Hello,
Could you please advice me what would be the wise way to print out the labels's accuracy while doing ner.batch-train alongside with the total accuracy?
Many thanks,
This isn't built into ner.batch-train
at the moment, but the latest version of spaCy supports an ents_by_type
property in the Scorer
. See here for details: https://spacy.io/api/scorer/
So you could write a separate evaluation script that takes the pre-trained model and the evaluation data and outputs the scores, including the scores per label.
Thanks Ines, I'll try this.