def testSplitWithSparseOutput(self, texts, expected): input_tensor = np.array(_nested_encode(texts, "UTF-8"), dtype=bytes) result = ragged_string_ops.unicode_split(input_tensor, "UTF-8").to_sparse() self.assertIsInstance(result, sparse_tensor.SparseTensor) self.assertAllEqual(expected.indices, result.indices) self.assertAllEqual(expected.values, result.values) self.assertAllEqual(expected.dense_shape, result.dense_shape)
def testSplitWithSparseOutput(self, texts, expected): input_tensor = np.array(_nested_encode(texts, "UTF-8"), dtype=bytes) result = ragged_string_ops.unicode_split(input_tensor, "UTF-8").to_sparse() self.assertIsInstance(result, sparse_tensor.SparseTensor) self.assertAllEqual(expected.indices, result.indices) self.assertAllEqual(expected.values, result.values) self.assertAllEqual(expected.dense_shape, result.dense_shape)
def testSplitWithPaddedOutput(self, texts, expected, ragged_rank=None):
input_tensor = ragged_factory_ops.constant_value(
_nested_encode(texts, "UTF-8"),
ragged_rank=ragged_rank,
dtype=bytes)
result = ragged_string_ops.unicode_split(
input_tensor, "UTF-8").to_tensor(default_value="")
self.assertAllEqual(np.array(expected, dtype=bytes), result)
def testBasicSplit(self, texts, ragged_rank=None):
input_tensor = ragged_factory_ops.constant_value(
_nested_encode(texts, "UTF-8"),
ragged_rank=ragged_rank,
dtype=bytes)
result = ragged_string_ops.unicode_split(input_tensor, "UTF-8")
expected = _nested_splitchars(texts, "UTF-8")
self.assertAllEqual(expected, result)
def testDocstringExamples(self):
texts = [s.encode("utf8") for s in [u"G\xf6\xf6dnight", u"\U0001f60a"]]
codepoints1 = ragged_string_ops.unicode_split(texts, "UTF-8")
codepoints2, offsets = ragged_string_ops.unicode_split_with_offsets(
texts, "UTF-8")
self.assertAllEqual(codepoints1, [[
b"G", b"\xc3\xb6", b"\xc3\xb6", b"d", b"n", b"i", b"g", b"h", b"t"
], [b"\xf0\x9f\x98\x8a"]])
self.assertAllEqual(codepoints2, [[
b"G", b"\xc3\xb6", b"\xc3\xb6", b"d", b"n", b"i", b"g", b"h", b"t"
], [b"\xf0\x9f\x98\x8a"]])
self.assertAllEqual(offsets, [[0, 1, 3, 5, 6, 7, 8, 9, 10], [0]])
def testDocstringExamples(self):
texts = [s.encode("utf8") for s in [u"G\xf6\xf6dnight", u"\U0001f60a"]]
codepoints1 = ragged_string_ops.unicode_split(texts, "UTF-8")
codepoints2, offsets = ragged_string_ops.unicode_split_with_offsets(
texts, "UTF-8")
self.assertRaggedEqual(
codepoints1,
[[b"G", b"\xc3\xb6", b"\xc3\xb6", b"d", b"n", b"i", b"g", b"h", b"t"],
[b"\xf0\x9f\x98\x8a"]])
self.assertRaggedEqual(
codepoints2,
[[b"G", b"\xc3\xb6", b"\xc3\xb6", b"d", b"n", b"i", b"g", b"h", b"t"],
[b"\xf0\x9f\x98\x8a"]])
self.assertRaggedEqual(offsets, [[0, 1, 3, 5, 6, 7, 8, 9, 10], [0]])
def tokenize(self, text_input):
"""Performs basic word tokenization for BERT.
Args:
text_input: A `Tensor` or `RaggedTensor` of untokenized UTF-8 strings.
Returns:
A `RaggedTensor` of tokenized strings from text_input.
