Problems occur when obtaining eigenvalues and eigenvectors from voxel data

[msmhi52]

Hi, Xutong.

I intermittently faced two issues when I tried to obtain eigenvalues and eigenvectors from voxel data.

First, pytorch.lobpcg invokes an assertion error, which is stated as

"AssertionError: the number of converged eigenpairs (was 1, got 0) cannot decrease"

Second, pytorch.lobpcg invokes a value error, which is stated as

ValueError: LPBPCG algorithm is not applicable when the number of A rows (=72) is smaller than 3 x the number of requested eigenpairs (=26)

I downloaded ten ABC dataset chunks (abc_0000_obj_v00.7z - abc_0009_obj_v00.7z) and tried to reproduce the result demonstrated in the NeuralSound paper. Is there any method to solve these issues? Also, which ABC dataset chucks did you use for the paper?

Thanks.