About symmetrically normalization of adjacency matrix

[empty-id]

Hi, I notice that in this PyTorch version code, the adjacency matrix is row-normalized instead of symmetrically normalized. However, the accuracy (82.5%) is higher than the TensorFlow version code (81.6%). Moreover, I also tried to symmetrically normalize the adjacency matrix in this PyTorch version, but the result dropped (to 79.9%). Nevertheless, result of TensorFlow version does not change after modification of normalization. For summarization, this is the experiments I did:

Cora dataset	TensorFlow	PyTorch
Symmetrically Normalization	81.6	79.9
Row Normalization	81.6	82.5

Is there any idea why does this happen?

[empty-id]

I find some differences between pytorch and tf version.

• In layers.py, pytorch version should use glorot initialization for weights and zero initialization for bias
• In models.py, pytorch doesn't perform dropout (or sparse_dropout) for the input x first.

After I modified the above two points, the result now is ~82% for symmetrically normalization in pytorch version.