def setUp(self):
r"""Setup both cased and uncased tokenizer instances."""
self.cased_tokenizer = CharDictTokenizer()
self.cased_tokenizer.build_vocab(self.__class__.vocab_source)
self.uncased_tokenizer = CharDictTokenizer(is_uncased=True)
self.uncased_tokenizer.build_vocab(self.__class__.vocab_source)
self.tokenizers = [self.cased_tokenizer, self.uncased_tokenizer]
def test_experiment_does_not_exist(self):
r"""Raise `FileNotFoundError` when `experiment` does not exist."""
msg1 = (
'Must raise `FileNotFoundError` when `experiment` does not exist.')
msg2 = 'Inconsistent error message.'
examples = (self.__class__.experiment, 'I-AM-A-TEST-AND-I-DONT-EXIST')
for experiment in examples:
with self.assertRaises(FileNotFoundError, msg=msg1) as ctx_man:
CharDictTokenizer.load(experiment=experiment)
test_path = os.path.join(DATA_PATH, experiment, 'tokenizer.json')
self.assertEqual(ctx_man.exception.args[0],
f'File {test_path} does not exist.',
msg=msg2)
def test_load_result(self):
r"""Load `tokenizer.json`."""
msg = 'Inconsistent `tokenizer.json` format.'
examples = (
{
'is_uncased': False,
'token_to_id': {
'A': 0,
'B': 1,
'C': 2,
},
'id_to_token': {
0: 'A',
1: 'B',
2: 'C',
},
},
{
'is_uncased': True,
'token_to_id': {
'a': 0,
'b': 1,
'c': 2,
},
'id_to_token': {
0: 'a',
1: 'b',
2: 'c',
},
},
)
test_path = os.path.join(self.__class__.test_dir, 'tokenizer.json')
for obj in examples:
try:
# Create test file.
with open(test_path, 'w', encoding='utf-8') as output_file:
tmp = {
'is_uncased': obj['is_uncased'],
'token_to_id': obj['token_to_id']
}
json.dump(tmp, output_file)
tokenizer = CharDictTokenizer.load(
experiment=self.__class__.experiment)
self.assertIsInstance(tokenizer, CharDictTokenizer, msg=msg)
for attr_key, attr_value in obj.items():
self.assertTrue(hasattr(tokenizer, attr_key), msg=msg)
self.assertIsInstance(getattr(tokenizer, attr_key),
type(attr_value),
msg=msg)
self.assertEqual(getattr(tokenizer, attr_key),
attr_value,
msg=msg)
finally:
# Clean up test file.
os.remove(test_path)
def test_invalid_input_experiment(self):
r"""Raise exception when input `experiment` is invalid."""
msg1 = (
'Must raise `TypeError` or `ValueError` when input `experiment` '
'is invalid.')
msg2 = 'Inconsistent error message.'
examples = (
False,
True,
0,
1,
-1,
0.0,
1.0,
math.nan,
-math.nan,
math.inf,
-math.inf,
0j,
1j,
'',
b'',
(),
[],
{},
set(),
object(),
lambda x: x,
type,
None,
NotImplemented,
...,
)
for invalid_input in examples:
with self.assertRaises((TypeError, ValueError),
msg=msg1) as ctx_man:
CharDictTokenizer.load(experiment=invalid_input)
if isinstance(ctx_man.exception, TypeError):
self.assertEqual(ctx_man.exception.args[0],
'`experiment` must be an instance of `str`.',
msg=msg2)
else:
self.assertEqual(ctx_man.exception.args[0],
'`experiment` must not be empty.',
msg=msg2)
def test_invalid_input_max_seq_len(self):
r"""Raise exception when input `max_seq_len` is invalid."""
msg1 = (
'Must raise `TypeError` or `ValueError` when input `max_seq_len` '
'is invalid.')
msg2 = 'Inconsistent error message.'
examples = (
False,
True,
0,
1,
-2,
0.0,
1.0,
math.nan,
-math.nan,
math.inf,
-math.inf,
0j,
1j,
'',
b'',
(),
[],
{},
set(),
object(),
lambda x: x,
type,
None,
NotImplemented,
...,
)
for invalid_input in examples:
with self.assertRaises((TypeError, ValueError),
msg=msg1) as cxt_man:
BaseDataset([]).create_collate_fn(
tokenizer=CharDictTokenizer(), max_seq_len=invalid_input)
if isinstance(cxt_man.exception, TypeError):
self.assertEqual(cxt_man.exception.args[0],
'`max_seq_len` must be an instance of `int`.',
msg=msg2)
else:
self.assertEqual(
cxt_man.exception.args[0],
'`max_seq_len` must be greater than `1` or equal to '
'`-1`.',
msg=msg2)
def test_reset_vocab_size(self):
r"""Reset vocabulary size after `reset_vocab`."""
msg = 'Must reset vocabulary size after `reset_vocab`.'
