SQUAD2.Q-Augmented-Dataset icon indicating copy to clipboard operation
SQUAD2.Q-Augmented-Dataset copied to clipboard

Published 20 hours ago • verified publisher icon ankit-ai

each question in v2.q exists twice, and the original question has been lost

Open freefuiiismyname opened this issue 5 years ago • 2 comments

Thank you for your work! When I analyzed the data, I found that each question in v2.q existed twice, and the original question had been lost.Maybe your code need a 'deepcopy' function, to ensure original question not be covered.

freefuiiismyname avatar Dec 13 '19 07:12 freefuiiismyname

And i tried v2.q, v2q50 with BERT-base, the score of dev set lower about 1 point than raw dataset

freefuiiismyname avatar Dec 18 '19 02:12 freefuiiismyname

Yeah it looks like he's copying the dict in line 51 of augment.py. Should instead use a deepcopy.

akfuster avatar Mar 08 '21 17:03 akfuster