search

BangHonor / SimGC

Simple Graph Condensation
8 stars 2 forks source link

Hyperparameters about Cora and Citeseer #1

Closed rockcor closed 1 month ago

rockcor commented 2 months ago

Thanks for your consice solution on Graph Condensation. I locally run your code and most of results reported on paper can be achieved.

The question is I cannot reproduce Cora and Citeseer. I got 76 of Cora and 62 of Citeseer. Also, the hyperparameters of Citeseer are missed in the code.

BangHonor commented 2 months ago

Hello ,thanks for your interest in our work. Please make sure that on these two datasets, you are using the SGC_Multi_PYG model, which contains a MLP instead of Linear layer inside. Besides, these two datasets are too small and they are highly sensitive to hyperparameters, so you can further tune the layers/hidden units of SGC and the evalutaion GNN, condensing_loop and leanring rates.

Original Email

From:"Rock"< @.*** >;

Sent Time:2024/8/8 10:25

To:"BangHonor/SimGC"< @.*** >;

Cc recipient:"Subscribed"< @.*** >;

Subject:[BangHonor/SimGC] Hyperparameters about Cora and Citeseer (Issue #1)

Thanks for your consice solution on Graph Condensation. I locally run your code and most of results reported on paper can be achieved.

The question is I cannot reproduce Cora and Citeseer. I got 76 of Cora and 62 of Citeseer. Also, the hyperparameters of Citeseer are missed in the code.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>

BangHonor commented 2 months ago

Besides, the citeseer hypermeters are the same as cora. If you get bad results using the same hyperparameters, please pay attention the to points I mentioned in the last e-mail. The results of these two datasets are rather unstable and they are too small to proceed graph condensation. 

Original Email

From:"Rock"< @.*** >;

Sent Time:2024/8/8 10:25

To:"BangHonor/SimGC"< @.*** >;

Cc recipient:"Subscribed"< @.*** >;

Subject:[BangHonor/SimGC] Hyperparameters about Cora and Citeseer (Issue #1)

Thanks for your consice solution on Graph Condensation. I locally run your code and most of results reported on paper can be achieved.

The question is I cannot reproduce Cora and Citeseer. I got 76 of Cora and 62 of Citeseer. Also, the hyperparameters of Citeseer are missed in the code.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>