spaCy/tests/tokenizer/test_surround_punct.py

33 lines
987 B
Python
Raw Normal View History

from __future__ import unicode_literals
import pytest
@pytest.fixture
def paired_puncts():
return [('(', ')'), ('[', ']'), ('{', '}'), ('*', '*')]
def test_token(paired_puncts, en_tokenizer):
word_str = 'Hello'
for open_, close_ in paired_puncts:
string = open_ + word_str + close_
tokens = en_tokenizer(string)
assert len(tokens) == 3
assert tokens[0].orth_ == open_
assert tokens[1].orth_ == word_str
assert tokens[2].orth_ == close_
def test_two_different(paired_puncts, en_tokenizer):
word_str = 'Hello'
for open_, close_ in paired_puncts:
string = "`" + open_ + word_str + close_ + "'"
tokens = en_tokenizer(string)
assert len(tokens) == 5
assert tokens[0].orth_ == "`"
assert tokens[1].orth_ == open_
assert tokens[2].orth_ == word_str
assert tokens[2].orth_ == word_str
assert tokens[3].orth_ == close_
assert tokens[4].orth_ == "'"