Hi,
I just started using prodigy and was watching some of the videos/reading (e.g. this one TRAINING AN INSULTS CLASSIFIER with Prodigy in ~1 hour - YouTube) the documentation and saw that running the "train" command generates a number of stats about the model, (e.g. recall, precision, f-stat, accuracy, #correct, #incorrect, etc.)
However, when I run:
$prodigy train ./model --textcat annotations
It only gives stats on loss, CATS_SCORE and SCORE.
I feel like I am missing something very obvious. Is there an easy way to generate these stats?
Thanks,
Aurelius
Hi @aureliusnoble !
The tutorial was created using an older version of Prodigy. that's why the console output differs. To see the per-label result, you can pass the --label-stats parameter in your train command:
prodigy train ./model --textcat annotations --label-stats
Note that you can also pass a --base-model for tokenization and sentence segmentation.
The reason why it didn't show up in the previous command is because the reported textcat scores depend upon the text classification architecture used. In addition, the training scheme for Prodigy v1.11+ is now identical to the v3 spacy train—the outputs are now similar for consistency.
1 Like