examples = (
('HeLlO WoRlD!', 'I aM a LeGeNd.'),
('y = f(x)',),
('',),
)
sp_tokens_size = len(list(CharDictTokenizer.special_tokens()))
for batch_sequences in examples:
for tokenizer in self.tokenizers:
tokenizer.build_vocab(batch_sequences)
tokenizer.reset_vocab()
self.assertEqual(
tokenizer.vocab_size,
sp_tokens_size,
msg=msg
)
def test_invalid_input_is_uncased(self):
r"""Raise `TypeError` when input `is_uncased` is invalid."""
msg1 = 'Must raise `TypeError` when input `is_uncased` is invalid.'
msg2 = 'Inconsistent error message.'
examples = (
0,
1,
-1,
0.0,
1.0,
math.nan,
-math.nan,
math.inf,
-math.inf,
0j,
1j,
'',
b'',
(),
[],
{},
set(),
object(),
lambda x: x,
type,
None,
NotImplemented,
...,
)
for invalid_input in examples:
with self.assertRaises(TypeError, msg=msg1) as ctx_man:
CharDictTokenizer(is_uncased=invalid_input)
self.assertEqual(ctx_man.exception.args[0],
'`is_uncased` must be an instance of `bool`.',
msg=msg2)
def test_increase_vocab_size(self):
r"""Increase vocabulary size after `build_vocab`."""
msg = 'Must increase vocabulary size after `build_vocab`.'
examples = (
(('HeLlO WoRlD!', 'I aM a LeGeNd.'), 18, 15),
(('y = f(x)',), 24, 21),
(('',), 24, 21),
)
sp_tokens_size = len(list(CharDictTokenizer.special_tokens()))
for batch_sequences, cased_vocab_size, uncased_vocab_size in examples:
self.cased_tokenizer.build_vocab(batch_sequences)
self.assertEqual(
self.cased_tokenizer.vocab_size,
cased_vocab_size + sp_tokens_size,
msg=msg
)
self.uncased_tokenizer.build_vocab(batch_sequences)
self.assertEqual(
self.uncased_tokenizer.vocab_size,
uncased_vocab_size + sp_tokens_size,
msg=msg
)
class TestDecode(unittest.TestCase):
r"""Test case for `lmp.tokenizer.CharDictTokenizer.decode`."""
@classmethod
def setUpClass(cls):
cls.vocab_source = [
'Hello World!',
'I am a legend.',
]
@classmethod
def tearDownClass(cls):
del cls.vocab_source
gc.collect()
def setUp(self):
r"""Setup both cased and uncased tokenizer instances."""
self.cased_tokenizer = CharDictTokenizer()
self.cased_tokenizer.build_vocab(self.__class__.vocab_source)
self.uncased_tokenizer = CharDictTokenizer(is_uncased=True)
self.uncased_tokenizer.build_vocab(self.__class__.vocab_source)
self.tokenizers = [self.cased_tokenizer, self.uncased_tokenizer]
def tearDown(self):
r"""Delete both cased and uncased tokenizer instances."""
del self.tokenizers
del self.cased_tokenizer
del self.uncased_tokenizer
gc.collect()
def test_signature(self):
r"""Ensure signature consistency."""
msg = 'Inconsistent method signature.'
self.assertEqual(
inspect.signature(CharDictTokenizer.decode),
inspect.Signature(parameters=[
inspect.Parameter(
name='self',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
),
inspect.Parameter(name='token_ids',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
annotation=Iterable[int],
default=inspect.Parameter.empty),
inspect.Parameter(name='remove_special_tokens',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
annotation=bool,
default=False)
],
return_annotation=str),
msg=msg)
def test_invalid_input_token_ids(self):
r"""Raise `TypeError` when input `token_ids` is invalid."""
msg1 = 'Must raise `TypeError` when input `token_ids` is invalid.'
msg2 = 'Inconsistent error message.'
examples = (
False,
True,
0,
1,
-1,
0.0,
1.0,
math.nan,
-math.nan,
math.inf,
-math.inf,
0j,
1j,
object(),
lambda x: x,
type,
None,
NotImplemented,
...,
[0.0],
[1.0],
[math.nan],
[-math.nan],
[math.inf],
[-math.inf],
[0j],
[1j],
[''],
[b''],
[()],
[[]],
[{}],
[set()],
[object()],
[lambda x: x],
[type],
[None],
[NotImplemented],
[...],
[0, 0.0],
[0, 1.0],
[0, math.nan],
[0, -math.nan],
[0, math.inf],
[0, -math.inf],
[0, 0j],
[0, 1j],
[0, ''],
[0, b''],
[0, ()],
[0, []],
[0, {}],
[0, set()],
[0, object()],
[0, lambda x: x],
[0, type],
[0, None],
[0, NotImplemented],
[0, ...],
)
for invalid_input in examples:
for tokenizer in self.tokenizers:
with self.assertRaises(TypeError, msg=msg1) as cxt_man:
tokenizer.decode(token_ids=invalid_input)
self.assertEqual(
cxt_man.exception.args[0],
'`token_ids` must be an instance of `Iterable[int]`.',
msg=msg2)
def test_invalid_input_remove_special_tokens(self):
r"""Raise `TypeError` when input `remove_special_tokens` is invalid."""
