Confirmation of NER Flowchart implementation

The NER Flowchart is an Engineering Masterpiece, which we treat as Ultimate Authority.

We know that we must train new models from scratch. Therefore the decision box "Can you write rules....?" is pivotal, since we expect to avoid ner.manual for production-level work. We infer that "[to]write rules .... (Matcher patterns)?...." can mean "using ner.teach to generate patterns." This is our current procedure, and we want to be certain that we're in line with the flowchart.

Prodigy + SpaCy + your voluminous documentation and patient care of your users makes for an outstanding environment for learning, training, and least-cost experimentation. Thank you (from a learner and trainer)!

1 Like

Hi Ron,

Thanks for the kind words! Not 100% sure I'm answering the whole of your question, but a quick clarification:

since we expect to avoid ner.manual for production-level work.

There's really nothing wrong with ner.manual, and in the new docs and website we've been working on (which if you love the current docs, I'm sure you'll be extra impressed by), we've tried to clarify that it's often very useful to do the initial annotations without a model or patterns. The ner.manual recipe is also great for annotating evaluation data.

So, I would say, just get started with ner.manual, and see how you go. You can always solve the problem of "this process isn't as fast as we'd like" later, when you have data to train models to help you, and you know which common cases are best addressed by pattern rules.