search

wujcan / SGL-TensorFlow

173 stars 43 forks source link

When will you release the codes? #1

Closed Coder-Yu closed 3 years ago

Coder-Yu commented 3 years ago

Can't wait to see the original implementation. I have implemented a version but it does not work well. Hope the codes can be released soon.

Coder-Yu commented 3 years ago

Thanks for sharing the codes. I have successfully implemented SGL by referring to your implementation. The loss curves and performance improvments over LightGCN are almost as the same as the results reported in your paper. Very reliable experimental details. However, it seems that, LightGCN can converge faster by l2_normalizing embeddings learned at each layer. While in your implementation, I notice that the layer embeddings are not normalized.