msg1 = ('Must raise `TypeError` when input `remove_special_tokens` is '
'invalid.')
msg2 = 'Inconsistent error message.'
examples = (
0,
1,
-1,
0.0,
1.0,
math.nan,
-math.nan,
math.inf,
-math.inf,
0j,
1j,
'',
b'',
(),
[],
{},
set(),
object(),
lambda x: x,
type,
None,
NotImplemented,
...,
)
for invalid_input in examples:
for tokenizer in self.tokenizers:
with self.assertRaises(TypeError, msg=msg1) as cxt_man:
tokenizer.decode(token_ids=[],
remove_special_tokens=invalid_input)
self.assertEqual(
cxt_man.exception.args[0],
'`remove_special_tokens` must be an instance of `bool`.',
msg=msg2)
def test_return_type(self):
r"""Return `str`."""
msg = 'Must return `str`.'
examples = (
[0, 1, 2, 3],
[4, 5, 6, 7, 8, 9],
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
[0],
[],
)
for token_ids in examples:
for tokenizer in self.tokenizers:
self.assertIsInstance(tokenizer.decode(token_ids=token_ids),
str,
msg=msg)
def test_remove_special_tokens(self):
r"""Remove special tokens."""
msg = 'Must remove special tokens.'
examples = (
(
False,
[0, 10, 6, 4, 4, 7, 5, 11, 7, 12, 4, 8, 13, 1, 2],
'[bos]Hello World![eos][pad]',
'[bos]hello world![eos][pad]',
),
(
False,
[0, 14, 5, 9, 15, 5, 9, 5, 3, 1, 2, 2],
'[bos]I am a [unk][eos][pad][pad]',
'[bos]i am a [unk][eos][pad][pad]',
),
(
False,
[0, 19, 4, 6, 16, 6, 17, 8, 18, 1],
'[bos][unk]legend.[eos]',
'[bos][unk]legend.[eos]',
),
(
True,
[0, 10, 6, 4, 4, 7, 5, 11, 7, 12, 4, 8, 13, 1, 2],
'Hello World!',
'hello world!',
),
(
True,
[0, 14, 5, 9, 15, 5, 9, 5, 3, 1, 2, 2],
'I am a [unk]',
'i am a [unk]',
),
(
True,
[0, 19, 4, 6, 16, 6, 17, 8, 18, 1],
'[unk]legend.',
'[unk]legend.',
),
)
for (remove_special_tokens, token_ids, cased_sequence,
uncased_sequence) in examples:
self.assertEqual(self.cased_tokenizer.decode(
token_ids=token_ids,
remove_special_tokens=remove_special_tokens),
cased_sequence,
msg=msg)
self.assertEqual(self.uncased_tokenizer.decode(
token_ids=token_ids,
remove_special_tokens=remove_special_tokens),
uncased_sequence,
msg=msg)
def setUp(self):
r"""Setup both cased and uncased tokenizer instances."""
self.cased_tokenizer = CharDictTokenizer()
self.uncased_tokenizer = CharDictTokenizer(is_uncased=True)
self.tokenizers = [self.cased_tokenizer, self.uncased_tokenizer]
class TestEncode(unittest.TestCase):
r"""Test case for `lmp.tokenizer.CharDictTokenizer.encode`."""
@classmethod
def setUpClass(cls):
cls.vocab_source = [
'Hello World!',
'I am a legend.',
]
@classmethod
def tearDownClass(cls):
del cls.vocab_source
gc.collect()
def setUp(self):
r"""Setup both cased and uncased tokenizer instances."""
self.cased_tokenizer = CharDictTokenizer()
self.cased_tokenizer.build_vocab(self.__class__.vocab_source)
self.uncased_tokenizer = CharDictTokenizer(is_uncased=True)
self.uncased_tokenizer.build_vocab(self.__class__.vocab_source)
self.tokenizers = [self.cased_tokenizer, self.uncased_tokenizer]
def tearDown(self):
r"""Delete both cased and uncased tokenizer instances."""
del self.tokenizers
del self.cased_tokenizer
del self.uncased_tokenizer
gc.collect()
def test_signature(self):
r"""Ensure signature consistency."""
msg = 'Inconsistent method signature.'
self.assertEqual(
inspect.signature(CharDictTokenizer.encode),
inspect.Signature(parameters=[
inspect.Parameter(
name='self',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
),
inspect.Parameter(name='sequence',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
annotation=str,
default=inspect.Parameter.empty),
inspect.Parameter(name='max_seq_len',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
annotation=int,
default=-1)
],
return_annotation=List[int]),
msg=msg)
def test_invalid_input_sequence(self):
r"""Raise `TypeError` when input `sequence` is invalid."""
