Bugfix/fix entity ruler from disk

[GuiGel]

Solve a bug in the EntityRuler from_disk method

Description

issue4651

The PhraseMatcher isn't initialized correctly in the EntityRuler from_disk method when the phrase_matcher_attr attribute is specified.
This is because the from_disk function is used before the recovery of the phrase_matcher_attr attribute.
A way to solve this bug is to split the deserializer in a 2 keys dict, the patterns and the cfg. By this way we can first deserialize the cfg and initialize the PhraseMatcher with it, and secondly use the patterns part in the from_disk function, in order to add the patterns the PhraseMatcher that have been correctly initialized.

My environnement:

spaCy version: 2.2.2
Platform: Linux-4.9.0-11-amd64-x86_64-with-debian-9.11
Python version: 3.7.3

Tests run

test_issue4651_with_phrase_matcher_attr --> test that failed without modification of the EntityRuler from_disk method and work after.
test_issue4651_without_phrase_matcher_attr --> test that work all the time.

Types of change

bug fix

Checklist

• I have submitted the spaCy Contributor Agreement.
• I ran the tests, and all new and existing tests passed.
• My changes don't require a change to the documentation, or if they do, I've added all required information.

[adrianeboyd]

Hi, thanks for submitting this PR! Your changes and tests look good, the main problems look like a few quirky things related to temporary files in the CI. (I ran into the same problems in my first PR with tests involving temporary files, it's not obvious!)

Try serializing the EntityRuler by itself (just to minimize the functionality that you're testing) and handling the temporary directories for to_disk() like this:

from ..util import make_tempdir

...
    nlp_reloaded = English()
    with make_tempdir() as d:
        file_path = d / "entityruler"
        ruler.to_disk(file_path)
        ruler_reloaded = EntityRuler(nlp_reloaded).from_disk(file_path)

    nlp_reloaded.add_pipe(ruler_reloaded)
    loaded_doc = nlp_reloaded(text)

You can see some examples in other tests like this one:

spaCy/spacy/tests/serialize/test_serialize_pipeline.py

Lines 48 to 60 in 5ad5c4b

	@pytest.mark.parametrize("Parser", test_parsers)
	def test_serialize_parser_roundtrip_disk(en_vocab, Parser):
	parser = Parser(en_vocab)
	parser.model, _ = parser.Model(0)
	with make_tempdir() as d:
	file_path = d / "parser"
	parser.to_disk(file_path)
	parser_d = Parser(en_vocab)
	parser_d.model, _ = parser_d.Model(0)
	parser_d = parser_d.from_disk(file_path)
	parser_bytes = parser.to_bytes(exclude=["model", "vocab"])
	parser_d_bytes = parser_d.to_bytes(exclude=["model", "vocab"])
	assert parser_bytes == parser_d_bytes

[ines]

Thanks a lot! 👍 Also really appreciate the extensive tests.

[GuiGel]

Happy to contribute a little to your great library and thanks for you help!