def create_examples(self, datas, set_type):
examples = []
for i, data in enumerate(datas):
guid = data[0]
text_a = data[1].strip()
# text_b = data[3].strip()
if set_type == 'test':
label = None
else:
label = str(int(data[2]))
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=None,
label=label))
return examples
def similarity_with_concepts(self, text, concepts):
examples = [InputExample(guid='0', text_a=text, text_b=concept) for concept in concepts]
features = glue_convert_examples_to_features(examples=examples,
tokenizer=tokenizer,
max_length=128,
output_mode='regression',
label_list=[None])
input_ids = torch.tensor([feature.input_ids for feature in features])
attention_mask = torch.tensor([feature.attention_mask for feature in features])
with torch.no_grad():
outputs = model(input_ids=input_ids, attention_mask=attention_mask)
outputs = outputs[0].T.tolist()[0]
return outputs
def _create_examples(self, lines, set_type):
"""Creates examples for the training, dev and test sets."""
examples = []
text_index = 0
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, i)
text_a = line[text_index]
label = line[1]
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=None,
label=label))
return examples
def _create_examples_adv(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
# if i == 0:
# continue
guid = "%s-%s" % (set_type, i)
text_a = line
label = "0" ## label here doesn't matter
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=None,
label=label))
return examples
def _create_examples(self, lines_in, set_type):
"""Creates examples for the training, dev and test sets."""
examples = []
for i, line in enumerate(lines_in[1:]):
guid = "%s-%s" % (set_type, i)
text_a = line[1]
text_b = ''
label = None if set_type == "predict" else line[0]
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=text_b,
label=label))
return examples
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, line[0])
text_a = line[7]
text_b = line[8]
label = line[-1]
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=text_b,
label=label))
return examples
def similarity_with_concept(self, text, concept):
example = InputExample(guid='0', text_a=text, text_b=concept)
feature = glue_convert_examples_to_features(examples=[example],
tokenizer=self.tokenizer,
max_length=128,
output_mode='regression',
label_list=[None])
input_ids = torch.tensor(feature[0].input_ids).unsqueeze(0)
attention_mask = torch.tensor(feature[0].attention_mask).unsqueeze(0)
with torch.no_grad():
outputs = self.model(input_ids=input_ids,
attention_mask=attention_mask)
return outputs[0].item()
def load_pr_curve_data(world, split):
prcp_pairs, tags = world.GetDPRCP(test=(split == 'test'))
examples = []
for i, (qid1, qid2) in enumerate(prcp_pairs):
q1 = world.id_to_text[qid1]
q2 = world.id_to_text[qid2]
# One QQP question is '' which screws up RoBERTa tokenizer. Change it to ' '
if not q1:
q1 = ' '
if not q2:
q2 = ' '
examples.append(
InputExample(guid='{}-{}-pr-{}-{}'.format(split, i, qid1, qid2),
text_a=q1,
text_b=q2,
label='0'))
return examples, tags
def get_train_examples(self, data_dir):
"""See base class."""