msg1 = 'Must raise `TypeError` when input `sequence` is invalid.'
msg2 = 'Inconsistent error message.'
examples = (
False,
True,
0,
1,
-1,
0.0,
1.0,
math.nan,
-math.nan,
math.inf,
-math.inf,
0j,
1j,
b'',
(),
[],
{},
set(),
object(),
lambda x: x,
type,
None,
NotImplemented,
...,
)
for invalid_input in examples:
for tokenizer in self.tokenizers:
with self.assertRaises(TypeError, msg=msg1) as cxt_man:
tokenizer.encode(sequence=invalid_input)
self.assertEqual(cxt_man.exception.args[0],
'`sequence` must be an instance of `str`.',
msg=msg2)
def test_invalid_input_max_seq_len(self):
r"""Raise exception when input `max_seq_len` is invalid."""
msg1 = (
'Must raise `TypeError` or `ValueError` when input `max_seq_len` '
'is invalid.')
msg2 = 'Inconsistent error message.'
examples = (
False,
True,
0,
1,
-2,
0.0,
1.0,
math.nan,
-math.nan,
math.inf,
-math.inf,
0j,
1j,
'',
b'',
(),
[],
{},
set(),
object(),
lambda x: x,
type,
None,
NotImplemented,
...,
)
for invalid_input in examples:
for tokenizer in self.tokenizers:
with self.assertRaises((TypeError, ValueError),
msg=msg1) as cxt_man:
tokenizer.encode(sequence='', max_seq_len=invalid_input)
if isinstance(cxt_man.exception, TypeError):
self.assertEqual(
cxt_man.exception.args[0],
'`max_seq_len` must be an instance of `int`.',
msg=msg2)
else:
self.assertEqual(
cxt_man.exception.args[0],
'`max_seq_len` must be greater than `1` or equal to '
'`-1`.',
msg=msg2)
def test_return_type(self):
r"""Return `List[int]`."""
msg = 'Must return `List[int]`.'
examples = (
'Hello world!',
'I am a legend.',
'y = f(x)',
'',
)
for sequence in examples:
for tokenizer in self.tokenizers:
token_ids = tokenizer.encode(sequence=sequence)
self.assertIsInstance(token_ids, list, msg=msg)
for token_id in token_ids:
self.assertIsInstance(token_id, int, msg=msg)
def test_encode_format(self):
r"""Follow encode format."""
msg = 'Must follow encode format: [bos] t1 t2 ... tn [eos].'
examples = (
(
'Hello World!',
[0, 10, 6, 4, 4, 7, 5, 11, 7, 12, 4, 8, 13, 1],
),
(
'I am a legend.',
[0, 14, 5, 9, 15, 5, 9, 5, 4, 6, 16, 6, 17, 8, 18, 1],
),
(
'y = f(x)',
[0, 3, 5, 3, 5, 3, 3, 3, 3, 1],
),
(
'',
[0, 1],
),
)
for sequence, token_ids in examples:
for tokenizer in self.tokenizers:
self.assertEqual(tokenizer.encode(sequence=sequence),
token_ids,
msg=msg)
def test_truncate(self):
r"""Token ids' length must not exceed `max_seq_len`."""
msg = 'Token ids\' length must not exceed `max_seq_len`.'
examples = (
(
'Hello World!',
[0, 10, 6, 4, 4, 7, 5, 11, 7, 1],
10,
),
(
'I am a legend.',
[0, 14, 5, 9, 1],
5,
),
(
'y = f(x)',
[0, 3, 1],
3,
),
(
'',
[0, 1],
2,
),
)
for sequence, token_ids, max_seq_len in examples:
for tokenizer in self.tokenizers:
self.assertEqual(tokenizer.encode(sequence=sequence,
max_seq_len=max_seq_len),
token_ids,
msg=msg)
def test_padding(self):
r"""Token ids' length must pad to `max_seq_len`."""
msg = 'Token ids\' length must pad to `max_seq_len`.'
examples = (
(
'Hello World!',
[0, 10, 6, 4, 4, 7, 5, 11, 7, 12, 4, 8, 13, 1, 2],
15,
),
(
'I am a legend.',
[
0,
14,
5,
9,
15,
5,
9,
5,
4,
6,
16,
6,
17,
8,
18,
1,
2,
2,
2,
2,
],
20,
),
(
'y = f(x)',
[
0,
3,
5,
3,
5,
3,
3,
3,
3,
1,
2,
2,
2,
2,
2,
2,
2,
2,
2,
2,
],
20,
),
(
'',
[0, 1, 2, 2, 2, 2, 2, 2, 2, 2],
10,
),
)
for sequence, token_ids, max_seq_len in examples:
for tokenizer in self.tokenizers:
self.assertEqual(tokenizer.encode(sequence=sequence,
max_seq_len=max_seq_len),
token_ids,
msg=msg)
class TestTokenize(unittest.TestCase):
r"""Test case for `lmp.tokenizer.CharDictTokenizer.tokenize`."""
