add spacy.Language as valid argument for 'spacy_pipeline'

[dominik-schwabe]

This commit allows to reuse an object from spacy.load for many different KeyphraseVectorizer objects. I noticed that the nlp objects gets loaded when fit is called, which makes extracting keyphrases from multiple documents super slow when a model link en_core_web_md is used.

[TimSchopf]

Hi Dominik,

thanks for the contribution. Can you also please add a short code example and explanation on how to use the new argument in the README.md file?

Also, you can extract keyphrases from multiple documents with the same object and calling fit only once by using a list of documents as inputs. This probably solves the issue already.

Best,
Tim

[dominik-schwabe]

I added a little example to the README.

Also, you can extract keyphrases from multiple documents with the same object and calling fit only once by using a list of documents as inputs. This probably solves the issue already.

Usually I make small experiments, where I inspect the results on one documents then change some things, try the changed document or try some different document. I also usually use en_core_web_md. In that setup loading a new nlp object for every new small experiment gives a delay of about 5s instead of being instantaneous.