def test_eq(self):
examples = self._regular_examples()
example_0 = examples[0]
example_1 = examples[1]
example_2 = examples[2]
assert example_0 != example_1
assert example_2 == example_2
assert example_0 == example_0
del example_2['2']
assert example_2 == example_1
# Ensure that it only relies on the text_id and not the content of the
# target text
example_2['2'] = TargetText('hello how', '2')
example_1['2'] = TargetText('another test', '2')
assert example_2 == example_1
# Testing instance, if not TargetTextCollection should return False
text = 'today was good'
text_id = '1'
dict_version = [{'text_id': text_id, 'text': text}]
collection_version = TargetTextCollection(
[TargetText(**dict_version[0])])
assert dict_version != collection_version
# Should return False as they have different text_id's but have the
# content
alt_collection_version = TargetTextCollection(
[TargetText(text_id='2', text=text)])
assert collection_version != alt_collection_version
assert len(collection_version) == len(alt_collection_version)
def test_one_sample_per_span(self, remove_empty: bool):
# Case where nothing should change with respect to the number of spans
# but will change the values target_sentiments to None etc
target_text = TargetText(
text_id='0',
spans=[Span(4, 15)],
text='The laptop case was great and cover was rubbish',
target_sentiments=[0],
targets=['laptop case'])
collection = TargetTextCollection([target_text])
new_collection = collection.one_sample_per_span(
remove_empty=remove_empty)
assert new_collection == collection
assert new_collection['0']['spans'] == [Span(4, 15)]
assert new_collection['0']['target_sentiments'] == None
assert collection['0']['target_sentiments'] == [0]
# Should change the number of Spans.
assert target_text['target_sentiments'] == [0]
target_text._storage['spans'] = [Span(4, 15), Span(4, 15)]
target_text._storage['targets'] = ['laptop case', 'laptop case']
target_text._storage['target_sentiments'] = [0, 1]
diff_collection = TargetTextCollection([target_text])
new_collection = diff_collection.one_sample_per_span(
remove_empty=remove_empty)
assert new_collection == collection
assert new_collection['0']['spans'] == [Span(4, 15)]
assert new_collection['0']['target_sentiments'] == None
assert diff_collection['0']['target_sentiments'] == [0, 1]
assert diff_collection['0']['spans'] == [Span(4, 15), Span(4, 15)]
def test_force_targets(self):
text = 'The laptop casewas great and cover was rubbish'
spans = [Span(4, 15), Span(29, 34)]
targets = ['laptop case', 'cover']
target_text = TargetText(text=text,
text_id='0',
targets=targets,
spans=spans)
text_1 = 'The laptop casewas great andcover was rubbish'
spans_1 = [Span(4, 15), Span(28, 33)]
target_text_1 = TargetText(text=text_1,
text_id='1',
targets=targets,
spans=spans_1)
perfect_text = 'The laptop case was great and cover was rubbish'
perfect_spans = [Span(4, 15), Span(30, 35)]
# Test the single case
test_collection = TargetTextCollection([target_text])
test_collection.force_targets()
assert test_collection['0']['text'] == perfect_text
assert test_collection['0']['spans'] == perfect_spans
# Test the multiple case
test_collection = TargetTextCollection([target_text, target_text_1])
test_collection.force_targets()
for target_key in ['0', '1']:
assert test_collection[target_key]['text'] == perfect_text
assert test_collection[target_key]['spans'] == perfect_spans
def no_pred_values(true_sentiment_key: str, predicted_sentiment_key: str
) -> Tuple[TargetTextCollection, List[str], List[List[str]]]:
example_1 = TargetText(text_id='1', text='some text')
example_1[true_sentiment_key] = []
example_1[predicted_sentiment_key] = []
example_2 = TargetText(text_id='2', text='some text')
example_2[true_sentiment_key] = []
example_2[predicted_sentiment_key] = []
true_labels = []
pred_labels = []
return TargetTextCollection([example_1, example_2]), true_labels, pred_labels
def passable_diff_num_labels(true_sentiment_key: str,
predicted_sentiment_key: str
) -> Tuple[TargetTextCollection, List[str], List[List[str]]]:
example_1 = TargetText(text_id='1', text='some text')
example_1[true_sentiment_key] = ['pos', 'neg']
example_1[predicted_sentiment_key] = [['neg', 'pos']]
example_2 = TargetText(text_id='2', text='some text')
example_2[true_sentiment_key] = ['pos', 'neg', 'neu']
example_2[predicted_sentiment_key] = [['neg', 'neg', 'pos']]
true_labels = ['pos', 'neg', 'pos', 'neg', 'neu']
pred_labels = [['neg', 'pos', 'neg', 'neg', 'pos']]
return TargetTextCollection([example_1, example_2]), true_labels, pred_labels
def empty_preds_examples(true_sentiment_key: str,
predicted_sentiment_key: str,
labels_per_text: bool = False
) -> Tuple[TargetTextCollection, List[str], List[List[str]]]:
example_1 = TargetText(text_id='1', text='some text')
example_1[true_sentiment_key] = ['pos', 'neg']
example_1[predicted_sentiment_key] = [[]]
example_2 = TargetText(text_id='2', text='some text')
example_2[true_sentiment_key] = ['pos', 'neg', 'neu']
example_2[predicted_sentiment_key] = [[]]
true_labels = ['pos', 'neg', 'pos', 'neg', 'neu']
pred_labels = [[]]
return TargetTextCollection([example_1, example_2]), true_labels, pred_labels
def passable_example_multiple_preds(true_sentiment_key: str,
predicted_sentiment_key: str
) -> TargetTextCollection:
example_1 = TargetText(text_id='1', text='some text', targets=['some', 'text'],
spans=[Span(0,4), Span(5, 9)])
example_1[true_sentiment_key] = ['pos', 'neg']
example_1[predicted_sentiment_key] = [['pos', 'neg'], ['pos', 'neg']]
example_2 = TargetText(text_id='2', text='some text is', targets=['some', 'text', 'is'],
spans=[Span(0,4), Span(5, 9), Span(10,12)])
example_2[true_sentiment_key] = ['pos', 'neg', 'neu']
if predicted_sentiment_key == 'model_2':
example_2[predicted_sentiment_key] = [['pos', 'neg', 'neu'], ['neg', 'neg', 'pos']]
else:
example_2[predicted_sentiment_key] = [['neu', 'neg', 'pos'], ['neg', 'neg', 'pos']]
return TargetTextCollection([example_1, example_2])
def test_strict_text_accuracy(true_sentiment_key: str,
predicted_sentiment_key: str):
# Test macro F1 works as should on one set of predictions
example, _, _ = passable_example(true_sentiment_key, predicted_sentiment_key)
score = strict_text_accuracy(example, true_sentiment_key,
predicted_sentiment_key, False, False, None)
assert 0.0 == score
# Test it works on multiple predictions
example, _, _ = passable_example_multiple_preds(true_sentiment_key,
predicted_sentiment_key)
score = strict_text_accuracy(example, true_sentiment_key,
predicted_sentiment_key, True, False, None)
assert 0.25 == score
# Test it works on multiple predictions
example, _, _ = passable_example_multiple_preds(true_sentiment_key,
predicted_sentiment_key)
score = strict_text_accuracy(example, true_sentiment_key,
predicted_sentiment_key, False, True, None)
