Flair: Why TP is always equal to TN in the final test?
Hi! I've seen that in the final test (evaluation of NER model after training) TP is always equal to the TN. Why does it happening?
Thank you.
honcharov-danylo
All 5 comments
I also encountered this problem, what is the reason?
ZouQunSheng
on 23 May 2019
@alanakbik Even I have encountered the same thing for my case of a very custom model. I had ignored it before assuming it to be a case of coincidence.
alipetiwala
on 23 May 2019
This is probably a bug in the way TNs are counted. We have a few open issues so will fix this hopefully soon!
alanakbik
on 23 May 2019
I wonder how one would count TNs in an entity-level evaluation. They are essentially all possible entities which were not tagged. So I would argue that they are simply not defined. As NER systems are usually evaluated according to precision, recall, F1, the true negatives do not play a role.
jantrienes
on 24 May 2019
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.
![stale[bot] picture](https://avatars.githubusercontent.com/in/1724?v=4&s=40)
stale[bot]
on 30 Apr 2020
Related issues
jannenev
路
3Comments
mnishant2
路
3Comments
gopalkalpande
路
3Comments
Aditya715
路
3Comments
Y4rd13
路
3Comments
Most helpful comment
I wonder how one would count TNs in an entity-level evaluation. They are essentially all possible entities which were not tagged. So I would argue that they are simply not defined. As NER systems are usually evaluated according to precision, recall, F1, the true negatives do not play a role.