Config error with nlp.add_pipe

[alejandrods]

I am refactoring my code from Spacy v2 to v3 and I am facing some problems. I got the following error using the below code:

E               thinc.config.ConfigValidationError: 
E               
E               Config validation error
E               
E               phrase_matcher.phrase_patterns -> self   field required
E               
E               {'@misc': 'phrase_matcher.patterns.v1'}

Code:

class PhraseMatcherComponent:
...

@Language.factory("phrase_matcher", default_config={"phrase_patterns": []})
def create_phrase_component(nlp: Language,
                            name: str,
                            phrase_patterns):
    return PhraseMatcherComponent(nlp, phrase_patterns)

class Metabase:
  def __init__(subscriptions):
    self.subscriptions = subscriptions
    self._init_matchers()

  @spacy.registry.misc("phrase_matcher.patterns.v1")
  def _init_phrase_matcher(self):
      phrase_patterns = []
      for query, case_sensitive in self.subscriptions.queries:
          tokens = self.language.make_doc(query)
          if not query.endswith("*"): 
              phrase_patterns.append(
                  (str(query), None, tokens, case_sensitive))
      return phrase_patterns

  def _init_matchers(self):
    self.language = English()

    self.language.add_pipe("phrase_matcher", config={"phrase_patterns": {"@misc": "phrase_matcher.patterns.v1"}}, last=True)

How should I then define the function _init_phrase_matcher()?

Thank you.

[polm]

A function in a registry should not be a method on an instance of an object. As you have the decorator now, the function you have for patterns has to be called on an instance of Metabase, but there's no place to get the Metabase object from. The error means that you're creating the function but you're not passing the required argument of self, but there's also no way to specify that in the config file.

Things that go in the config need to be JSON serializable. Assuming Metabase is not JSON serializable, the easiest thing here is probably to have a function on the PhraseMatcherComponent that allows you to set the Metabase after the component has been initialized, such as set_metabase(self, metabase).

Also, you might want to look at the EntityRuler component, which is basically a component that wraps the PhraseMatcher and normal Matcher. Even if you're creating your own component because the EntityRuler doesn't work for you for some reason, the way the EntityRuler handles patterns - using methods that let you set them outside the config - is worth looking at.