assert [(0.0), (0.5)] == score
# Test the case where the TargetCollection has a sentence/text with no
# Targets/Predictions. This should raise a ValueError.
no_target = TargetText(text='hello how are you', text_id='10',
target_sentiments=[], targets=[], spans=[])
no_target[true_sentiment_key] = []
no_target[predicted_sentiment_key] = []
target_examples, _, _ = passable_example_multiple_preds(true_sentiment_key,
predicted_sentiment_key)
all_targets = list(target_examples.values())
all_targets.append(no_target)
test_collection = TargetTextCollection(all_targets)
assert 3 == len(test_collection)
with pytest.raises(ValueError):
strict_text_accuracy(test_collection, true_sentiment_key,
predicted_sentiment_key, True, False, None)
def test_sanitize(self):
# The normal case where no errors should be raised.
target_text = TargetText(
text_id='0',
spans=[Span(4, 15)],
text='The laptop case was great and cover was rubbish',
target_sentiments=[0],
targets=['laptop case'])
collection = TargetTextCollection([target_text])
collection.sanitize()
# The case where an error should be raised
with pytest.raises(ValueError):
target_text._storage['spans'] = [Span(3, 15)]
collection = TargetTextCollection([target_text])
collection.sanitize()
def _target_text_not_align_example(self) -> TargetText:
text = 'The laptop case; was awful'
text_id = 'inf'
spans = [Span(4, 15)]
targets = ['laptop case']
return TargetText(text=text,
text_id=text_id,
spans=spans,
targets=targets)
def passable_subset_multiple_preds(true_sentiment_key: str,
predicted_sentiment_key: str,
labels_per_text: bool = False
) -> Tuple[TargetTextCollection, List[str], List[List[str]]]:
example_1 = TargetText(text_id='1', text='some text')
example_1[true_sentiment_key] = ['pos', 'neg']
example_1[predicted_sentiment_key] = [['neg', 'neg'], ['neg', 'neg']]
example_2 = TargetText(text_id='2', text='some text')
example_2[true_sentiment_key] = ['pos', 'neg', 'neu']
example_2[predicted_sentiment_key] = [['neu', 'neg', 'neu'], ['neg', 'neg', 'neu']]
true_labels = ['pos', 'neg', 'pos', 'neg', 'neu']
pred_labels = [['neg', 'neg', 'neu', 'neg', 'neu'],
['neg', 'neg', 'neg', 'neg', 'neu']]
if labels_per_text:
true_labels = [['pos', 'neg'], ['pos', 'neg', 'neu']]
pred_labels = [[['neg', 'neg'], ['neu', 'neg', 'neu']],
[['neg', 'neg'], ['neg', 'neg', 'neu']]]
return TargetTextCollection([example_1, example_2]), true_labels, pred_labels
def _target_text_measure_examples(self) -> List[TargetText]:
text = 'The laptop case was great and cover was rubbish'
text_id = '0'
spans = [Span(4, 15), Span(30, 35)]
targets = ['laptop case', 'cover']
target_text_0 = TargetText(text=text,
text_id=text_id,
spans=spans,
targets=targets)
text = 'The laptop price was awful'
text_id = '1'
spans = [Span(4, 16)]
targets = ['laptop price']
target_text_1 = TargetText(text=text,
text_id=text_id,
spans=spans,
targets=targets)
return [target_text_0, target_text_1]
def test_get_item(self):
examples = self._regular_examples()
example_2 = examples[2]
last_target_text = self._target_text_example()
assert example_2['2'] == last_target_text
assert example_2['0'] == TargetText(
'can be any text as long as id is correct', '0')
with pytest.raises(KeyError):
example_2['any key']
def test_target_count(self):
# Start with an empty collection
test_collection = TargetTextCollection()
nothing = test_collection.target_count()
assert len(nothing) == 0
assert not nothing
# Collection that contains TargetText instances but with no targets
test_collection.add(TargetText(text='some text', text_id='1'))
assert len(test_collection) == 1
nothing = test_collection.target_count()
assert len(nothing) == 0
assert not nothing
# Collection now contains at least one target
test_collection.add(
TargetText(text='another item today',
text_id='2',
spans=[Span(0, 12)],
targets=['another item']))
assert len(test_collection) == 2
one = test_collection.target_count()
assert len(one) == 1
assert one == {'another item': 1}
# Collection now contains 3 targets but 2 are the same
test_collection.add(
TargetText(text='another item today',
text_id='3',
spans=[Span(0, 12)],
targets=['another item']))
test_collection.add(
TargetText(text='item today',
text_id='4',
spans=[Span(0, 4)],
targets=['item']))
assert len(test_collection) == 4
two = test_collection.target_count()
assert len(two) == 2
assert two == {'another item': 2, 'item': 1}
def test_samples_with_targets(self):
# Test the case where all of the TargetTextCollection contain targets
test_collection = TargetTextCollection(self._target_text_examples())
sub_collection = test_collection.samples_with_targets()
assert test_collection == sub_collection
assert len(test_collection) == 3
# Test the case where none of the TargetTextCollection contain targets
for sample_id in list(test_collection.keys()):
del test_collection[sample_id]
test_collection.add(TargetText(text='nothing', text_id=sample_id))
assert len(test_collection) == 3
sub_collection = test_collection.samples_with_targets()
assert len(sub_collection) == 0
assert sub_collection != test_collection
# Test the case where only a 2 of the the three TargetTextCollection
# contain targets.
test_collection = TargetTextCollection(self._target_text_examples())
del test_collection['another_id']
easy_case = TargetText(text='something else', text_id='another_id')
test_collection.add(easy_case)
sub_collection = test_collection.samples_with_targets()
assert len(sub_collection) == 2
assert sub_collection != test_collection
# Test the case where the targets are just an empty list rather than
# None
test_collection = TargetTextCollection(self._target_text_examples())
del test_collection['another_id']
edge_case = TargetText(text='something else',
text_id='another_id',
targets=[],
spans=[])
test_collection.add(edge_case)
sub_collection = test_collection.samples_with_targets()
assert len(sub_collection) == 2
assert sub_collection != test_collection
def parse_tweet(tweet_data: Dict[str, Any], annotation_data: Dict[str,
Any],
tweet_id: str) -> TargetText:
'''
:params tweet_data: Data containing the Tweet information
:params annotation_data: Data containing the annotation data on the
Tweet
:params tweet_id: ID of the Tweet
:returns: The Tweet data in
:class:`target_extraction.data_types.TargetText` format
:raises ValueError: If the Target offset cannot be found.