datafile = os.path.join(data_dir, 'doordash_categorized.pkl')
print(datafile)
with open(datafile, 'rb') as f:
json = pickle.load(f)
examples = []
for entry in json:
guid = entry['key']
title = entry['title']
label = entry['labels']['category']
if label in self.excluded_labels:
continue
examples.append(
InputExample(guid=guid, text_a=title, label=label)
)
return examples
def _create_examples(self, lines, set_type):
examples = []
for (_, line) in enumerate(lines):
guid = "%s-%s" % (set_type, line[0])
try:
text_a = line[1].lower()
text_b = line[2].lower()
label = line[3]
except IndexError:
print('cannot read the line: ' + line)
continue
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=text_b,
label=label))
return examples
def create_examples(filename):
datas = pd.read_csv(filename).values.tolist()
examples = []
for i, data in enumerate(datas):
guid = data[1]
text_a = data[2].strip()
# print(text_a)
# text_b = data[2].strip()
examples.append(
InputExample(
guid=guid,
text_a=text_a,
text_b=None,
label=None
)
)
return examples
def _create_examples(self, lines: List[List[str]],
set_type: str) -> List[InputExample]:
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, i)
text_a = line[5]
text_b = line[6]
label = self._preprocess_label(line[0])
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=text_b,
label=label))
return examples
def _get_test_or_dev_helper(self, data_dir, stage):
if stage == "dev":
language = self.dev_language
elif stage == "test":
language = self.test_language
rows = self._read_xnli_tsv(
os.path.join(data_dir, "XNLI-1.0", f"xnli.{stage}.tsv"))
examples = [
InputExample(
guid=f"{stage}-{i}",
text_a=row["sentence1"],
text_b=row["sentence2"],
label=row["gold_label"],
) for i, row in enumerate(rows) if row["language"] == language
]
return examples
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
guid = "%s-%s" % (set_type, i)
text_a = line[0]
text_b = line[1]
if "imho" in text_a or "imo" in text_a: #Gengyu: shoud I remove this two words from data
label = "1"
else:
label = "0"
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=text_b,
label=label))
return examples
def _create_examples(self, lines, set_type):
"""Creates examples for the training, dev and test sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, line[0])
text_a = line[1]
text_b = line[2]
label = self.get_labels()[0] if set_type == 'test' else line[-1]
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=text_b,
label=label))
return examples
def _create_examples(self, lines, set_type):
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, i)
if set_type == 'test':
text_a = line[1]
label = '-1'
else:
label = line[0]
text_a = line[1]
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=None,
label=label))
return examples
def get_train_examples(self, data_dir):
rows = self._read_xnli_tsv(
os.path.join(
data_dir,
"XNLI-MT-1.0",
"multinli",
f"multinli.train.{self.train_language}.tsv",
))
examples = [
InputExample(
guid=f"train-{i}",
text_a=row["premise"],
text_b=row["hypo"],
label=("contradiction"
if row["label"] == "contradictory" else row["label"]),
) for i, row in enumerate(rows)
]
return examples
def _create_examples(self, lines, set_type):
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, line[0])
text_a = line[1]
text_b = line[2]
if set_type == "test":
label = -1
else:
label = line[0]
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=text_b,
label=label))
return examples
def _create_examples(self, lines: List[List[str]],
set_type: str) -> List[InputExample]:
"""Creates examples for the training, dev and test sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, line[0])
text_a = line[8]
text_b = line[9]
label = None if set_type.startswith(
"test") else self._preprocess_label(line[-1])
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=text_b,
label=label))
return examples
def _create_examples(self, lines, set_type):
"""Creates examples for the training, dev and test sets."""
test_mode = set_type == "test"
q1_index = 1 if test_mode else 3
q2_index = 2 if test_mode else 4
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, line[0])
try:
text_a = line[q1_index]
text_b = line[q2_index]
label = None if test_mode else line[5]
except IndexError:
continue
examples.append(InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
def _create_examples(self, set_type):
"""Creates examples for the training, dev and test sets."""
