You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am trying to reproduce the results of dropedge on semi-supervised setting. I used scripts in script/semi-supervised/ to run the program, but I can't get the results showned in readme. I get around 81 accuracy on Cora with 4 layers JKnet(dropedge), not 83.3. How can I reproduce the results shown in readme?
And have you tried to calculate the mean accuracy using different random weight initializations which used in GCN and GAT?
Hi Weeley743,
To exactly reproduce the result, you need to satisfy the requirements listed in requirements.txt as well as python version. I've tried to re-run the scripts on my own machine. The logs are attached in the end.
We haven't tested the mean accuracy using different random seeds. There should be some variances among the results from the different random seeds and different environments.
Activity
Weesley743 commentedon Jan 18, 2020
I used CUDA 9.0.176 (same as you) and I got 67.9 test acc. on citeseer with 16 layers JKnet(dropedge)
DropEdge commentedon Jan 21, 2020
Hi Weeley743,
To exactly reproduce the result, you need to satisfy the requirements listed in requirements.txt as well as python version. I've tried to re-run the scripts on my own machine. The logs are attached in the end.
We haven't tested the mean accuracy using different random seeds. There should be some variances among the results from the different random seeds and different environments.
Weesley743 commentedon Jan 21, 2020
I got 83.3 acc. after configuring the environment in requirements.txt
thanks