'''
def get_offsets(from_offset: int, tweet_text: str,
target: str) -> Span:
offset_shifts = [0, -1, 1]
for offset_shift in offset_shifts:
from_offset_shift = from_offset + offset_shift
to_offset = from_offset_shift + len(target)
offsets = Span(from_offset_shift, to_offset)
offset_text = tweet_text[from_offset_shift:to_offset].lower()
if offset_text == target.lower():
return offsets
raise ValueError(
f'Offset {from_offset} does not match target text'
f' {target}. Full text {tweet_text}\nid {tweet_id}')
target_id = str(tweet_id)
target_text = tweet_data['content']
target_categories = None
target_category_sentiments = None
targets = []
target_spans = []
target_sentiments = []
for entity in tweet_data['entities']:
target_sentiment = annotation_data['items'][str(entity['id'])]
if target_sentiment == 'doesnotapply':
continue
target = entity['entity']
target_span = get_offsets(entity['offset'], target_text, target)
# Take the target from the text as sometimes the original label
# is lower cased when it should not be according to the text.
target = target_text[target_span.start:target_span.end]
targets.append(target)
target_spans.append(target_span)
target_sentiments.append(target_sentiment)
return TargetText(target_text, target_id, targets, target_spans,
target_sentiments, target_categories,
target_category_sentiments)
def test_set_item(self):
new_collection = TargetTextCollection()
# Full target text.
new_collection['2'] = self._target_text_example()
# Minimum target text
new_collection['2'] = TargetText('minimum example', '2')
example_args = {'text': 'minimum example', 'text_id': '2'}
new_collection['2'] = TargetText(**example_args)
with pytest.raises(ValueError):
new_collection['2'] = TargetText('minimum example', '3')
with pytest.raises(TypeError):
new_collection['2'] = example_args
# Ensure that if the given TargetText changes it does not change in
# the collection
example_instance = TargetText(**example_args)
example_collection = TargetTextCollection()
example_collection['2'] = example_instance
example_instance['target_sentiments'] = [0]
assert example_instance['target_sentiments'] is not None
assert example_collection['2']['target_sentiments'] is None
def _target_text_examples(self) -> List[TargetText]:
text = 'The laptop case was great and cover was rubbish'
text_ids = ['0', 'another_id', '2']
spans = [[Span(4, 15)], [Span(30, 35)], [Span(4, 15), Span(30, 35)]]
target_sentiments = [[0], [1], [0, 1]]
targets = [['laptop case'], ['cover'], ['laptop case', 'cover']]
categories = [['LAPTOP#CASE'], ['LAPTOP'], ['LAPTOP#CASE', 'LAPTOP']]
target_text_examples = []
for i in range(3):
example = TargetText(text,
text_ids[i],
targets=targets[i],
spans=spans[i],
target_sentiments=target_sentiments[i],
categories=categories[i])
target_text_examples.append(example)
return target_text_examples
def test_target_length_plot():
# standard/normal case
ax = target_length_plot([TRAIN_COLLECTION], 'targets', whitespace())
del ax
# cumulative percentage True
ax = target_length_plot([TRAIN_COLLECTION],
'targets',
whitespace(),
cumulative_percentage=True)
del ax
# Max target length
ax = target_length_plot([TRAIN_COLLECTION],
'targets',
whitespace(),
cumulative_percentage=True,
max_target_length=1)
del ax
# Can take consume an axes
fig, alt_ax = plt.subplots(1, 1)
ax = target_length_plot([TRAIN_COLLECTION],
'targets',
whitespace(),
cumulative_percentage=True,
max_target_length=1,
ax=alt_ax)
assert alt_ax == ax
del ax
plt.close(fig)
# Can take more than one collection
alt_collection = copy.deepcopy(list(TRAIN_COLLECTION.dict_iterator()))
alt_collection = [TargetText(**v) for v in alt_collection]
alt_collection = TargetTextCollection(alt_collection)
alt_collection.name = 'Another'
ax = target_length_plot([TRAIN_COLLECTION, alt_collection],
'targets',
whitespace(),
cumulative_percentage=True,
max_target_length=1)
assert alt_ax != ax
del alt_ax
del ax
def test_overall_metric_results(true_sentiment_key: str,
include_metadata: bool,
strict_accuracy_metrics: bool):
if true_sentiment_key is None:
true_sentiment_key = 'target_sentiments'
model_1_collection = passable_example_multiple_preds(true_sentiment_key, 'model_1')
model_1_collection.add(TargetText(text='a', text_id='200', spans=[Span(0,1)], targets=['a'],
model_1=[['pos'], ['neg']], **{f'{true_sentiment_key}': ['pos']}))
model_1_collection.add(TargetText(text='a', text_id='201', spans=[Span(0,1)], targets=['a'],
model_1=[['pos'], ['neg']], **{f'{true_sentiment_key}': ['neg']}))
model_1_collection.add(TargetText(text='a', text_id='202', spans=[Span(0,1)], targets=['a'],
model_1=[['pos'], ['neg']], **{f'{true_sentiment_key}': ['neu']}))
print(true_sentiment_key)
print(model_1_collection['1'])
print(model_1_collection['200'])
model_2_collection = passable_example_multiple_preds(true_sentiment_key, 'model_2')
combined_collection = TargetTextCollection()
standard_columns = ['Dataset', 'Macro F1', 'Accuracy', 'run number',
'prediction key']
if strict_accuracy_metrics:
standard_columns = standard_columns + ['STAC', 'STAC 1', 'STAC Multi']
if include_metadata:
metadata = {'predicted_target_sentiment_key': {'model_1': {'CWR': True},
'model_2': {'CWR': False}}}
combined_collection.name = 'name'
combined_collection.metadata = metadata
standard_columns.append('CWR')
number_df_columns = len(standard_columns)
for key, value in model_1_collection.items():
if key in ['200', '201', '202']:
combined_collection.add(value)
combined_collection[key]['model_2'] = [['neg'], ['pos']]
continue
combined_collection.add(value)
combined_collection[key]['model_2'] = model_2_collection[key]['model_2']
if true_sentiment_key is None:
result_df = util.overall_metric_results(combined_collection,
['model_1', 'model_2'],
strict_accuracy_metrics=strict_accuracy_metrics)
else:
result_df = util.overall_metric_results(combined_collection,
['model_1', 'model_2'],
true_sentiment_key,
strict_accuracy_metrics=strict_accuracy_metrics)
assert (4, number_df_columns) == result_df.shape
assert set(standard_columns) == set(result_df.columns)
if include_metadata:
assert ['name'] * 4 == result_df['Dataset'].tolist()
else:
assert [''] * 4 == result_df['Dataset'].tolist()
# Test the case where only one model is used
if true_sentiment_key is None:
result_df = util.overall_metric_results(combined_collection,
['model_1'],
strict_accuracy_metrics=strict_accuracy_metrics)
else:
result_df = util.overall_metric_results(combined_collection,
['model_1'],
true_sentiment_key,
strict_accuracy_metrics=strict_accuracy_metrics)
assert (2, number_df_columns) == result_df.shape
# Test the case where the model names come from the metadata
if include_metadata:
result_df = util.overall_metric_results(combined_collection,
true_sentiment_key=true_sentiment_key,
strict_accuracy_metrics=strict_accuracy_metrics)
assert (4, number_df_columns) == result_df.shape
else:
with pytest.raises(KeyError):
util.overall_metric_results(combined_collection,
true_sentiment_key=true_sentiment_key,
strict_accuracy_metrics=strict_accuracy_metrics)
def multi_aspect_multi_sentiment_atsa(
dataset: str,
cache_dir: Optional[Path] = None,
original: bool = True) -> TargetTextCollection:
'''
The data for this function when downloaded is stored within:
`Path(cache_dir, 'Jiang 2019 MAMS ATSA')
:NOTE: That as each sentence/`TargetText` object has to have
a `text_id`, as no ids exist in this dataset the ids are created
based on when the sentence occurs in the dataset e.g. the first
sentence/`TargetText` object id is '0'
:param dataset: Either `train`, `val` or `test`, determines the dataset that
is returned.