@classmethod
def setUpClass(cls):
cls.vocab_source = [
'Hello World!',
'I am a legend.',
]
@classmethod
def tearDownClass(cls):
del cls.vocab_source
gc.collect()
def setUp(self):
r"""Setup both cased and uncased tokenizer instances."""
self.cased_tokenizer = CharDictTokenizer()
self.cased_tokenizer.build_vocab(self.__class__.vocab_source)
self.uncased_tokenizer = CharDictTokenizer(is_uncased=True)
self.uncased_tokenizer.build_vocab(self.__class__.vocab_source)
self.tokenizers = [self.cased_tokenizer, self.uncased_tokenizer]
def tearDown(self):
r"""Delete both cased and uncased tokenizer instances."""
del self.tokenizers
del self.cased_tokenizer
del self.uncased_tokenizer
gc.collect()
def test_signature(self):
r"""Ensure signature consistency."""
msg = 'Inconsistent method signature.'
self.assertEqual(
inspect.signature(CharDictTokenizer.tokenize),
inspect.Signature(parameters=[
inspect.Parameter(name='self',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
default=inspect.Parameter.empty),
inspect.Parameter(name='sequence',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
annotation=str,
default=inspect.Parameter.empty)
],
return_annotation=List[str]),
msg=msg)
def test_invalid_input_sequence(self):
r"""Raise `TypeError` when input `sequence` is invalid."""
msg1 = 'Must raise `TypeError` when input `sequence` is invalid.'
msg2 = 'Inconsistent error message.'
examples = (
False,
True,
0,
1,
-1,
0.0,
1.0,
math.nan,
-math.nan,
math.inf,
-math.inf,
b'',
0j,
1j,
(),
[],
{},
set(),
object(),
lambda x: x,
type,
None,
NotImplemented,
...,
)
for invalid_input in examples:
for tokenizer in self.tokenizers:
with self.assertRaises(TypeError, msg=msg1) as ctx_man:
tokenizer.tokenize(invalid_input)
self.assertEqual(ctx_man.exception.args[0],
'`sequence` must be an instance of `str`.',
msg=msg2)
def test_return_type(self):
r"""Return `List[str]`."""
msg = 'Must return `List[str]`.'
examples = (
'Hello world!',
'H',
'',
)
for sequence in examples:
for tokenizer in self.tokenizers:
tokens = tokenizer.tokenize(sequence)
self.assertIsInstance(tokens, list, msg=msg)
for token in tokens:
self.assertIsInstance(token, str, msg=msg)
def test_normalize(self):
r"""Return sequence is normalized."""
msg = 'Return sequence must be normalized.'
examples = (
(
' HeLlO WoRlD!',
[
'H',
'e',
'L',
'l',
'O',
' ',
'W',
'o',
'R',
'l',
'D',
'!',
],
[
'h',
'e',
'l',
'l',
'o',
' ',
'w',
'o',
'r',
'l',
'd',
'!',
],
),
(
'HeLlO WoRlD! ',
[
'H',
'e',
'L',
'l',
'O',
' ',
'W',
'o',
'R',
'l',
'D',
'!',
],
[
'h',
'e',
'l',
'l',
'o',
' ',
'w',
'o',
'r',
'l',
'd',
'!',
],
),
(
' HeLlO WoRlD! ',
[
'H',
'e',
'L',
'l',
'O',
' ',
'W',
'o',
'R',
'l',
'D',
'!',
],
[
'h',
'e',
'l',
'l',
'o',
' ',
'w',
'o',
'r',
'l',
'd',
'!',
],
),
(
'０',
['0'],
['0'],
),
(
'é',
[unicodedata.normalize('NFKC', 'é')],
[unicodedata.normalize('NFKC', 'é')],
),
(
'０é',
[
unicodedata.normalize('NFKC', '0'),
unicodedata.normalize('NFKC', 'é'),
],
[
unicodedata.normalize('NFKC', '0'),
unicodedata.normalize('NFKC', 'é'),
],
),
(
'',
[],
[],
),
)
for sequence, cased_tokens, uncased_tokens in examples:
self.assertEqual(self.cased_tokenizer.tokenize(sequence),
cased_tokens,
msg=msg)
self.assertEqual(self.uncased_tokenizer.tokenize(sequence),
uncased_tokens,
msg=msg)
class TestVocabSize(unittest.TestCase):
r"""Test case for `lmp.tokenizer.CharDictTokenizer.vocab_size`."""
def setUp(self):
r"""Setup both cased and uncased tokenizer instances."""
self.cased_tokenizer = CharDictTokenizer()
self.uncased_tokenizer = CharDictTokenizer(is_uncased=True)
self.tokenizers = [self.cased_tokenizer, self.uncased_tokenizer]
def tearDown(self):
r"""Delete both cased and uncased tokenizer instances."""
