I am doubtful about the working of the recipe ner.correct
.
Do I have to retrain the model using train
or ner.correct
internally trains the model?
Yes, you'll always need to train the model. ner.correct
(and other recipes) may use a model to show you suggestions and help you annotate more efficiently – but they won't just silently overwrite your model. You'll always want to update your model separately and in batches and by making multiple passes over the data (e.g. using the train
recipe). You can also read more about training after annotating with a model in the loop here.
1 Like