torch.eig is deprecated for a long time and is being removed

[kit1980]

PyTorch's torch.eig was deprecated since version 1.9 and is being removed by pytorch/pytorch#70982. Please use the torch.linalg.eig function instead if you want your code to continue to work with the latest PyTorch.

Affected file:

neural-scs/automl21/enr.py

Line 242 in 29743a1

L = torch.eig(ATA).eigenvalues[:,0].max() + mu/2.

[lezcano]

Interestingly enough, you can save the ATA multiplication there, and simply write that as

L = torch.svdvals(A).max() ** 2 + mu/2

. For the why, see https://en.wikipedia.org/wiki/Singular_value_decomposition#Relation_to_eigenvalue_decomposition
This is a much more stable and efficient way to compute this quantity.