I try to train a custom POS tagger with Spacy, but when I run training:
python -m spacy train config.cfg --output ./output
I get the following training metrics:
E # LOSS TOK2VEC LOSS TAGGER TAG_ACC SCORE
--- ------ ------------ ----------- ------- ------
0 0 0.00 56.47 34.57 0.35
0 200 220.60 9159.67 75.71 0.76
0 400 416.10 8008.25 80.36 0.80
0 600 574.27 8974.13 82.15 0.82
1 800 683.66 8606.33 82.96 0.83
However, in different tutorials and examples, I can see additional metrics that would be nice to see as well:
E # LOSS TOK2VEC LOSS NER ENTS_F ENTS_P ENTS_R SCORE
(True, these are mostly indicated in case of NER training, not POS training examples, but I cannot see any reason why not to indicate them in case of POS as well).
I suspect that it could be customized by the following line in config.cfg:
scorer = {"@scorers":"spacy.tagger_scorer.v1"}
but I'm not sure how to actually edit this to get the desired output.
By the way, what is score metric in this context, and why does loss tok2vec keep increasing during the training (while accuracy improves)?