SpaCy 3 Cross validation

Question

I was guessing whether there are some build-in options using SpaCy 3 in combination with k-folds cross-validation.

I would like to run a 10 folds cross validation on a dataset of approx. 17.5k job descriptions to receive the best performing model in the end.

Documentation isn't telling anything about these options.

I think I can do some workaround by hand if the option does not exist.

Thanks in advance!

Answer 1

spaCy does not have any built-in handling for cross validation, no.

With v3 using the config training, you'd probably want to generate train/dev split .spacy files before several training runs to do cross validation.