:param cache_dir: The directory where all of the data is stored for
this code base. If None then the cache directory is
`dataset_parsers.CACHE_DIRECTORY`
:param original: This does not affect `val` or `test`. If True then it will
download the original training data from the `original paper
<https://www.aclweb.org/anthology/D19-1654.pdf>`_ . Else
it will download the cleaned Training dataset version. The
cleaned version only contains a few sample differences
but these differences are with respect to overlapping
targets. See this `notebook for full differences
<https://github.com/apmoore1/target-extraction/blob/master/tutorials/Difference_between_MAMS_ATSA_original_and_MAMS_ATSA_cleaned.ipynb>`_:
:returns: The `train`, `val`, or `test` dataset from the
Multi-Aspect-Multi-Sentiment dataset (MAMS) ATSA version.
Dataset came from the `A Challenge Dataset and Effective Models
for Aspect-Based Sentiment Analysis, EMNLP 2019
<https://www.aclweb.org/anthology/D19-1654.pdf>`_
:raises ValueError: If the `dataset` value is not `train`, `val`, or `test`
'''
accepted_datasets = {'train', 'val', 'test'}
if dataset not in accepted_datasets:
raise ValueError('dataset has to be one of these values '
f'{accepted_datasets}, not {dataset}')
if cache_dir is None:
cache_dir = CACHE_DIRECTORY
data_folder = Path(cache_dir, 'Jiang 2019 MAMS ATSA')
data_folder.mkdir(parents=True, exist_ok=True)
dataset_url = {
'train':
'https://github.com/siat-nlp/MAMS-for-ABSA/raw/master/data/MAMS-ATSA/raw/train.xml',
'val':
'https://github.com/siat-nlp/MAMS-for-ABSA/raw/master/data/MAMS-ATSA/raw/val.xml',
'test':
'https://github.com/siat-nlp/MAMS-for-ABSA/raw/master/data/MAMS-ATSA/raw/test.xml'
}
url = dataset_url[dataset]
if dataset == 'train' and not original:
url = 'https://raw.githubusercontent.com/apmoore1/target-extraction/master/data/MAMS/MAMS_ATSA_cleaned_train.xml'
data_fp = Path(cached_path(url, cache_dir=data_folder))
# Parsing the data
target_text_collection = TargetTextCollection()
tree = ET.parse(data_fp)
sentences = tree.getroot()
for sentence_id, sentence in enumerate(sentences):
targets: List[str] = []
target_sentiments: List[Union[str, int]] = []
spans: List[Span] = []
for data in sentence:
if data.tag == 'text':
text = data.text
text = text.replace(u'\xa0', u' ')
elif data.tag == 'aspectTerms':
for target in data:
target_sentiment = target.attrib['polarity']
target_sentiments.append(target_sentiment)
targets.append(target.attrib['term'].replace(
u'\xa0', u' '))
span_from = int(target.attrib['from'])
span_to = int(target.attrib['to'])
spans.append(Span(span_from, span_to))
else:
raise ValueError(f'This tag {data.tag} should not occur '
'within a sentence tag')
target_text_kwargs = {
'targets': targets,
'spans': spans,
'text_id': f'{dataset}${str(sentence_id)}',
'target_sentiments': target_sentiments,
'categories': None,
'text': text,
'category_sentiments': None
}
for key in target_text_kwargs:
if not target_text_kwargs[key]:
target_text_kwargs[key] = None
target_text = TargetText(**target_text_kwargs)
target_text_collection.add(target_text)
return target_text_collection
def multi_aspect_multi_sentiment_acsa(
dataset: str,
cache_dir: Optional[Path] = None) -> TargetTextCollection:
'''
The data for this function when downloaded is stored within:
`Path(cache_dir, 'Jiang 2019 MAMS ACSA')
:NOTE: That as each sentence/`TargetText` object has to have
a `text_id`, as no ids exist in this dataset the ids are created
based on when the sentence occurs in the dataset e.g. the first
sentence/`TargetText` object id is '0'
For reference this dataset has 8 different aspect categories.
:param dataset: Either `train`, `val` or `test`, determines the dataset that
is returned.
:param cache_dir: The directory where all of the data is stored for
this code base. If None then the cache directory is
`dataset_parsers.CACHE_DIRECTORY`
:returns: The `train`, `val`, or `test` dataset from the
Multi-Aspect-Multi-Sentiment dataset (MAMS) ACSA version.
