].matmul(t_mat[0].inverse()))
                self.low_rank_right = Variable(q_mat[0].transpose(-1, -2))
                self.has_computed_low_rank = True

            // Calculate mean
            if isinstance(full_covar, InterpolatedLazyVariable):
                left_interp_indices = test_train_covar.left_interp_indices
                left_interp_values = test_train_covar.left_interp_values
                predictive_mean = left_interp(left_interp_indices, left_interp_values, self.alpha) + test_mean
            elif isinstance(test_train_covar, LazyVariable):
                predictive_mean = test_train_covar.matmul(self.alpha) + test_mean
            else:
                predictive_mean = torch.addmv(test_mean, test_train_covar, self.alpha)

            // Calculate covar
            if gpytorch.functions.fast_pred_var:
                if not isinstance(test_test_covar, LazyVariable):
                    test_test_covar = NonLazyVariable(test_test_covar)
                covar_correction_left = test_train_covar.matmul(self.low_rank_left)
                covar_correction_right = test_train_covar.matmul(self.low_rank_right.transpose(-1, -2))
                covar_correction_right = covar_correction_right.transpose(-1, -2)
                covar_correction = MatmulLazyVariable(covar_correction_left, covar_correction_right).mul(-1)
                predictive_covar = test_test_covar + covar_correction
            else:
                if isinstance(train_test_covar, LazyVariable):
                    train_test_covar = train_test_covar.evaluate()
                if isinstance(test_train_covar, LazyVariable):

Instances