192b0dcede721b827e5347861204b88bbdaf4a52,gpytorch/lazy/kronecker_product_lazy_tensor.py,KroneckerProductLazyTensor,_symeig,#KroneckerProductLazyTensor#Any#,152

Before Change

        evals, evecs = [], []
        for lazy_tensor in self.lazy_tensors:
            // TODO: replace with lazy_tensor.symeig() once that is added in.
            evals_, evecs_ = lazy_tensor.evaluate().symeig(eigenvectors=eigenvectors)
            evals.append(evals_)
            evecs.append(evecs_)
        evals = KroneckerProductLazyTensor(*[DiagLazyTensor(evals_) for evals_ in evals])

After Change

            // TODO: replace with lazy_tensor.symeig() once that is added in.
            // TODO: ensure that the symeig call is also done in this manner

            eval_tensor = lazy_tensor.evaluate()
            tensor_dtype = eval_tensor.dtype

            evals_, evecs_ = eval_tensor.double().symeig(eigenvectors=eigenvectors)

            // we chop any negative eigenvalues
            neg_evals = zeros_like(evals_)
            neg_evals[evals_.data < 0] = -evals_.data[evals_.data < 0]
            evals_ = evals_ + neg_evals

            evals_ = evals_.type(tensor_dtype)
            evecs_ = evecs_.type(tensor_dtype)

            evals.append(evals_)
            evecs.append(evecs_)

Instances