Dataset came from the `A Challenge Dataset and Effective Models
for Aspect-Based Sentiment Analysis, EMNLP 2019
<https://www.aclweb.org/anthology/D19-1654.pdf>`_
:raises ValueError: If the `dataset` value is not `train`, `val`, or `test`
'''
accepted_datasets = {'train', 'val', 'test'}
if dataset not in accepted_datasets:
raise ValueError('dataset has to be one of these values '
f'{accepted_datasets}, not {dataset}')
if cache_dir is None:
cache_dir = CACHE_DIRECTORY
data_folder = Path(cache_dir, 'Jiang 2019 MAMS ACSA')
data_folder.mkdir(parents=True, exist_ok=True)
dataset_url = {
'train':
'https://github.com/siat-nlp/MAMS-for-ABSA/raw/master/data/MAMS-ACSA/raw/train.xml',
'val':
'https://github.com/siat-nlp/MAMS-for-ABSA/raw/master/data/MAMS-ACSA/raw/val.xml',
'test':
'https://github.com/siat-nlp/MAMS-for-ABSA/raw/master/data/MAMS-ACSA/raw/test.xml'
}
url = dataset_url[dataset]
data_fp = Path(cached_path(url, cache_dir=data_folder))
# Parsing the data
category_text_collection = TargetTextCollection()
tree = ET.parse(data_fp)
sentences = tree.getroot()
for sentence_id, sentence in enumerate(sentences):
categories: List[str] = []
category_sentiments: List[Union[str, int]] = []
for data in sentence:
if data.tag == 'text':
text = data.text
text = text.replace(u'\xa0', u' ')
elif data.tag == 'aspectCategories':
for category in data:
category_sentiment = category.attrib['polarity']
category_sentiments.append(category_sentiment)
categories.append(category.attrib['category'].replace(
u'\xa0', u' '))
else:
raise ValueError(f'This tag {data.tag} should not occur '
'within a sentence tag')
category_text_kwargs = {
'targets': None,
'spans': None,
'text_id': f'{dataset}${str(sentence_id)}',
'target_sentiments': None,
'categories': categories,
'text': text,
'category_sentiments': category_sentiments
}
for key in category_text_kwargs:
if not category_text_kwargs[key]:
category_text_kwargs[key] = None
category_text = TargetText(**category_text_kwargs)
category_text_collection.add(category_text)
return category_text_collection
parser.add_argument("augmented_training_dataset", type=parse_path,
help='File path to the augmented training dataset')
parser.add_argument("expanded_targets_fp", type=parse_path,
help='File path to the expanded targets json file')
args = parser.parse_args()
with args.expanded_targets_fp.open('r') as expanded_targets_file:
targets_equivalents: Dict[str, str] = json.load(expanded_targets_file)
assert len(targets_equivalents) > 1
expanded_target_counts = Counter()
number_training_samples = 0
number_targets_expanded = 0
with args.augmented_training_dataset.open('r') as training_file:
for line in training_file:
training_sample = TargetText.from_json(line)
number_targets = len(training_sample['targets'])
number_training_samples += number_targets
for target_index in range(number_targets):
original_target = training_sample['targets'][target_index]
if original_target.lower() not in targets_equivalents:
continue
number_targets_expanded += 1
expanded_target_key = f'target {target_index}'
expanded_targets = training_sample[expanded_target_key]
assert original_target in expanded_targets
number_expanded_targets = len(expanded_targets) - 1
assert len(expanded_targets) == len(set(expanded_targets))
expanded_target_counts.update([number_expanded_targets])
def test_exact_match_score(self):
# Simple case where it should get perfect score
test_collection = TargetTextCollection([self._target_text_example()])
test_collection.tokenize(spacy_tokenizer())
test_collection.sequence_labels()
measures = test_collection.exact_match_score('sequence_labels')
for index, measure in enumerate(measures):
if index == 3:
assert measure['FP'] == []
assert measure['FN'] == []
assert measure['TP'] == [('2', Span(4, 15)),
('2', Span(30, 35))]
else:
assert measure == 1.0
# Something that has perfect precision but misses one therefore does
# not have perfect recall nor f1
test_collection = TargetTextCollection(
self._target_text_measure_examples())
test_collection.tokenize(str.split)
# text = 'The laptop case was great and cover was rubbish'
sequence_labels_0 = ['O', 'O', 'O', 'O', 'O', 'O', 'B', 'O', 'O']
test_collection['0']['sequence_labels'] = sequence_labels_0
# text = 'The laptop price was awful'
sequence_labels_1 = ['O', 'B', 'I', 'O', 'O']
test_collection['1']['sequence_labels'] = sequence_labels_1
recall, precision, f1, error_analysis = test_collection.exact_match_score(
'sequence_labels')
assert precision == 1.0
assert recall == 2.0 / 3.0
assert f1 == 0.8
assert error_analysis['FP'] == []
assert error_analysis['FN'] == [('0', Span(4, 15))]
assert error_analysis['TP'] == [('0', Span(30, 35)), ('1', Span(4,
16))]
# Something that has perfect recall but not precision as it over
# predicts
sequence_labels_0 = ['O', 'B', 'I', 'B', 'O', 'O', 'B', 'O', 'O']
test_collection['0']['sequence_labels'] = sequence_labels_0
sequence_labels_1 = ['O', 'B', 'I', 'O', 'O']
test_collection['1']['sequence_labels'] = sequence_labels_1
recall, precision, f1, error_analysis = test_collection.exact_match_score(
'sequence_labels')
assert precision == 3 / 4
assert recall == 1.0
assert round(f1, 3) == 0.857
assert error_analysis['FP'] == [('0', Span(16, 19))]
assert error_analysis['FN'] == []
assert error_analysis['TP'] == [('0', Span(4, 15)), ('0', Span(30,
35)),
('1', Span(4, 16))]
# Does not predict anything for a whole sentence therefore will have
# perfect precision but bad recall (mainly testing the if not
# getting anything for a sentence matters)
sequence_labels_0 = ['O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O']
test_collection['0']['sequence_labels'] = sequence_labels_0
sequence_labels_1 = ['O', 'B', 'I', 'O', 'O']
test_collection['1']['sequence_labels'] = sequence_labels_1
recall, precision, f1, error_analysis = test_collection.exact_match_score(
'sequence_labels')
assert precision == 1.0
assert recall == 1 / 3
assert f1 == 0.5
assert error_analysis['FP'] == []
fn_error = sorted(error_analysis['FN'], key=lambda x: x[1].start)
