Summary

Key points

• We explored how to customize spaCy statistical models according to our own domain and data.

• We learned the key points of deciding whether we really need custom model training.

• We went through an essential part of statistical algorithm design—data collection and labeling.

• We also learned about two annotation tools—Prodigy and Brat.

• We started model training by updating spaCy's NER component with our navigation domain data samples.

• We learned the necessary model training steps, including disabling the other pipeline components, creating example objects to hold our examples, and feeding our examples to the training code.

• We learned how to train an NER model from scratch on a small toy dataset and on a real medical domain dataset.