System (please complete the following information):
OS: Mac os
Python version: 3.6
AllenNLP version: latest
PyTorch version: 1.2
Question
I want to compute span-f1 with 'BIOUL' label encoding, and I find that the function 'bioul_tags_to_spans' does not allow ill-formed spans. But the prediction output of the model
may be ill-formed, and will raise 'InvalidTagSequence' Error, so that the span-f1 can not compute.
doesn't handle the case of ill formed tag sequences. If you wanted to contribute a fix to allow it to handle ill formed sequences, that would be great!
System (please complete the following information):
Question