Documenation for using a trained model for inference?

[jpmorris]

I've looked high and low and I don't see any example for how to actually use a trained model to score on unseened data--say in a production pipeline with new data coming in at a regular cadence. I did find a stackoverflow post which said I need to load the train model and create an nlp object. I assume since the nlp object is just a pipline that uses existing trained models to score examples, I can just use nlp.pipe to run over a DocBin of new data? But is there really not a single example of how to do this in any of the documentation? Can I use the CLI for this or do I have to make a script to do it?

[honnibal]

Does this command do what you need? https://spacy.io/api/cli#apply

[jpmorris]

That looks good to me! Thanks for the response and quick reply!