GraphSage Embedding Issue

Hi, I was wondering that GraphSage SageEncoder use_id=True, which will look_up table and then get the embedding vector of input.

My question is i don't see stop_gradient in the embedding process, so i suppose that the embedding will update with training process, but i think in the paper of GraphSage, the embedding is the input(don't update).

It's a bug?

Start by locating the SageEncoder path for use_id=True and tracing whether the lookup embedding receives gradients during training. Compare that behavior with the GraphSAGE paper's treatment of input embeddings; done means the expected behavior is confirmed and covered by the relevant test or documentation.

Written by the indexing model from the issue text.