search

PetarV- / DGI

Deep Graph Infomax (https://arxiv.org/abs/1809.10341)
MIT License
630 stars 135 forks source link

Mean accuracy on fixed representation #19

Closed nihuo76 closed 2 years ago

nihuo76 commented 2 years ago

HI, great work ! But I have a question regarding how to repeat the experiments. While getting the mean accuracy, you used the fixed representation and only repeated the logistic regression part. Shall we repeat the whole pretext training + downstream task altogether ? Or is there any reference for the reason of doing so ? Many thanks

PetarV- commented 2 years ago

Hi Hexin,

Thank you for your issue, and kind comments on our work. You should repeat the entire training procedure for every random seed -- at each seed, averaging over many repeated logistic regressions. If I remember correctly, I took the average-of-averages to get the number in the paper, but it's been a while!

Hope that helps.

Thanks, Petar

nihuo76 commented 2 years ago

Hi Petar, thank you so much !