scibert throws InvalidTagSequence(tag_sequence) for BIOUL
Opened this issue · 1 comments
kamalravi commented
I have a used custom dataset where I have encoded entities in BIOUL format. While training on this encoded dataset, I am getting InvalidTagSequence(tag_sequence) error.
Later I changed encoding to IOB1 and the training was smooth, no InvalidTagSequence(tag_sequence) error.
Q1.
While predicting on a new sentence, I get the entities coded in BIOUL format. Is this happening because IOB1 being mapped to BIOUL as mentioned here #50 (comment) ?
Q2.
If it is, since BIOUL is also a valid format. Why scibert throws such error?
ibeltagy commented
This is an AllenNLP issue. Can you share the error stack trace?