def _whitespace_tokenize_with_offsets_encode_decode_wrapper(
self, input_tensor):
"""Tokenizes a tensor of UTF-8 strings with rank of 1.
Args:
input_tensor: The single dimensional Tensor to tokenize.
Returns:
Tuple of RaggedTensors of tokenized text and byte offsets, with shapes
[num_strings, (num_tokens or num_offsets)].
"""
# Decode the strings and get byte offsets
(codepoints, byte_start_offsets) = (
ragged_string_ops.unicode_decode_with_offsets(input_tensor, "UTF-8"))
byte_limit_offsets = array_ops.concat([
byte_start_offsets[:, 1:],
math_ops.cast(
array_ops.expand_dims(string_ops.string_length(input_tensor), 1),
dtypes.int64)
], 1)
# Tokenize
(codepoint_tokens, codepoint_start_offsets, codepoint_limit_offsets) = (
self._whitespace_tokenize_codepoints_with_offsets(codepoints))
# Encode the codepoints and translate the codepoint offsets to byte offsets.
return (ragged_string_ops.unicode_encode(codepoint_tokens, "UTF-8"),
array_ops.batch_gather(byte_start_offsets, codepoint_start_offsets),
array_ops.batch_gather(
byte_limit_offsets,
math_ops.subtract(codepoint_limit_offsets, [1])))
def testLegacyPositionalName(self):
# Code that predates the 'unit' parameter may have used a positional
# argument for the 'name' parameter. Check that we don't break such code.
strings = [[["1", "12"], ["123", "1234"], ["12345", "123456"]]]
lengths = string_ops.string_length(strings, "some_name")
with self.test_session():
self.assertAllEqual(lengths.eval(), [[[1, 2], [3, 4], [5, 6]]])
def tokenize_with_offsets(self, input): # pylint: disable=redefined-builtin
"""Tokenizes a tensor of UTF-8 strings to Unicode characters.
Returned token tensors are of integer type.
Args:
input: A `RaggedTensor`or `Tensor` of UTF-8 strings with any shape.
Returns:
A tuple `(tokens, start_offsets, end_offsets)` where:
* `tokens`: A `RaggedTensor` of codepoints (integer type).
* `start_offsets`: A `RaggedTensor` of the tokens' starting byte offset.
* `end_offsets`: A `RaggedTensor` of the tokens' ending byte offset.
"""
name = None
with ops.name_scope(name, "UnicodeCharTokenize", [input]):
input_tensor = ragged_tensor.convert_to_tensor_or_ragged_tensor(
input)
(codepoints, byte_start_offsets) = (
ragged_string_ops.unicode_decode_with_offsets(
input_tensor, "UTF-8"))
strlens = math_ops.cast(
array_ops.expand_dims(string_ops.string_length(input_tensor),
-1), dtypes.int64)
# Adjust strlens to set 0-length strings to empty array (there will be no
# tokens in this case).
final_ends = ragged_array_ops.boolean_mask(strlens, strlens > 0)
byte_end_offsets = array_ops.concat(
[byte_start_offsets[..., 1:], final_ends], -1)
return codepoints, byte_start_offsets, byte_end_offsets
def testStringLength(self):
strings = [[["1", "12"], ["123", "1234"], ["12345", "123456"]]]
with self.cached_session() as sess:
lengths = string_ops.string_length(strings)
values = sess.run(lengths)
self.assertAllEqual(values, [[[1, 2], [3, 4], [5, 6]]])
def testLegacyPositionalName(self):
# Code that predates the 'unit' parameter may have used a positional
# argument for the 'name' parameter. Check that we don't break such code.
strings = [[["1", "12"], ["123", "1234"], ["12345", "123456"]]]
lengths = string_ops.string_length(strings, "some_name")
with self.test_session():
self.assertAllEqual(lengths.eval(), [[[1, 2], [3, 4], [5, 6]]])
def testStringLength(self):
strings = [[["1", "12"], ["123", "1234"], ["12345", "123456"]]]
with self.cached_session() as sess:
lengths = string_ops.string_length(strings)
values = sess.run(lengths)
self.assertAllEqual(values, [[[1, 2], [3, 4], [5, 6]]])
def testUnit(self):
unicode_strings = [u"H\xc3llo", u"\U0001f604"]
utf8_strings = [s.encode("utf-8") for s in unicode_strings]
expected_utf8_byte_lengths = [6, 4]
expected_utf8_char_lengths = [5, 1]
with self.test_session() as sess:
utf8_byte_lengths = string_ops.string_length(utf8_strings, unit="BYTE")
utf8_char_lengths = string_ops.string_length(
utf8_strings, unit="UTF8_CHAR")
self.assertAllEqual(
sess.run(utf8_byte_lengths), expected_utf8_byte_lengths)
self.assertAllEqual(
sess.run(utf8_char_lengths), expected_utf8_char_lengths)
with self.assertRaisesRegexp(
ValueError, "Attr 'unit' of 'StringLength' Op passed string 'XYZ' "
'not in: "BYTE", "UTF8_CHAR"'):
string_ops.string_length(utf8_strings, unit="XYZ")
def testUnit(self):
unicode_strings = [u"H\xc3llo", u"\U0001f604"]
utf8_strings = [s.encode("utf-8") for s in unicode_strings]
expected_utf8_byte_lengths = [6, 4]
expected_utf8_char_lengths = [5, 1]
with self.session() as sess:
utf8_byte_lengths = string_ops.string_length(utf8_strings, unit="BYTE")
utf8_char_lengths = string_ops.string_length(
utf8_strings, unit="UTF8_CHAR")
self.assertAllEqual(
self.evaluate(utf8_byte_lengths), expected_utf8_byte_lengths)
self.assertAllEqual(
self.evaluate(utf8_char_lengths), expected_utf8_char_lengths)
with self.assertRaisesRegexp(
ValueError, "Attr 'unit' of 'StringLength' Op passed string 'XYZ' "
'not in: "BYTE", "UTF8_CHAR"'):
string_ops.string_length(utf8_strings, unit="XYZ")
def f1(self, x):
return string_ops.string_length(
string_ops.string_format('{}', x))
def fn(x):
return string_ops.string_length(
string_ops.string_format('{}', x)) # COMMENT4
