Explanation scores using Remove and debias is none

[MasterEndless]

trafficstars

During the reproduction of the blog: [https://jacobgil.github.io/pytorch-gradcam-book/CAM%20Metrics%20And%20Tuning%20Tutorial.html#road-remove-and-debias], I found the scores calculated by cam_metric using remove and debias is always none. I have located the source of the error, the "NoisyLinearImputer" function will produce a super large output tensor that will cause the target model to create a very large posterior. Do you have any ideas on how to solve this problem? I guess it may be due to the following code: res = torch.tensor(spsolve(csc_matrix(A), b), dtype=torch.float) Will this equation vulnerable to some numerical instability, that will output large values? Thanks in advance!