search

tkipf / gcn

Implementation of Graph Convolutional Networks in TensorFlow
MIT License
7.09k stars 2k forks source link

about the training loss #31

Open gelang93 opened 6 years ago

gelang93 commented 6 years ago

When my label is a high-dimensional vector, and the label is not sparse. I find that the training loss has not been reduced（or the reduction is particularly small）,and the end result is not good.how to solve the problem ？ Thanks

tkipf commented 6 years ago

Sounds like a regression problem to me. Have you tried mean-squared error (MSE) instead of cross entropy?

On 7 May 2018, at 09:45, gelang93 notifications@github.com wrote:

When my label is a high-dimensional vector, and the label is not sparse. I find that the training loss has not been reduced（or the reduction is particularly small）,and the end result is not good.how to solve the problem ？ Thanks

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/tkipf/gcn/issues/31, or mute the thread https://github.com/notifications/unsubscribe-auth/AHAcYKnTb7sW1CrPs6TalakqVaSP2GxIks5tv_uvgaJpZM4T0iom.