LUOyk1999 / tunedGNN

[NeurIPS 2024] Implementation of "Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification"
https://neurips.cc/virtual/2024/poster/97440
73 stars 6 forks source link

Question about POLYNORMER #4

Open W-rudder opened 3 hours ago

W-rudder commented 3 hours ago

您好,我注意到您的代码中对POLYNORMER的Global attention设置为False(不启用),请问为什么这样设置?这样会导致POLYNORMER的效果变差吗?

LUOyk1999 commented 3 hours ago

您好,这是来源自https://github.com/cornell-zhang/Polynormer/blob/master/model.py#L92 代码仅供示例,具体两种情况我都调过,一种是不开启,另外一种是训练一段epoch后再开启,我汇报的是更好的结果

W-rudder commented 3 hours ago

您好,这是来源自https://github.com/cornell-zhang/Polynormer/blob/master/model.py#L92 代码仅供示例,具体两种情况我都调过,一种是不开启,另外一种是训练一段epoch后再开启,我汇报的是更好的结果

感谢您的回复,您的文章给出了GraphSAGE,GAT,GCN的最佳参数,请问您有保存GT_baseliens的最佳参数吗?是否可以麻烦您提供一下

LUOyk1999 commented 3 hours ago

GT_baseliens的最佳参数都在原论文的最佳参数附近,例如POLYNORMER的在这: https://github.com/cornell-zhang/Polynormer/blob/master/run.sh https://github.com/LUOyk1999/tunedGNN/blob/main/medium_graph/GTs_baselines/polynormer_fixed.sh

W-rudder commented 3 hours ago

GT_baseliens的最佳参数都在原论文的最佳参数附近,例如POLYNORMER的在这: https://github.com/cornell-zhang/Polynormer/blob/master/run.sh https://github.com/LUOyk1999/tunedGNN/blob/main/medium_graph/GTs_baselines/polynormer_fixed.sh

十分重要的工作!感谢!

LUOyk1999 commented 3 hours ago

也感谢您的支持!