THUDM / GraphMAE

GraphMAE: Self-Supervised Masked Graph Autoencoders in KDD'22
478 stars 75 forks source link

Find implement issue in PyG version #47

Closed fs302 closed 1 year ago

fs302 commented 1 year ago

Hi, I have encountered an error while running the PyG version of GraphMAE using the command python main_transductive.py --dataset cora --encoder gat --decoder gat --seed 0 --device 0. The error message states:

File "../GraphMAE/graphmae/evaluation.py", line 52, in linear_probing_for_transductive_node_classiifcation     
out = model(graph, x) 
TypeError: dropout(): argument 'input' (position 1) must be Tensor, not Data 

While it works fine by sh scripts/run_transductive.sh cora 0. I think it is because the args setting linear_prob cause the model mis-assign in linear_probing_for_transductive_node_classiifcation.

THINK2TRY commented 1 year ago

@fs302 Thanks for your attention, and sorry for the late response. You are right as the evaluation procedures are different depending on whether setting--linear_prob. Initially, we would like to implement finetuning without freezing GNN parameters but this setting is discarded in later evaluation. And the code is retained. Now all reported results are under the --linear_prob setting.