"""
# lowercase and strip accents (if option is set)
if self._lower_case:
text_input = case_fold_utf8(text_input)
text_input = normalize_utf8(text_input, "NFD")
text_input = string_ops.regex_replace(text_input, r"\p{Mn}", "")
else:
# utf8 normalization
if self._normalization_form is not None:
text_input = normalize_utf8(text_input, self._normalization_form)
# strip out control characters
text_input = string_ops.regex_replace(text_input, r"\p{Cc}|\p{Cf}", " ")
# For chinese and emoji characters, tokenize by unicode codepoints
unicode_tokenizer = UnicodeScriptTokenizer(
keep_whitespace=self._keep_whitespace)
script_tokenized = unicode_tokenizer.tokenize(text_input)
split_cond = self._should_split(script_tokenized)
unicode_char_split = ragged_string_ops.unicode_split(
script_tokenized, "UTF-8")
unicode_split_tokens = array_ops.where(
array_ops.squeeze(split_cond),
y=array_ops.expand_dims(script_tokenized.values, axis=1),
x=unicode_char_split.values)
final_tokens = script_tokenized.with_flat_values(unicode_split_tokens)
return final_tokens.merge_dims(-2, -1)
def testScalarSplit(self):
text = constant_op.constant(u"仅今年前".encode("UTF-8"))
chars = ragged_string_ops.unicode_split(text, "UTF-8")
self.assertAllEqual(chars, [c.encode("UTF-8") for c in u"仅今年前"])
def testSplitWithDifferentEncodings(self, encoding, texts): expected = _nested_splitchars(texts, encoding) input_tensor = constant_op.constant(_nested_encode(texts, encoding)) result = ragged_string_ops.unicode_split(input_tensor, encoding) self.assertAllEqual(expected, result)
def testExceptions(self, exception=None, message=None, **args):
with self.assertRaisesRegex(exception, message):
self.evaluate(ragged_string_ops.unicode_split(**args))
def testBasicSplit(self, texts, ragged_rank=None):
input_tensor = ragged_factory_ops.constant_value(
_nested_encode(texts, "UTF-8"), ragged_rank=ragged_rank, dtype=bytes)
result = ragged_string_ops.unicode_split(input_tensor, "UTF-8")
expected = _nested_splitchars(texts, "UTF-8")
self.assertRaggedEqual(expected, result)
def testErrorModes(self, expected=None, **args): result = ragged_string_ops.unicode_split(**args) self.assertAllEqual(expected, result)
def testScalarSplit(self):
text = constant_op.constant(u"仅今年前".encode("UTF-8"))
chars = ragged_string_ops.unicode_split(text, "UTF-8")
self.assertAllEqual(chars, [c.encode("UTF-8") for c in u"仅今年前"])
def testVectorSplit(self):
text = constant_op.constant([u"仅今年前".encode("UTF-8"), b"hello"])
chars = ragged_string_ops.unicode_split(text, "UTF-8")
expected_chars = [[c.encode("UTF-8") for c in u"仅今年前"],
[c.encode("UTF-8") for c in u"hello"]]
self.assertAllEqual(chars, expected_chars)
def testSplitWithPaddedOutput(self, texts, expected, ragged_rank=None):
input_tensor = ragged_factory_ops.constant_value(
_nested_encode(texts, "UTF-8"), ragged_rank=ragged_rank, dtype=bytes)
result = ragged_string_ops.unicode_split(
input_tensor, "UTF-8").to_tensor(default_value="")
self.assertAllEqual(np.array(expected, dtype=bytes), result)
def testErrorModes(self, expected=None, **args): result = ragged_string_ops.unicode_split(**args) self.assertRaggedEqual(expected, result)
def testVectorSplit(self):
text = constant_op.constant([u"仅今年前".encode("UTF-8"), b"hello"])
chars = ragged_string_ops.unicode_split(text, "UTF-8")
expected_chars = [[c.encode("UTF-8") for c in u"仅今年前"],
[c.encode("UTF-8") for c in u"hello"]]
self.assertRaggedEqual(chars, expected_chars)
def testExceptions(self, exception=None, message=None, **args):
with self.assertRaisesRegexp(exception, message):
self.evaluate(ragged_string_ops.unicode_split(**args))
def testSplitWithDifferentEncodings(self, encoding, texts): expected = _nested_splitchars(texts, encoding) input_tensor = constant_op.constant(_nested_encode(texts, encoding)) result = ragged_string_ops.unicode_split(input_tensor, encoding) self.assertRaggedEqual(expected, result)
