Hi,
I'm evaluating Flair and so far it's very cool!
I have a dataset which is already annotated as BILOU. I can technically translate it to BIO but was wondering if it should or shouldn't work off the shelf with BILOU tags.
Thanks!
You can replace BIEOS which is managed by Flair
annotations = [a.replace('L-', 'E-') for a in annotations]
annotations = [a.replace('U-', 'S-') for a in annotations]
(for Spacy annotations)
Thanks.
If I train a SequenceTagger with pretrained embeddings, does it matter which annotation scheme I use? Does the model use any external data to the training data I provide it with?
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.
Hi @pommedeterresautee
I have confirmed that BIEOS works.
In the other hand, would it be a problem if someone uses BIO scheme?
No problem with other schemes, but some may achieve better results.
Thanks @djstrong
I have just confirmed that using BIO works as well.
I hope this info could be useful as reference for future user.
Here is the summary to train NER with BILOU tags using Flair:
annotations = [a.replace('L-', 'E-') for a in annotations]
annotations = [a.replace('U-', 'S-') for a in annotations]
annotations = [a.replace('L-', 'I-') for a in annotations]
annotations = [a.replace('U-', 'B-') for a in annotations]
We can close the thread now.
Most helpful comment
Thanks @djstrong
I have just confirmed that using BIO works as well.
I hope this info could be useful as reference for future user.
Here is the summary to train NER with BILOU tags using Flair:
We can close the thread now.