scikit-prodigy
Helpers to leverage scikit-learn pipelines in Prodigy.
Recipes
textcat.sklearn.binary
This recipe assumes binary text classification done via scikit-learn.
You're able to annotate as you would normally, but you can also set the
--correct flag which will train a scikit-learn model just before annotation.
You can then annotate more positive, negative or uncertain examples based
on the --prefer setting in the recipe.
The default usage, which you should use to start with is:
python -m prodigy textcat.sklearn sklearn-demo examples.jsonl --label insult -F recipes/binary_textcat.py
Then, once we have positive/negative examples that sklearn could train on, you can use it for model-in-the-loop annotation.
python -m prodigy textcat.sklearn sklearn-demo examples.jsonl --label insult --correct --prefer uncertain -F recipes/binary_textcat.py
Feel free to take this recipe as a starting point to customise further!