Mark lemmatizer tests as models since they use installed data

This commit is contained in:
Ines Montani 2017-01-13 14:30:15 +01:00
parent 138deb80a1
commit 49186b34a1

View File

@ -6,6 +6,7 @@ from ...lemmatizer import read_index, read_exc
import pytest
@pytest.mark.models
@pytest.mark.parametrize('text,lemmas', [("aardwolves", ["aardwolf"]),
("aardwolf", ["aardwolf"]),
("planets", ["planet"]),
@ -17,6 +18,7 @@ def test_tagger_lemmatizer_noun_lemmas(lemmatizer, text, lemmas):
assert lemmatizer.noun(text) == set(lemmas)
@pytest.mark.models
def test_tagger_lemmatizer_base_forms(lemmatizer):
if lemmatizer is None:
return None
@ -24,12 +26,14 @@ def test_tagger_lemmatizer_base_forms(lemmatizer):
assert lemmatizer.noun('dive', {'number': 'plur'}) == set(['diva'])
@pytest.mark.models
def test_tagger_lemmatizer_base_form_verb(lemmatizer):
if lemmatizer is None:
return None
assert lemmatizer.verb('saw', {'verbform': 'past'}) == set(['see'])
@pytest.mark.models
def test_tagger_lemmatizer_punct(lemmatizer):
if lemmatizer is None:
return None
@ -37,6 +41,7 @@ def test_tagger_lemmatizer_punct(lemmatizer):
assert lemmatizer.punct('') == set(['"'])
@pytest.mark.models
def test_tagger_lemmatizer_read_index(path):
if path is not None:
with (path / 'wordnet' / 'index.noun').open() as file_:
@ -46,6 +51,7 @@ def test_tagger_lemmatizer_read_index(path):
assert 'plant' in index
@pytest.mark.models
@pytest.mark.parametrize('text,lemma', [("was", "be")])
def test_tagger_lemmatizer_read_exc(path, text, lemma):
if path is not None: