search

yangheng95 / LCF-ATEPC

codes for paper A Multi-task Learning Model for Chinese-oriented Aspect Polarity Classification and Aspect Term Extraction
MIT License
191 stars 45 forks source link

Evaluation #35

Closed Astudnew closed 3 years ago

Astudnew commented 3 years ago

I have question related to evaluation, why you evaluate the model for every 20 steps? Also, what is the difference in each train (seed from1-5)? what happened on each train?

yangheng95 commented 3 years ago

Mainly bcause the performance of APC and ATE do not reach the peak at the same steps, and performance will be differennt by using different random seed.