search

pyg-team / pytorch_geometric

Graph Neural Network Library for PyTorch
https://pyg.org
MIT License
21.28k stars 3.65k forks source link

GraphSAGE example #1460

Closed yuanx749 closed 4 years ago

yuanx749 commented 4 years ago

❓ Questions & Help

Hi, In examples/reddit.py (line 61), why is relu not applied to x in the last layer in the inference method? Shouldn't activation function be used at each layer?

rusty1s commented 4 years ago

The last layer typically does not contain an activation function since you do not want to restrict the output of your model, e.g., for regression. For inference, we are only interested in the argmax of the output, and this does not get changed by the log_softmax call, so this operation is not really of need here.

yuanx749 commented 4 years ago

Thank you, you're right. I was thinking in unsupervised learning like Deep Graph Infomax, whether activation function should be used in the last layer to get the node representation. I will try both to see which gets better performance.