spacy example does not work as advertised

[reckart]

When putting the spacy example shown in the README file into a py file and running it, it produces an error:

OSError: [E941] Can't find model 'en'. It looks like you're trying to load a model from a shortcut, which is deprecated as of spaCy v3.0. To load the model, use its full name instead:

nlp = spacy.load("en_core_web_sm")

[DavidHuebner]

Sorry, to intervene here, but I believe that this issue needs another little change. Right now, when I execute the SpacyPosClassifier pipelines, then all pos tags are empty strings. The reason are twofold:

1. The tagger pipelines needs the existence of word embeddings. Hence, one needs to manually call self._model.get_pipe("tok2vec")(doc) before calling self._model.get_pipe("tagger")(doc)
2. The information about the PoS tags in Spacy v3 is now saved in spacy_token.tag_ instead of spacy_token.pos_.

Please also see my discussion here: https://github.com/explosion/spaCy/issues/7105#

BTW: Unfortunately, the spacy test is not capturing the problem because all PoS-values are empty strings, i.e. not None and hence, the following assert is not triggered.

for prediction in predictions:
    assert getattr(prediction, PREDICTED_FEATURE) is not None

[jcklie]

Thank you for the report. I changed the recommender and the test, I hope it works now.