def _convert_to_ragged_inputs(self, inputs):
"""Transforms the text batch inputs to a ragged shape."""
if isinstance(self.input_data, ragged_tensor.RaggedTensor):
return inputs
inputs = text_ops.WhitespaceTokenizer().tokenize(inputs)
return inputs
def benchmark_pad_along_dimension(self):
self.input_data = text_ops.WhitespaceTokenizer().tokenize(self.input_data)
self._run(text_ops.pad_along_dimension, {
"axis": -1,
"right_pad": ["RP"],
"left_pad": ["LP"]
})
def benchmark_wordpiece_tokenizer(self):
self.input_data = text_ops.WhitespaceTokenizer().tokenize(
self.input_data)
tokenizer = text_ops.WordpieceTokenizer(
vocab_lookup_table=self._create_table((_BERT_VOCAB_PATH)),
unknown_token=None,
token_out_type=dtypes.int64)
self._run(tokenizer)
def benchmark_ngrams(self):
self.input_data = text_ops.WhitespaceTokenizer().tokenize(self.input_data)
self._run(
text_ops.ngrams, {
"width": 2,
"axis": -1,
"reduction_type": text_ops.Reduction.STRING_JOIN,
"string_separator": "|"
})
def benchmark_sliding_window(self):
self.input_data = text_ops.WhitespaceTokenizer().tokenize(
self.input_data)
self._run(text_ops.sliding_window, {"width": 3, "axis": -1})