assert fn_error == [('0', Span(4, 15)), ('0', Span(30, 35))]
assert error_analysis['TP'] == [('1', Span(4, 16))]
# Handle the edge case of not getting anything
sequence_labels_0 = ['O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O']
test_collection['0']['sequence_labels'] = sequence_labels_0
sequence_labels_1 = ['O', 'O', 'O', 'O', 'O']
test_collection['1']['sequence_labels'] = sequence_labels_1
recall, precision, f1, error_analysis = test_collection.exact_match_score(
'sequence_labels')
assert precision == 0.0
assert recall == 0.0
assert f1 == 0.0
assert error_analysis['FP'] == []
fn_error = sorted(error_analysis['FN'], key=lambda x: x[1].start)
assert fn_error == [('0', Span(4, 15)), ('1', Span(4, 16)),
('0', Span(30, 35))]
assert error_analysis['TP'] == []
# The case where the tokens and the text do not align
not_align_example = self._target_text_not_align_example()
# text = 'The laptop case; was awful'
sequence_labels_align = ['O', 'B', 'I', 'O', 'O']
test_collection.add(not_align_example)
test_collection.tokenize(str.split)
test_collection['inf']['sequence_labels'] = sequence_labels_align
sequence_labels_0 = ['O', 'B', 'I', 'O', 'O', 'O', 'B', 'O', 'O']
test_collection['0']['sequence_labels'] = sequence_labels_0
sequence_labels_1 = ['O', 'B', 'I', 'O', 'O']
test_collection['1']['sequence_labels'] = sequence_labels_1
recall, precision, f1, error_analysis = test_collection.exact_match_score(
'sequence_labels')
assert recall == 3 / 4
assert precision == 3 / 4
assert f1 == 0.75
assert error_analysis['FP'] == [('inf', Span(4, 16))]
assert error_analysis['FN'] == [('inf', Span(4, 15))]
tp_error = sorted(error_analysis['TP'], key=lambda x: x[1].start)
assert tp_error == [('0', Span(4, 15)), ('1', Span(4, 16)),
('0', Span(30, 35))]
# This time it can get a perfect score as the token alignment will be
# perfect
test_collection.tokenize(spacy_tokenizer())
sequence_labels_align = ['O', 'B', 'I', 'O', 'O', 'O']
test_collection['inf']['sequence_labels'] = sequence_labels_align
recall, precision, f1, error_analysis = test_collection.exact_match_score(
'sequence_labels')
assert recall == 1.0
assert precision == 1.0
assert f1 == 1.0
assert error_analysis['FP'] == []
assert error_analysis['FN'] == []
tp_error = sorted(error_analysis['TP'], key=lambda x: x[1].end)
assert tp_error == [('0', Span(4, 15)), ('inf', Span(4, 15)),
('1', Span(4, 16)), ('0', Span(30, 35))]
# Handle the case where one of the samples has no spans
test_example = TargetText(text="I've had a bad day", text_id='50')
other_examples = self._target_text_measure_examples()
other_examples.append(test_example)
test_collection = TargetTextCollection(other_examples)
test_collection.tokenize(str.split)
test_collection.sequence_labels()
measures = test_collection.exact_match_score('sequence_labels')
for index, measure in enumerate(measures):
if index == 3:
assert measure['FP'] == []
assert measure['FN'] == []
tp_error = sorted(measure['TP'], key=lambda x: x[1].end)
assert tp_error == [('0', Span(4, 15)), ('1', Span(4, 16)),
('0', Span(30, 35))]
else:
assert measure == 1.0
# Handle the case where on the samples has no spans but has predicted
# there is a span there
test_collection['50']['sequence_labels'] = ['B', 'I', 'O', 'O', 'O']
recall, precision, f1, error_analysis = test_collection.exact_match_score(
'sequence_labels')
assert recall == 1.0
assert precision == 3 / 4
assert round(f1, 3) == 0.857
assert error_analysis['FP'] == [('50', Span(start=0, end=8))]
assert error_analysis['FN'] == []
tp_error = sorted(error_analysis['TP'], key=lambda x: x[1].end)
assert tp_error == [('0', Span(4, 15)), ('1', Span(4, 16)),
('0', Span(30, 35))]
# See if it can handle a collection that only contains no spans
test_example = TargetText(text="I've had a bad day", text_id='50')
test_collection = TargetTextCollection([test_example])
test_collection.tokenize(str.split)
test_collection.sequence_labels()
measures = test_collection.exact_match_score('sequence_labels')
for index, measure in enumerate(measures):
if index == 3:
assert measure['FP'] == []
assert measure['FN'] == []
assert measure['TP'] == []
else:
assert measure == 0.0
# Handle the case the collection contains one spans but a mistake
test_collection['50']['sequence_labels'] = ['B', 'I', 'O', 'O', 'O']
measures = test_collection.exact_match_score('sequence_labels')
for index, measure in enumerate(measures):
if index == 3:
assert measure['FP'] == [('50', Span(0, 8))]
assert measure['FN'] == []
assert measure['TP'] == []
else:
assert measure == 0.0
# Should raise a KeyError if one of the TargetText instances does
# not have a Span key
del test_collection['50']._storage['spans']
with pytest.raises(KeyError):
test_collection.exact_match_score('sequence_labels')
# should raise a KeyError if one of the TargetText instances does
# not have a predicted sequence key
test_collection = TargetTextCollection([self._target_text_example()])
test_collection.tokenize(spacy_tokenizer())
test_collection.sequence_labels()
with pytest.raises(KeyError):
measures = test_collection.exact_match_score('nothing')
# Should raise a ValueError if there are multiple same true spans
a = TargetText(text='hello how are you I am good',
text_id='1',
targets=['hello', 'hello'],
spans=[Span(0, 5), Span(0, 5)])
test_collection = TargetTextCollection([a])
test_collection.tokenize(str.split)
test_collection['1']['sequence_labels'] = [
'B', 'O', 'O', 'O', 'O', 'O', 'O'
]
with pytest.raises(ValueError):
test_collection.exact_match_score('sequence_labels')
def _semeval_extract_data(sentence_tree: Element,
conflict: bool) -> TargetTextCollection:
'''
:param sentence_tree: The root element of the XML tree that has come
from a SemEval XML formatted XML File.
:param conflict: Whether or not to include targets or categories that
have the `conflict` sentiment value. True is to include
conflict targets and categories.
:returns: The SemEval data formatted into a
`target_extraction.data_types.TargetTextCollection` object.