del self.tokenizers
del self.cased_tokenizer
del self.uncased_tokenizer
gc.collect()
def test_signature(self):
r"""Ensure signature consistency."""
msg = 'Inconsistent property signature.'
self.assertTrue(
inspect.isdatadescriptor(CharDictTokenizer.vocab_size),
msg=msg
)
self.assertFalse(
inspect.isfunction(CharDictTokenizer.vocab_size),
msg=msg
)
self.assertFalse(
inspect.ismethod(CharDictTokenizer.vocab_size),
msg=msg
)
def test_return_type(self):
r"""Return `int`"""
msg = 'Must return `int`.'
for tokenizer in self.tokenizers:
self.assertIsInstance(tokenizer.vocab_size, int, msg=msg)
def test_return_value(self):
r"""Return vocabulary size."""
msg = 'Inconsistent vocabulary size.'
for tokenizer in self.tokenizers:
self.assertEqual(tokenizer.vocab_size, 4, msg=msg)
def test_increase_vocab_size(self):
r"""Increase vocabulary size after `build_vocab`."""
msg = 'Must increase vocabulary size after `build_vocab`.'
examples = (
(('HeLlO WoRlD!', 'I aM a LeGeNd.'), 18, 15),
(('y = f(x)',), 24, 21),
(('',), 24, 21),
)
sp_tokens_size = len(list(CharDictTokenizer.special_tokens()))
for batch_sequences, cased_vocab_size, uncased_vocab_size in examples:
self.cased_tokenizer.build_vocab(batch_sequences)
self.assertEqual(
self.cased_tokenizer.vocab_size,
cased_vocab_size + sp_tokens_size,
msg=msg
)
self.uncased_tokenizer.build_vocab(batch_sequences)
self.assertEqual(
self.uncased_tokenizer.vocab_size,
uncased_vocab_size + sp_tokens_size,
msg=msg
)
def test_reset_vocab_size(self):
r"""Reset vocabulary size after `reset_vocab`."""
msg = 'Must reset vocabulary size after `reset_vocab`.'
examples = (
('HeLlO WoRlD!', 'I aM a LeGeNd.'),
('y = f(x)',),
('',),
)
sp_tokens_size = len(list(CharDictTokenizer.special_tokens()))
for batch_sequences in examples:
for tokenizer in self.tokenizers:
tokenizer.build_vocab(batch_sequences)
tokenizer.reset_vocab()
self.assertEqual(
tokenizer.vocab_size,
sp_tokens_size,
msg=msg
)
class TestTokenize(unittest.TestCase):
r"""Test Case for `lmp.tokenizer.CharDictTokenizer.tokenize`."""
def setUp(self):
r"""Setup both cased and uncased tokenizer instances."""
self.cased_tokenizer = CharDictTokenizer()
self.uncased_tokenizer = CharDictTokenizer(is_uncased=True)
self.tokenizers = [self.cased_tokenizer, self.uncased_tokenizer]
def tearDown(self):
r"""Delete both cased and uncased tokenizer instances."""
del self.tokenizers
del self.cased_tokenizer
del self.uncased_tokenizer
gc.collect()
def test_signature(self):
r"""Ensure signature consistency."""
msg = 'Inconsistent method signature.'
self.assertEqual(
inspect.signature(CharDictTokenizer.tokenize),
inspect.Signature(parameters=[
inspect.Parameter(name='self',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
default=inspect.Parameter.empty),
inspect.Parameter(name='sequence',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
annotation=str,
default=inspect.Parameter.empty)
],
return_annotation=List[str]),
msg=msg)
def test_invalid_input(self):
r"""Raise `TypeError` when input is invalid."""
msg1 = 'Must raise `TypeError` when input is invalid.'
msg2 = 'Inconsistent error message.'
examples = (
0,
1,
-1,
0.0,
1.0,
math.nan,
math.inf,
True,
False,
b'',
[],
(),
{},
set(),
object(),
lambda x: x,
type,
None,
)
for invalid_input in examples:
for tokenizer in self.tokenizers:
with self.assertRaises(TypeError, msg=msg1) as ctx_man:
tokenizer.tokenize(invalid_input)
self.assertEqual(ctx_man.exception.args[0],
'`sequence` must be instance of `str`.',
msg=msg2)
def test_return_type(self):
r"""Return `List[str]`."""
msg = 'Must return `List[str]`.'
examples = (
'Hello world!',
'',
)
for sequence in examples:
for tokenizer in self.tokenizers:
tokens = tokenizer.tokenize(sequence)
self.assertIsInstance(tokens, list, msg=msg)
for token in tokens:
self.assertIsInstance(token, str, msg=msg)
def test_unicode_normalize(self):
r"""Return NFKC normalized characters."""
msg = 'Must return NFKC normalized characters.'
