SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Fixing Dataset for TQA + token-classification. (#14658)

Browse Source 
* Fixing Dataset for TQA + token-classification.

* Fixing the tests.

* Making sure `offset_mappings` is a valid argument.
This commit is contained in:
Nicolas Patry
2021-12-08 09:54:24 +01:00
committed by GitHub
parent fae0b9faef
commit 2e12d90b9e
5 changed files with 58 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/transformers/pipelines/table_question_answering.py
View File

@@ -1,4 +1,5 @@
import collections
import types
import numpy as np
@@ -9,7 +10,7 @@ from ..file_utils import (
is_torch_available,
requires_backends,
)
from .base import PIPELINE_INIT_ARGS, ArgumentHandler, Pipeline, PipelineException
from .base import PIPELINE_INIT_ARGS, ArgumentHandler, Dataset, Pipeline, PipelineException
if is_torch_available():
@@ -58,6 +59,8 @@ class TableQuestionAnsweringArgumentHandler(ArgumentHandler):
f"If keyword argument `table` is a list of dictionaries, each dictionary should have a `table` "
f"and `query` key, but only dictionary has keys {table[0].keys()} `table` and `query` keys."
)
elif Dataset is not None and isinstance(table, Dataset) or isinstance(table, types.GeneratorType):
return table
else:
raise ValueError(
f"Invalid input. Keyword argument `table` should be either of type `dict` or `list`, but "

35
src/transformers/pipelines/token_classification.py
View File

@@ -1,3 +1,4 @@
import types
import warnings
from typing import List, Optional, Tuple, Union
@@ -5,7 +6,7 @@ import numpy as np
from ..file_utils import ExplicitEnum, add_end_docstrings, is_tf_available, is_torch_available
from ..models.bert.tokenization_bert import BasicTokenizer
from .base import PIPELINE_INIT_ARGS, ArgumentHandler, Pipeline
from .base import PIPELINE_INIT_ARGS, ArgumentHandler, Dataset, Pipeline
if is_tf_available():
@@ -28,6 +29,8 @@ class TokenClassificationArgumentHandler(ArgumentHandler):
elif isinstance(inputs, str):
inputs = [inputs]
batch_size = 1
elif Dataset is not None and isinstance(inputs, Dataset) or isinstance(inputs, types.GeneratorType):
return inputs, None
else:
raise ValueError("At least one input is required.")
@@ -112,8 +115,13 @@ class TokenClassificationPipeline(Pipeline):
grouped_entities: Optional[bool] = None,
ignore_subwords: Optional[bool] = None,
aggregation_strategy: Optional[AggregationStrategy] = None,
offset_mapping: Optional[List[Tuple[int, int]]] = None,
):
preprocess_params = {}
if offset_mapping is not None:
preprocess_params["offset_mapping"] = offset_mapping
postprocess_params = {}
if grouped_entities is not None or ignore_subwords is not None:
if grouped_entities and ignore_subwords:
@@ -147,7 +155,7 @@ class TokenClassificationPipeline(Pipeline):
postprocess_params["aggregation_strategy"] = aggregation_strategy
if ignore_labels is not None:
postprocess_params["ignore_labels"] = ignore_labels
return {}, {}, postprocess_params
return preprocess_params, {}, postprocess_params
def __call__(self, inputs: Union[str, List[str]], **kwargs):
"""
@@ -174,12 +182,13 @@ class TokenClassificationPipeline(Pipeline):
Only exists if the offsets are available within the tokenizer
"""
_inputs, offset_mappings = self._args_parser(inputs, **kwargs)
self.offset_mappings = offset_mappings
_inputs, offset_mapping = self._args_parser(inputs, **kwargs)
if offset_mapping:
kwargs["offset_mapping"] = offset_mapping
return super().__call__(inputs, **kwargs)
def preprocess(self, sentence):
def preprocess(self, sentence, offset_mapping=None):
truncation = True if self.tokenizer.model_max_length and self.tokenizer.model_max_length > 0 else False
model_inputs = self.tokenizer(
sentence,
@@ -189,8 +198,7 @@ class TokenClassificationPipeline(Pipeline):
return_special_tokens_mask=True,
return_offsets_mapping=self.tokenizer.is_fast,
)
if self.offset_mappings:
offset_mapping = self.offset_mappings[0]
if offset_mapping:
model_inputs["offset_mapping"] = offset_mapping
model_inputs["sentence"] = sentence
@@ -262,12 +270,13 @@ class TokenClassificationPipeline(Pipeline):
word = self.tokenizer.convert_ids_to_tokens(int(input_ids[idx]))
if offset_mapping is not None:
start_ind, end_ind = offset_mapping[idx]
if self.framework == "pt":
start_ind = start_ind.item()
end_ind = end_ind.item()
else:
start_ind = int(start_ind.numpy())
end_ind = int(end_ind.numpy())
if not isinstance(start_ind, int):
if self.framework == "pt":
start_ind = start_ind.item()
end_ind = end_ind.item()
else:
start_ind = int(start_ind.numpy())
end_ind = int(end_ind.numpy())
word_ref = sentence[start_ind:end_ind]
if getattr(self.tokenizer._tokenizer.model, "continuing_subword_prefix", None):
# This is a BPE, word aware tokenizer, there is a correct way

