mirror of
https://github.com/explosion/spaCy.git
synced 2024-12-30 20:06:30 +03:00
31 lines
683 B
Python
31 lines
683 B
Python
|
import pytest
|
||
|
from spacy.lang.sv.lex_attrs import like_num
|
||
|
|
||
|
|
||
|
@pytest.mark.parametrize(
|
||
|
"text,match",
|
||
|
[
|
||
|
("10", True),
|
||
|
("1", True),
|
||
|
("10.000", True),
|
||
|
("10.00", True),
|
||
|
("999,0", True),
|
||
|
("en", True),
|
||
|
("två", True),
|
||
|
("miljard", True),
|
||
|
("hund", False),
|
||
|
(",", False),
|
||
|
("1/2", True),
|
||
|
],
|
||
|
)
|
||
|
def test_lex_attrs_like_number(sv_tokenizer, text, match):
|
||
|
tokens = sv_tokenizer(text)
|
||
|
assert len(tokens) == 1
|
||
|
assert tokens[0].like_num == match
|
||
|
|
||
|
|
||
|
@pytest.mark.parametrize("word", ["elva"])
|
||
|
def test_sv_lex_attrs_capitals(word):
|
||
|
assert like_num(word)
|
||
|
assert like_num(word.upper())
|