examples = (
('０', ['0']),
('é', ['é']),
('０é', ['0', 'é']),
)
for sequence, ans_tokens in examples:
for tokenizer in self.tokenizers:
out_tokens = tokenizer.tokenize(sequence)
self.assertEqual(out_tokens, ans_tokens, msg=msg)
for out_token in out_tokens:
self.assertEqual(len(out_token), 1, msg=msg)
def test_cased_sensitive(self):
r"""Return cased sensitive characters when `is_uncased=False`."""
msg = ('Return result must be case-sensitive when construct with '
'`is_uncased=False`.')
examples = (
('HeLlO WoRlD!',
['H', 'e', 'L', 'l', 'O', ' ', 'W', 'o', 'R', 'l', 'D', '!']),
('HELLO WORLD!',
['H', 'E', 'L', 'L', 'O', ' ', 'W', 'O', 'R', 'L', 'D', '!']),
('hello world!',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!']),
('H', ['H']),
('h', ['h']),
)
for sequence, ans_tokens in examples:
self.assertEqual(self.cased_tokenizer.tokenize(sequence),
ans_tokens,
msg=msg)
def test_cased_insensitive(self):
r"""Return cased insensitive characters when `is_uncased=True`."""
msg = ('Return result must be case-insensitive when construct with '
'`is_uncased=True`.')
examples = (
('HeLlO WoRlD!',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!']),
('HELLO WORLD!',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!']),
('hello world!',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!']),
('H', ['h']),
('h', ['h']),
)
for sequence, ans_tokens in examples:
self.assertEqual(self.uncased_tokenizer.tokenize(sequence),
ans_tokens,
msg=msg)
def test_whitespace_strip(self):
r"""Strip input sequence."""
msg = (
'Input sequence must strip both leading and trailing whitespace '
'characters.')
examples = (
(' hello world! ',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!']),
(' hello world!',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!']),
('hello world! ',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!']),
('\nhello world!\n',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!']),
(' ', []),
('', []),
)
for sequence, ans_tokens in examples:
for tokenizer in self.tokenizers:
self.assertEqual(tokenizer.tokenize(sequence),
ans_tokens,
msg=msg)
def test_whitespace_collapse(self):
r"""Collapse whitespace characters."""
msg = ('Input sequence must convert consecutive whitespace characters '
'into single whitespace character.')
examples = (
('hello world !',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', ' ',
'!']),
('hello world !',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', ' ',
'!']),
('hello world !',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', ' ',
'!']),
('hello world\n\n!',
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', ' ',
'!']),
)
for sequence, ans_tokens in examples:
for tokenizer in self.tokenizers:
self.assertEqual(tokenizer.tokenize(sequence),
ans_tokens,
msg=msg)
class TestDetokenize(unittest.TestCase):
r"""Test Case for `lmp.tokenizer.CharDictTokenizer.detokenize`."""
def setUp(self):
r"""Setup both cased and uncased tokenizer instances."""
self.cased_tokenizer = CharDictTokenizer()
self.uncased_tokenizer = CharDictTokenizer(is_uncased=True)
self.tokenizers = [self.cased_tokenizer, self.uncased_tokenizer]
def tearDown(self):
r"""Delete both cased and uncased tokenizer instances."""
del self.tokenizers
del self.cased_tokenizer
del self.uncased_tokenizer
gc.collect()
def test_signature(self):
r"""Ensure signature consistency."""
msg = 'Inconsistent method signature.'
self.assertEqual(
inspect.signature(CharDictTokenizer.detokenize),
inspect.Signature(parameters=[
inspect.Parameter(name='self',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
default=inspect.Parameter.empty),
inspect.Parameter(name='tokens',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
annotation=Iterable[str],
default=inspect.Parameter.empty)
],
return_annotation=str),
msg=msg)
def test_invalid_input(self):
r"""Raise `TypeError` when input is invalid."""
msg1 = 'Must raise `TypeError` when input is invalid.'
msg2 = 'Inconsistent error message.'
examples = (
0,
1,
-1,
0.0,
1.0,
math.nan,
math.inf,
True,
False,
(1, 2, 3),
[1, 2, 3],
{1, 2, 3},
None,
)
for invalid_input in examples:
for tokenizer in self.tokenizers:
with self.assertRaises(TypeError, msg=msg1) as ctx_man:
tokenizer.detokenize(invalid_input)
self.assertEqual(
ctx_man.exception.args[0],
'`tokens` must be instance of `Iterable[str]`.',
msg=msg2)
def test_expected_return(self):
r"""Return expected strings."""
msg = 'Inconsistent detokenization result.'
examples = (([
'H', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!'
