DushyantaDhyani / kdtf

Knowledge Distillation using Tensorflow
MIT License
142 stars 47 forks source link

checkpoint condition should be modified #2

Open minxueric opened 5 years ago

minxueric commented 5 years ago

Thanks for your clear implementation of KD. There is a little mistake about the condition that a new epoch should be saved as checkpoint. You should assign current value of "acc" to "max_accuracy", after each checkpoint.

DushyantaDhyani commented 5 years ago

Surprisingly missed that!! Feel free to send a PR. Or I will see when I can get a chance to fix it