Very different testing result each time with everything unchanged

kyzhouhzau / BERT-NER

Use Google's BERT for named entity recognition （CoNLL-2003 as the dataset）.

MIT License

1.23k stars 335 forks source link

Very different testing result each time with everything unchanged #60

Open Wanjun0511 opened 5 years ago

Wanjun0511 commented 5 years ago

I'm doing the sentence tagging task. I repeat my experiment with same training and testing data, and exactly same code, just repeat. However, get different predictions every time. The testing accuracy ranges from 84.9% to 90.3%.

I've tried to add random seed but no use at all. Anyone has the same problem or any suggestions ?