], 'Hello world!'), ([], ''))
for tokens, ans_sequence in examples:
for tokenizer in self.tokenizers:
out_sequence = tokenizer.detokenize(tokens)
self.assertIsInstance(out_sequence, str, msg=msg)
self.assertEqual(out_sequence, ans_sequence, msg=msg)
def test_case_insensitive(self):
r"""Detokenize does not consider cases."""
msg = 'Inconsistent detokenization result.'
examples = (
['H', 'e', 'L', 'l', 'O', ' ', 'W', 'o', 'R', 'l', 'D', '!'],
['h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd', '!'],
)
for tokens in examples:
self.assertEqual(self.cased_tokenizer.detokenize(tokens),
self.uncased_tokenizer.detokenize(tokens),
msg=msg)
class TestNormalize(unittest.TestCase):
r"""Test case for `lmp.tokenizer.CharDictTokenizer.normalize`."""
def setUp(self):
r"""Setup both cased and uncased tokenizer instances."""
self.cased_tokenizer = CharDictTokenizer()
self.uncased_tokenizer = CharDictTokenizer(is_uncased=True)
self.tokenizers = [self.cased_tokenizer, self.uncased_tokenizer]
def tearDown(self):
r"""Delete both cased and uncased tokenizer instances."""
del self.tokenizers
del self.cased_tokenizer
del self.uncased_tokenizer
gc.collect()
def test_signature(self):
r"""Ensure signature consistency."""
msg = 'Inconsistent method signature.'
self.assertEqual(
inspect.signature(CharDictTokenizer.normalize),
inspect.Signature(parameters=[
inspect.Parameter(name='self',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
default=inspect.Parameter.empty),
inspect.Parameter(name='sequence',
kind=inspect.Parameter.POSITIONAL_OR_KEYWORD,
annotation=str,
default=inspect.Parameter.empty)
],
return_annotation=str),
msg=msg)
def test_invalid_input_sequence(self):
r"""Raise `TypeError` when input `sequence` is invalid."""
msg1 = 'Must raise `TypeError` when input `sequence` is invalid.'
msg2 = 'Inconsistent error message.'
examples = (
False,
True,
0,
1,
-1,
0.0,
1.0,
math.nan,
-math.nan,
math.inf,
-math.inf,
0j,
1j,
b'',
(),
[],
{},
set(),
object(),
lambda x: x,
type,
None,
NotImplemented,
...,
)
for invalid_input in examples:
for tokenizer in self.tokenizers:
with self.assertRaises(TypeError, msg=msg1) as cxt_man:
tokenizer.normalize(sequence=invalid_input)
self.assertEqual(cxt_man.exception.args[0],
'`sequence` must be an instance of `str`.',
msg=msg2)
def test_return_type(self):
r"""Return `str`."""
msg = 'Must return `str`.'
examples = (
'Hello world!',
'I am a legend.',
'y = f(x)',
'',
)
for sequence in examples:
for tokenizer in self.tokenizers:
self.assertIsInstance(tokenizer.normalize(sequence=sequence),
str,
msg=msg)
def test_unicode_normalize(self):
r"""Return NFKC normalized characters."""
msg = 'Must return NFKC normalized characters.'
examples = (
('０', '0', 1),
('é', 'é', 1),
('０é', '0é', 2),
)
for sequence, normalized_sequence, sequence_len in examples:
for tokenizer in self.tokenizers:
out_sequence = tokenizer.normalize(sequence=sequence)
self.assertEqual(out_sequence, normalized_sequence, msg=msg)
self.assertEqual(len(out_sequence), sequence_len, msg=msg)
def test_cased_sensitive(self):
r"""Return cased sensitive sequence."""
msg = 'Return sequence must be cased sensitive.'
examples = (
('HeLlO WoRlD!', 'HeLlO WoRlD!', 'hello world!'),
('HELLO WORLD!', 'HELLO WORLD!', 'hello world!'),
('hello world!', 'hello world!', 'hello world!'),
('H', 'H', 'h'),
('h', 'h', 'h'),
)
for sequence, cased_sequence, uncased_sequence in examples:
self.assertEqual(self.cased_tokenizer.normalize(sequence),
cased_sequence,
msg=msg)
self.assertEqual(self.uncased_tokenizer.normalize(sequence),
uncased_sequence,
msg=msg)
def test_whitespace_strip(self):
r"""Strip input sequence."""
msg = 'Must strip both leading and trailing whitespace characters.'
examples = (
(' hello world!', 'hello world!'),
('hello world! ', 'hello world!'),
(' hello world! ', 'hello world!'),
(' hello world! ', 'hello world!'),
('\nhello world!\n', 'hello world!'),
(' ', ''),
('', ''),
)
for sequence, stripped_sequence in examples:
for tokenizer in self.tokenizers:
self.assertEqual(tokenizer.normalize(sequence),
stripped_sequence,
msg=msg)
def test_whitespace_collapse(self):
r"""Collapse whitespace characters."""
msg = ('Must convert consecutive whitespace characters into single '
'whitespace character.')
examples = (
('hello world !', 'hello world !'),
('hello world !', 'hello world !'),
('hello world !', 'hello world !'),
('hello world\n\n!', 'hello world !'),
)
for sequence, ans_tokens in examples:
for tokenizer in self.tokenizers:
self.assertEqual(tokenizer.normalize(sequence),
ans_tokens,
msg=msg)