examples = []
label_map = {
0: "entailment",
1: "neutral",
2: "contradiction"
}
from nlp import load_dataset
dataset = load_dataset('anli')
for (i, data) in enumerate(dataset[set_type]):
guid = "%s-%s" % (set_type, data['uid'])
text_a = data['premise']
text_b = data['hypothesis']
# label = None if set_type.startswith("test") else label_map[data['label']]
label = label_map[data['label']]
examples.append(InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
def _rows2examples(self, rows):
# Spanish sentence tokenizer
tokenizer = nltk.data.load("tokenizers/punkt/PY3/spanish.pickle")
examples = []
logger.info("Reading examples")
for i, row in enumerate(tqdm(rows)):
# the text column was saved as a string with the python syntax
# for bytes literals, so it must be converted to a string literal
tokens = tokenizer.tokenize(eval(row[1]).decode())
example = InputExample(
f"test-{i}",
tokens[0],
tokens[1] if len(tokens) > 1 else None,
label=row[0],
)
examples.append(example)
return examples
def _create_test_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples, rows = [], []
for (_, line) in enumerate(lines):
guid = "%s-%s" % (set_type, line[0])
try:
text_a = line[1].lower()
text_b = line[2].lower()
label = line[3]
except IndexError:
print('cannot read the line: ' + line)
continue
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=text_b,
label=label))
rows.append(line)
return examples, rows
def _load_data(self, dir, set_type):
examples = []
for filename in os.listdir(dir):
if filename.endswith(".ann"):
lines = self._read_tsv(os.path.join(dir, filename))
#print(os.path.join(directory, filename))
for (i, line) in enumerate(lines):
#line = line.split("\t")
if line[0].startswith("T"):
guid = "%s-%s" % (set_type, len(examples))
text_a = line[2].replace("CMV:", "").strip()
if line[0].startswith("A"):
label = line[1].split(" ")[2].strip()
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=None,
label=label))
return examples
def _read_tfds_and_create_examples(self, mode):
if not is_tfds_available():
raise RuntimeError(
"The package tensorflow_datasets can't be imported")
import tensorflow_datasets as tfds
try:
ds, dsinfo = tfds.load(self.dataset_name,
split=mode,
with_info=True)
except KeyError:
raise ValueError(
"The dataset {} does not exists in the tensorflow_datasets catalog."
.format(self.dataset_name))
seen_labels = set()
for ex_index, entry in enumerate(ds):
if ex_index % 10000 == 0:
logger.info("Creating example %d", ex_index)
if self.guid in list(dsinfo.features.keys()):
guid = entry[self.guid].numpy()
else:
guid = id
if self.text_b in list(dsinfo.features.keys()):
text_b = entry[self.text_b].numpy().decode("utf-8")
else:
text_b = None
label = dsinfo.features[self.label].int2str(
entry[self.label].numpy())
seen_labels.add(label)
example = InputExample(guid,
entry[self.text_a].numpy().decode("utf-8"),
text_b, label)
self.examples[mode].append(example)
self.labels = list(set(self.labels).union(seen_labels))
def get_dev_examples(self, data_dir):
"""See base class."""
examples = []
with open(os.path.join(data_dir, "dev.jsonl"), "r") as f:
for i, line in enumerate(f.readlines()):
json_content = json.loads(line.replace("\n", ""))
guid = "%s-%s" % (i, "dev")
text_a = json_content['q1']
text_b = json_content['q2']
label = json_content['label']
assert label == '1' or label == '0'
label = "paraphrase" if label == "1" else "not-paraphrase"
assert isinstance(text_a, str), f"Training input {text_a} is not a string"
assert isinstance(text_b, str), f"Training input {text_b} is not a string"
assert isinstance(label, str), f"Training label {label} is not a string"
example = InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label)
if i < 10:
print(example)
examples.append(example)
return examples
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
# if i == 0:
# continue
guid = "%s-%s" % (set_type, i)
if len(line) < 2:
continue
text_a = line[0]
label = line[1]
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=None,
label=label))
# if set_type == "train":
# random.shuffle(examples)
# print ("Examples are shuffled")
return examples
def _create_examples(self, data, set_type):
examples = []
raw_texts = data.tweet.values.tolist()
raw_labels = data.label.values.tolist()
for i in range(0, len(raw_texts)):
guid = "%s-%s" % (set_type, i)
raw_text = raw_texts[i]
raw_label = raw_labels[i]
label = raw_label
text = self._preprocess_text(raw_text)
examples.append(
InputExample(guid=guid,
text_a=text,
text_b=None,
label=label))
return examples
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
# guid = "%s-%s" % (set_type, i)
guid = line[0]
if set_type == 'test':
# text_a = line[1]
text_a = line[2]
label = None
else:
text_a = line[2]
label = line[1]
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=None,
label=label))
return examples
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
try:
if i == 0:
continue
guid = "%s-%s" % (set_type, line[0])
text_a = line[1]
text_b = None
label = line[0]
examples.append(
InputExample(guid=guid,
text_a=text_a,
text_b=text_b,
label=label))
except Exception as e:
import pdb
pdb.set_trace()
print(e)
return examples
