search

SwordYork / DCNMT

Deep Character-Level Neural Machine Translation
GNU General Public License v3.0
72 stars 19 forks source link

Is there a faster way to evaluate the BLEU score ? #8

Closed kadir-gunel closed 7 years ago

kadir-gunel commented 7 years ago

Hello @SwordYork ,

Since there are multiple models for a training, I want to get BLEU score of each model. The testing.py prints each translation and its score one by one. Is there a way to just print the BLEU score without printing out the scores?

SwordYork commented 7 years ago

Sorry, I think the printing of score (cost) is not related to speed. The sampling process has not been optimized, thus may be slower. You could adjust beam_size or the validation size. It may take 60 min to translate 3000 sentences.