You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi,
It is written in your AAAI paper "To generate the answer vocabulary, we choose the top 1,000 most frequent answers in the train split as the answer vocabulary, which covers 84.7% / 86.2% / 85.6% of the train/val/test answers, respectively".
However, choosing the same vocabulary, I obtain significantly lower coverage ( about 78% for test and 80% for val). Do you have an idea why? Do you process further the answers?
Best,
Antoine Yang
The text was updated successfully, but these errors were encountered:
Hi,
It is written in your AAAI paper "To generate the answer vocabulary, we choose the top 1,000 most frequent answers in the train split as the answer vocabulary, which covers 84.7% / 86.2% / 85.6% of the train/val/test answers, respectively".
However, choosing the same vocabulary, I obtain significantly lower coverage ( about 78% for test and 80% for val). Do you have an idea why? Do you process further the answers?
Best,
Antoine Yang
The text was updated successfully, but these errors were encountered: