Why the baseline performance is different with the original version?
liu-jc opened this issue · 1 comments
liu-jc commented
In the original version of GraphSAGE, the micro f1 score after 50 epochs in PPI dataset is only around 0.57. I don't modify the args in the shell script. In your experiments using uniform sampling, I use the exactly same args and the performance can achieve over 0.65. But why? I think they should be similar. Have you modified evaluation settings or done any improvement for uniform sampling?
oj9040 commented
In original GraphSAGE, the default hidden dimension is 128 if you did not specified additionally. However, my shell script specified to use 512. Maybe this caused the difference.