Eigen Decomposition #71
Description
Hi! I noticed that the author used the eigh function from scipy.linalg for eigen decomposition. I want to speed up the computation, so I tried replacing it with the equivalent function from PyTorch/NumPy. However, I found that the results were significantly different from those obtained with SciPy, especially in subgraph segmentation tasks.
Could it be that the eigen decomposition in SciPy is more reliable and stable?