'''
target_text_collection = TargetTextCollection()
for sentence in sentence_tree:
text_id = sentence.attrib['id']
targets: List[str] = []
target_sentiments: List[Union[str, int]] = []
spans: List[Span] = []
category_sentiments: List[Union[str, int]] = []
categories: List[str] = []
for data in sentence:
if data.tag == 'text':
text = data.text
text = text.replace(u'\xa0', u' ')
elif data.tag == 'aspectTerms':
for target in data:
# If it is a conflict sentiment and conflict argument True
# skip this target
target_sentiment = target.attrib['polarity']
if not conflict and target_sentiment == 'conflict':
continue
targets.append(target.attrib['term'].replace(
u'\xa0', u' '))
target_sentiments.append(target_sentiment)
span_from = int(target.attrib['from'])
span_to = int(target.attrib['to'])
spans.append(Span(span_from, span_to))
elif data.tag == 'aspectCategories':
for category in data:
# If it is a conflict sentiment and conflict argument True
# skip this category
category_sentiment = category.attrib['polarity']
if not conflict and category_sentiment == 'conflict':
continue
categories.append(category.attrib['category'])
category_sentiments.append(category.attrib['polarity'])
elif data.tag == 'Opinions':
for opinion in data:
category_target_sentiment = opinion.attrib['polarity']
if not conflict and category_target_sentiment == 'conflict':
continue
# Handle the case where some of the SemEval 16 files do
# not contain targets and are only category sentiment files
if 'target' in opinion.attrib:
# Handle the case where there is a category but no
# target
target_text = opinion.attrib['target'].replace(
u'\xa0', u' ')
span_from = int(opinion.attrib['from'])
span_to = int(opinion.attrib['to'])
# Special cases for poor annotation in SemEval 2016
# task 5 subtask 1 Restaurant dataset
if text_id == 'DBG#2:15' and target_text == 'NULL':
span_from = 0
span_to = 0
if text_id == "en_Patsy'sPizzeria_478231878:2"\
and target_text == 'NULL':
span_to = 0
if text_id == "en_MercedesRestaurant_478010602:1" \
and target_text == 'NULL':
span_to = 0
if text_id == "en_MiopostoCaffe_479702043:9" \
and target_text == 'NULL':
span_to = 0
if text_id == "en_MercedesRestaurant_478010600:1" \
and target_text == 'NULL':
span_from = 0
span_to = 0
if target_text == 'NULL':
target_text = None
# Special cases for poor annotation in SemEval 2016
# task 5 subtask 1 Restaurant dataset
if text_id == '1490757:0':
target_text = 'restaurant'
if text_id == 'TR#1:0' and span_from == 27:
target_text = 'spot'
if text_id == 'TFS#5:26':
target_text = "environment"
if text_id == 'en_SchoonerOrLater_477965850:10':
target_text = 'Schooner or Later'
targets.append(target_text)
spans.append(Span(span_from, span_to))
categories.append(opinion.attrib['category'])
target_sentiments.append(category_target_sentiment)
target_text_kwargs = {
'targets': targets,
'spans': spans,
'text_id': text_id,
'target_sentiments': target_sentiments,
'categories': categories,
'text': text,
'category_sentiments': category_sentiments
}
for key in target_text_kwargs:
if not target_text_kwargs[key]:
target_text_kwargs[key] = None
target_text = TargetText(**target_text_kwargs)
target_text_collection.add(target_text)
return target_text_collection
def test_dataset_target_extraction_statistics(lower: bool,
incl_sentence_statistics: bool):
if lower is not None:
target_stats = dataset_target_extraction_statistics(
[TRAIN_COLLECTION],
lower_target=lower,
incl_sentence_statistics=incl_sentence_statistics)
else:
target_stats = dataset_target_extraction_statistics(
[TRAIN_COLLECTION],
incl_sentence_statistics=incl_sentence_statistics)
tl_1 = round((17 / 19.0) * 100, 2)
tl_2 = round((2 / 19.0) * 100, 2)
true_stats = {
'Name': 'train',
'No. Sentences': 6,
'No. Sentences(t)': 5,
'No. Targets': 19,
'No. Uniq Targets': 13,
'ATS': round(19 / 6.0, 2),
'ATS(t)': round(19 / 5.0, 2),
'TL 1 %': tl_1,
'TL 2 %': tl_2,
'TL 3+ %': 0.0,
'Mean Sentence Length': 15.33,
'Mean Sentence Length(t)': 16.6
}
if not incl_sentence_statistics:
del true_stats['Mean Sentence Length(t)']
del true_stats['Mean Sentence Length']
if lower == False:
true_stats['No. Uniq Targets'] = 14
assert 1 == len(target_stats)
target_stats = target_stats[0]
assert len(true_stats) == len(target_stats)
for stat_name, stat in true_stats.items():
if re.search(r'^TL', stat_name):
assert math.isclose(stat, target_stats[stat_name], rel_tol=0.001)
else:
assert stat == target_stats[stat_name], stat_name
# Multiple collections, where one collection is just the subset of the other
subcollection = TargetTextCollection(name='sub')
subcollection.add(TRAIN_COLLECTION["81207500773427072"])
subcollection.add(TRAIN_COLLECTION["78522643479064576"])
long_target = TargetText(
text='some text that contains a long target or two',
spans=[Span(0, 14), Span(15, 37)],
targets=['some text that', 'contains a long target'],
target_sentiments=['positive', 'negative'],
text_id='100')
subcollection.add(long_target)
subcollection.tokenize(whitespace())
if lower is not None:
target_stats = dataset_target_extraction_statistics(
[subcollection, TRAIN_COLLECTION],
lower_target=lower,
incl_sentence_statistics=incl_sentence_statistics)
else:
target_stats = dataset_target_extraction_statistics(
[subcollection, TRAIN_COLLECTION],
incl_sentence_statistics=incl_sentence_statistics)
tl_1 = round((6 / 9.0) * 100, 2)
tl_2 = round((1 / 9.0) * 100, 2)
tl_3 = round((2 / 9.0) * 100, 2)
sub_stats = {
'Name': 'sub',
'No. Sentences': 3,
'No. Sentences(t)': 3,
'No. Targets': 9,
'No. Uniq Targets': 9,
'ATS': round(9 / 3.0, 2),
'ATS(t)': round(9 / 3.0, 2),
'TL 1 %': tl_1,
'TL 2 %': tl_2,
'TL 3+ %': tl_3,
'Mean Sentence Length': 11.67,
'Mean Sentence Length(t)': 11.67
}
if not incl_sentence_statistics:
del sub_stats['Mean Sentence Length(t)']
del sub_stats['Mean Sentence Length']
true_stats = [sub_stats, true_stats]
assert len(true_stats) == len(target_stats)
for stat_index, stat in enumerate(true_stats):
test_stat = target_stats[stat_index]
assert len(stat) == len(test_stat)
for stat_name, stat_value in stat.items():
if re.search(r'^TL', stat_name):
assert math.isclose(stat_value,
test_stat[stat_name],
rel_tol=0.001)
else:
assert stat_value == test_stat[stat_name], stat_name
train_targets = set(list(train_data.target_count(lower=True).keys()))
acceptable_confidence = args.confidence_score
all_targets: List[TargetText] = []
with args.predicted_target_data_fp.open('r') as predicted_file:
for index, line in enumerate(predicted_file):
target_data = json.loads(line)
target_id = str(index)
target_data_dict = {
'text': target_data['text'],
'text_id': target_id,
'confidences': target_data['confidence'],
'sequence_labels': target_data['sequence_labels'],
'tokenized_text': target_data['tokens']
}
target_data = TargetText.target_text_from_prediction(
**target_data_dict, confidence=acceptable_confidence)
if target_data['targets']:
all_targets.append(target_data)
print(len(all_targets))
all_targets: TargetTextCollection = TargetTextCollection(all_targets)
pred_target_dict = all_targets.target_count(lower=True)
pred_targets = set(list(pred_target_dict.keys()))
print(f'Number of unique targets in training dataset {len(train_targets)}')
print(f'Number of unique predicted targets {len(pred_targets)}')
pred_difference = pred_targets.difference(train_targets)
print(
f'Number of unique targets in predicted but not in training {len(pred_difference)}'
)
train_difference = train_targets.difference(pred_targets)
print(
def text_to_instance(self, text: str,
targets: Optional[List[str]] = None,
target_sentiments: Optional[List[Union[str, int]]] = None,
spans: Optional[List[List[int]]] = None,
categories: Optional[List[str]] = None,
category_sentiments: Optional[List[Union[str, int]]] = None,
**kwargs) -> Instance:
'''
The original text, text tokens as well as the targets and target
tokens are stored in the MetadataField.