7
tests/test_pipelines_common.py
View File

@@ -183,9 +183,12 @@ class PipelineTestCaseMeta(type):
# 10 examples with batch size 4 means there needs to be a unfinished batch
# which is important for the unbatcher
dataset = [copy.deepcopy(random.choice(examples)) for i in range(10)]
def data(n):
for _ in range(n):
# Need to copy because Conversation object is mutated
yield copy.deepcopy(random.choice(examples))
for item in pipeline(dataset, batch_size=4):
for item in pipeline(data(10), batch_size=4):
pass
run_batch_test(pipeline, examples)

18
tests/test_pipelines_table_question_answering.py
View File

@@ -35,17 +35,16 @@ from transformers.testing_utils import (
from .test_pipelines_common import PipelineTestCaseMeta
@require_tensorflow_probability
@require_torch_scatter
@require_torch
@require_pandas
@is_pipeline_test
class TQAPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
# Putting it there for consistency, but TQA do not have fast tokenizer
# which are needed to generate automatic tests
model_mapping = MODEL_FOR_TABLE_QUESTION_ANSWERING_MAPPING
@require_tensorflow_probability
@require_pandas
@require_tf
@require_torch
def test_small_model_tf(self):
model_id = "lysandre/tiny-tapas-random-wtq"
model = TFAutoModelForTableQuestionAnswering.from_pretrained(model_id, from_pt=True)
@@ -147,6 +146,7 @@ class TQAPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
)
@require_torch
@require_torch_scatter
def test_small_model_pt(self):
model_id = "lysandre/tiny-tapas-random-wtq"
model = AutoModelForTableQuestionAnswering.from_pretrained(model_id)
@@ -248,6 +248,7 @@ class TQAPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
)
@require_torch
@require_torch_scatter
def test_slow_tokenizer_sqa_pt(self):
model_id = "lysandre/tiny-tapas-random-sqa"
model = AutoModelForTableQuestionAnswering.from_pretrained(model_id)
@@ -366,6 +367,9 @@ class TQAPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
)
@require_tf
@require_tensorflow_probability
@require_pandas
@require_torch
def test_slow_tokenizer_sqa_tf(self):
model_id = "lysandre/tiny-tapas-random-sqa"
model = TFAutoModelForTableQuestionAnswering.from_pretrained(model_id, from_pt=True)
@@ -484,6 +488,7 @@ class TQAPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
)
@slow
@require_torch_scatter
def test_integration_wtq_pt(self):
table_querier = pipeline("table-question-answering")
@@ -528,6 +533,8 @@ class TQAPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
self.assertListEqual(results, expected_results)
@slow
@require_tensorflow_probability
@require_pandas
def test_integration_wtq_tf(self):
model_id = "google/tapas-base-finetuned-wtq"
model = TFAutoModelForTableQuestionAnswering.from_pretrained(model_id)
@@ -575,6 +582,7 @@ class TQAPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
self.assertListEqual(results, expected_results)
@slow
@require_torch_scatter
def test_integration_sqa_pt(self):
table_querier = pipeline(
"table-question-answering",
@@ -598,6 +606,8 @@ class TQAPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
self.assertListEqual(results, expected_results)
@slow
@require_tensorflow_probability
@require_pandas
def test_integration_sqa_tf(self):
model_id = "google/tapas-base-finetuned-sqa"
model = TFAutoModelForTableQuestionAnswering.from_pretrained(model_id)

13
tests/test_pipelines_token_classification.py
View File

@@ -636,6 +636,19 @@ class TokenClassificationPipelineTests(unittest.TestCase, metaclass=PipelineTest
[],
)
token_classifier = pipeline(task="token-classification", model=model_name, framework="pt")
# Overload offset_mapping
outputs = token_classifier(
"This is a test !", offset_mapping=[(0, 0), (0, 1), (0, 2), (0, 0), (0, 0), (0, 0), (0, 0)]
)
self.assertEqual(
nested_simplify(outputs),
[
{"entity": "I-MISC", "score": 0.115, "index": 1, "word": "this", "start": 0, "end": 1},
{"entity": "I-MISC", "score": 0.115, "index": 2, "word": "is", "start": 0, "end": 2},
],
)
@require_torch
def test_pt_ignore_subwords_slow_tokenizer_raises(self):
model_name = "sshleifer/tiny-dbmdz-bert-large-cased-finetuned-conll03-english"