Sean-Blank / AMRcoref

9 stars 1 forks source link

Reproducing reported numbers #3

Open wenchaodudu opened 2 years ago

wenchaodudu commented 2 years ago

Hello,

I ran the training and eval script for one time and got F1 at around 54, which is much lower than the reported number in the paper, 62. Is that number obtained by choosing the best across multiple runs? If yes, how many times do you have to run?