:NOTE: At least targets and/or categories must be present.
:NOTE: That the left and right contexts returned in the instance are
a List of a List of tokens. A list for each Target.
:param text: The text that contains the target(s) and/or categories.
:param targets: The targets that are within the text
:param target_sentiments: The sentiment of the targets. To be used if
training the classifier
:param spans: The spans that represent the character offsets for each
of the targets given in the targets list.
:param categories: The categories that are within the text
:param category_sentiments: The sentiment of the categories
:returns: An Instance object with all of the above encoded for a
PyTorch model.
:raises ValueError: If either targets and categories are both None
:raises ValueError: If `self._target_sequences` is True and the passed
`spans` argument is None.
:raises ValueError: If `self._left_right_contexts` is True and the
passed `spans` argument is None.
'''
if targets is None and categories is None:
raise ValueError('Either targets or categories must be given if you '
'want to be predict the sentiment of a target '
'or a category')
instance_fields: Dict[str, Field] = {}
# Metadata field
metadata_dict = {}
if targets is not None:
# need to change this so that it takes into account the case where
# the positions are True but not the target sequences.
if self._target_sequences or self._position_embeddings or self._position_weights:
if spans is None:
raise ValueError('To create target sequences requires `spans`')
spans = [Span(span[0], span[1]) for span in spans]
target_text_object = TargetText(text=text, spans=spans,
targets=targets, text_id='anything')
target_text_object.force_targets()
text = target_text_object['text']
allen_tokens = self._tokenizer.tokenize(text)
tokens = [x.text for x in allen_tokens]
target_text_object['tokenized_text'] = tokens
target_text_object.sequence_labels(per_target=True)
target_sequences = target_text_object['sequence_labels']
# Need to add the target sequences to the instances
in_label = {'B', 'I'}
number_targets = len(targets)
all_target_tokens: List[List[Token]] = [[] for _ in range(number_targets)]
target_sequence_fields = []
target_indicators: List[List[int]] = []
for target_index in range(number_targets):
one_values = []
target_ones = [0] * len(allen_tokens)
for token_index, token in enumerate(allen_tokens):
target_sequence_value = target_sequences[target_index][token_index]
in_target = 1 if target_sequence_value in in_label else 0
if in_target:
all_target_tokens[target_index].append(allen_tokens[token_index])
one_value_list = [0] * len(allen_tokens)
one_value_list[token_index] = 1
one_values.append(one_value_list)
target_ones[token_index] = 1
one_values = np.array(one_values)
target_sequence_fields.append(ArrayField(one_values, dtype=np.int32))
target_indicators.append(target_ones)
if self._position_embeddings:
target_distances = self._target_indicators_to_distances(target_indicators,
max_distance=self._max_position_distance,
as_string=True)
target_text_distances = []
for target_distance in target_distances:
token_distances = [Token(distance) for distance in target_distance]
token_distances = TextField(token_distances, self._position_indexers)
target_text_distances.append(token_distances)
instance_fields['position_embeddings'] = ListField(target_text_distances)
if self._position_weights:
target_distances = self._target_indicators_to_distances(target_indicators,
max_distance=self._max_position_distance,
as_string=False)
target_distances = np.array(target_distances)
instance_fields['position_weights'] = ArrayField(target_distances,
dtype=np.int32)
if self._target_sequences:
instance_fields['target_sequences'] = ListField(target_sequence_fields)
instance_fields['tokens'] = TextField(allen_tokens, self._token_indexers)
metadata_dict['text words'] = tokens
metadata_dict['text'] = text
# update target variable as the targets could have changed due
# to the force_targets function
targets = target_text_object['targets']
else:
all_target_tokens = [self._tokenizer.tokenize(target)
for target in targets]
target_fields = [TextField(target_tokens, self._token_indexers)
for target_tokens in all_target_tokens]
target_fields = ListField(target_fields)
instance_fields['targets'] = target_fields
# Add the targets and the tokenised targets to the metadata
metadata_dict['targets'] = [target for target in targets]
metadata_dict['target words'] = [[x.text for x in target_tokens]
for target_tokens in all_target_tokens]
# Target sentiment if it exists
if target_sentiments is not None:
target_sentiments_field = SequenceLabelField(target_sentiments,
target_fields,
label_namespace='target-sentiment-labels')
instance_fields['target_sentiments'] = target_sentiments_field
if categories is not None and self._use_categories:
category_fields = TextField([Token(category) for category in categories],
self._token_indexers)
instance_fields['categories'] = category_fields
# Category sentiment if it exists
if category_sentiments is not None:
category_sentiments_field = SequenceLabelField(category_sentiments,
category_fields,
label_namespace='category-sentiment-labels')
instance_fields['category_sentiments'] = category_sentiments_field
# Add the categories to the metadata
metadata_dict['categories'] = [category for category in categories]
if 'tokens' not in instance_fields:
tokens = self._tokenizer.tokenize(text)
instance_fields['tokens'] = TextField(tokens, self._token_indexers)
metadata_dict['text'] = text
metadata_dict['text words'] = [x.text for x in tokens]
# If required processes the left and right contexts
left_contexts = None
right_contexts = None
if self._left_right_contexts:
if spans is None:
raise ValueError('To create left, right, target contexts requires'
' the `spans` of the targets which is None')
spans = [Span(span[0], span[1]) for span in spans]
target_text_object = TargetText(text=text, spans=spans,
targets=targets, text_id='anything')
# left, right, and target contexts for each target in the
# the text
left_right_targets = target_text_object.left_right_target_contexts(incl_target=self._incl_target)
left_contexts: List[str] = []
right_contexts: List[str] = []
for left_right_target in left_right_targets:
left, right, _ = left_right_target
left_contexts.append(left)
if self._reverse_right_context:
right_tokens = self._tokenizer.tokenize(right)
reversed_right_tokens = []
for token in reversed(right_tokens):
reversed_right_tokens.append(token.text)
right = ' '.join(reversed_right_tokens)
right_contexts.append(right)
if left_contexts is not None:
left_field = self._add_context_field(left_contexts)
instance_fields["left_contexts"] = left_field
if right_contexts is not None:
right_field = self._add_context_field(right_contexts)
instance_fields["right_contexts"] = right_field
instance_fields["metadata"] = MetadataField(metadata_dict)
return Instance(instance_fields)