https://www.machinelearningplus.com/nlp/custom-text-classification-spacy/
https://v2.spacy.io/usage/training
https://towardsdatascience.com/sarcasm-text-classification-using-spacy-in-python-7cd39074f32e
https://v2.spacy.io/api/textcategorizer
Evaluate:
https://github.com/explosion/spaCy/blob/v2.3.5/spacy/language.py
Issue:
https://github.com/explosion/spaCy/issues/5322
https://course.spacy.io/en/chapter4
https://explosion.ai/blog/pseudo-rehearsal-catastrophic-forgetting
IMPORTANT:
Labels allow only letters. Categorization will not work if label contains Symbols (Numbers are not tested).
ITERATION DATA: i get excellent results with 500 items per category per training. With more than 500 item per training, you get bad score near 50% even after 500 trainings it stay on 50%. So don't use a lot of item per training!