search

tanatosuu / GDE

SIGIR 2022 Paper: Less is More: Reweighting Important Spectral Graph Features for Recommendation
31 stars 5 forks source link

loss is nan #1

Open CreaterLL opened 2 years ago

CreaterLL commented 2 years ago

Hi,i'm very interested in your resaerch ,but when i try to reproduce the results in the thesis,i encountered a irregular phenomenon as follows: image

i'm confuesed with the "loss=nan" there are any instructions for me? thank you!

tanatosuu commented 2 years ago

Hi, thanks for your question. Although drop_out leads to improvement, it also causes a unstable training, you can reduce the learning rate or drop_out ratio. drop_out ratio setting to 0 can make the model training more stable, but also sacrifices some improvement. Feel free to post any further questions.

CreaterLL commented 2 years ago

Thank you for your instructions,i 'll have a try later.