search

shuaishiliu / SGCN

Code for "SGCN:Sparse Graph Convolution Network for Pedestrian Trajectory Prediction" CVPR 2021
MIT License
88 stars 20 forks source link

Is the smaller the loss the better,or the loss close to 0 the better? #11

Closed AnsonDrew7 closed 2 years ago

AnsonDrew7 commented 2 years ago

For example,loss =-0.002 and loss =-0.033,which better?

shuaishiliu commented 2 years ago

For example,loss =-0.002 and loss =-0.033,which better?

Hi, the negative log likelihood loss function will be negative when the pdf is greater than 1. So, the more small loss more better, but be careful overfitting.