mirror of https://github.com/explosion/spaCy.git
* Add test for indices
This commit is contained in:
parent
03cc2ee08e
commit
613a195f92
|
@ -0,0 +1,39 @@
|
||||||
|
"""Test that token.idx correctly computes index into the original string."""
|
||||||
|
|
||||||
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
|
import pytest
|
||||||
|
from spacy.en import English
|
||||||
|
|
||||||
|
|
||||||
|
@pytest.fixture
|
||||||
|
def nlp():
|
||||||
|
nlp = English()
|
||||||
|
return nlp.tokenizer
|
||||||
|
|
||||||
|
|
||||||
|
def test_simple_punct(nlp):
|
||||||
|
text = 'to walk, do foo'
|
||||||
|
tokens = nlp(text)
|
||||||
|
assert tokens[0].idx == 0
|
||||||
|
assert tokens[1].idx == 3
|
||||||
|
assert tokens[2].idx == 7
|
||||||
|
assert tokens[3].idx == 9
|
||||||
|
assert tokens[4].idx == 12
|
||||||
|
|
||||||
|
|
||||||
|
def test_complex_punct(nlp):
|
||||||
|
text = 'Tom (D., Ill.)!'
|
||||||
|
tokens = nlp(text)
|
||||||
|
assert tokens[0].idx == 0
|
||||||
|
assert len(tokens[0]) == 3
|
||||||
|
assert tokens[1].idx == 4
|
||||||
|
assert len(tokens[1]) == 1
|
||||||
|
assert tokens[2].idx == 5
|
||||||
|
assert len(tokens[2]) == 2
|
||||||
|
assert tokens[3].idx == 7
|
||||||
|
assert len(tokens[3]) == 1
|
||||||
|
assert tokens[4].idx == 9
|
||||||
|
assert len(tokens[4]) == 4
|
||||||
|
assert tokens[5].idx == 13
|
||||||
|
assert tokens[6].idx == 14
|
Loading…
Reference in New Issue