fb7acacd42
I have been using the main example of the spaCy web page for testing updates of spacy (and its transitive dependencies). Let's convert this into a proper test to take out manual testing.
70 lines
1.7 KiB
Python
70 lines
1.7 KiB
Python
import pytest
|
||
import spacy
|
||
|
||
en_text = (
|
||
"When Sebastian Thrun started working on self-driving cars at "
|
||
"Google in 2007, few people outside of the company took him "
|
||
"seriously. “I can tell you very senior CEOs of major American "
|
||
"car companies would shake my hand and turn away because I wasn’t "
|
||
"worth talking to,” said Thrun, in an interview with Recode earlier "
|
||
"this week.")
|
||
|
||
|
||
@pytest.fixture
|
||
def en_core_web_sm():
|
||
return spacy.load("en_core_web_sm")
|
||
|
||
|
||
@pytest.fixture
|
||
def doc_en_core_web_sm(en_core_web_sm):
|
||
return en_core_web_sm(en_text)
|
||
|
||
|
||
def test_entities(doc_en_core_web_sm):
|
||
entities = list(map(lambda e: (e.text, e.label_),
|
||
doc_en_core_web_sm.ents))
|
||
|
||
assert entities == [
|
||
('Sebastian Thrun', 'PERSON'),
|
||
('Google', 'ORG'), ('2007', 'DATE'),
|
||
('American', 'NORP'),
|
||
('Thrun', 'ORG'),
|
||
('earlier this week', 'DATE')
|
||
]
|
||
|
||
|
||
def test_nouns(doc_en_core_web_sm):
|
||
assert [
|
||
chunk.text for chunk in doc_en_core_web_sm.noun_chunks] == [
|
||
'Sebastian Thrun',
|
||
'self-driving cars',
|
||
'Google',
|
||
'few people',
|
||
'the company',
|
||
'him',
|
||
'I',
|
||
'you',
|
||
'very senior CEOs',
|
||
'major American car companies',
|
||
'my hand',
|
||
'I',
|
||
'Thrun',
|
||
'an interview',
|
||
'Recode']
|
||
|
||
|
||
def test_verbs(doc_en_core_web_sm):
|
||
assert [
|
||
token.lemma_ for token in doc_en_core_web_sm if token.pos_ == "VERB"] == [
|
||
'start',
|
||
'work',
|
||
'drive',
|
||
'take',
|
||
'can',
|
||
'tell',
|
||
'would',
|
||
'shake',
|
||
'turn',
|
||
'talk',
|
||
'say']
|