def __repr__(self):
s = ""
s += "qas_id: %s" % (printable_text(self.qas_id))
s += ", question_text: %s" % (printable_text(self.question_text))
s += ", doc_tokens: [%s]" % (" ".join(self.doc_tokens))
s += ", answerable: [%s]" % (self.answerable)
return s
def __repr__(self):
s = ""
s += "qas_id: %s" % (printable_text(self.qas_id))
s += ", question_text: %s" % (printable_text(self.question_text))
s += ", doc_tokens: [%s]" % (" ".join(self.doc_tokens))
if self.start_position:
s += ", start_position: %d" % (self.start_position)
if self.start_position:
s += ", end_position: %d" % (self.end_position)
return s
def convert_examples_to_features(examples, label_list, max_seq_length,
tokenizer):
"""Loads a data file into a list of `InputBatch`s."""
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
features = []
for (ex_index, example) in enumerate(examples):
tokens_a = tokenizer.tokenize(example.text_a)
tokens_b = None
if example.text_b:
tokens_b = tokenizer.tokenize(example.text_b)
if tokens_b:
# Modifies `tokens_a` and `tokens_b` in place so that the total
# length is less than the specified length.
# Account for [CLS], [SEP], [SEP] with "- 3"
_truncate_seq_pair(tokens_a, tokens_b, max_seq_length - 3)
else:
# Account for [CLS] and [SEP] with "- 2"
if len(tokens_a) > max_seq_length - 2:
tokens_a = tokens_a[0:(max_seq_length - 2)]
# The convention in BERT is:
# (a) For sequence pairs:
# tokens: [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]
# type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1
# (b) For single sequences:
# tokens: [CLS] the dog is hairy . [SEP]
# type_ids: 0 0 0 0 0 0 0
#
# Where "type_ids" are used to indicate whether this is the first
# sequence or the second sequence. The embedding vectors for `type=0` and
# `type=1` were learned during pre-training and are added to the wordpiece
# embedding vector (and position vector). This is not *strictly* necessary
# since the [SEP] token unambigiously separates the sequences, but it makes
# it easier for the model to learn the concept of sequences.
#
# For classification tasks, the first vector (corresponding to [CLS]) is
# used as as the "sentence vector". Note that this only makes sense because
# the entire model is fine-tuned.
tokens = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for token in tokens_a:
tokens.append(token)
segment_ids.append(0)
tokens.append("[SEP]")
segment_ids.append(0)
if tokens_b:
for token in tokens_b:
tokens.append(token)
segment_ids.append(1)
tokens.append("[SEP]")
segment_ids.append(1)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
label_id = label_map[example.label]
if ex_index < 5:
logger.info("*** Example ***")
logger.info("guid: %s" % (example.guid))
logger.info("tokens: %s" %
" ".join([printable_text(x) for x in tokens]))
logger.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
logger.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
logger.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
logger.info("label: %s (id = %d)" % (example.label, label_id))
features.append(
InputFeatures(input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id))
return features
def convert_examples_to_features(examples, tokenizer, max_seq_length,
doc_stride, max_query_length, is_training):
"""Loads a data file into a list of `InputBatch`s."""
unique_id = 1000000000
features = []
for (example_index, example) in enumerate(examples):
query_tokens = tokenizer.tokenize(example.question_text)
if len(query_tokens) > max_query_length:
query_tokens = query_tokens[0:max_query_length]
tok_to_orig_index = []
orig_to_tok_index = []
all_doc_tokens = []
for (i, token) in enumerate(example.doc_tokens):
orig_to_tok_index.append(len(all_doc_tokens))
sub_tokens = tokenizer.tokenize(token)
for sub_token in sub_tokens:
tok_to_orig_index.append(i)
all_doc_tokens.append(sub_token)
tok_start_position = None
tok_end_position = None
if is_training:
tok_start_position = orig_to_tok_index[example.start_position]
if example.end_position < len(example.doc_tokens) - 1:
tok_end_position = orig_to_tok_index[example.end_position +
1] - 1
else:
tok_end_position = len(all_doc_tokens) - 1
(tok_start_position, tok_end_position) = _improve_answer_span(
all_doc_tokens, tok_start_position, tok_end_position,
tokenizer, example.orig_answer_text)
# The -3 accounts for [CLS], [SEP] and [SEP]
max_tokens_for_doc = max_seq_length - len(query_tokens) - 3
# We can have documents that are longer than the maximum sequence length.
# To deal with this we do a sliding window approach, where we take chunks
# of the up to our max length with a stride of `doc_stride`.
_DocSpan = collections.namedtuple( # pylint: disable=invalid-name
"DocSpan", ["start", "length"])
doc_spans = []
start_offset = 0
while start_offset < len(all_doc_tokens):
length = len(all_doc_tokens) - start_offset
if length > max_tokens_for_doc:
length = max_tokens_for_doc
doc_spans.append(_DocSpan(start=start_offset, length=length))
if start_offset + length == len(all_doc_tokens):
break
start_offset += min(length, doc_stride)
for (doc_span_index, doc_span) in enumerate(doc_spans):
tokens = []
token_to_orig_map = {}
token_is_max_context = {}
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for token in query_tokens:
tokens.append(token)
segment_ids.append(0)
tokens.append("[SEP]")
segment_ids.append(0)
for i in range(doc_span.length):
split_token_index = doc_span.start + i
token_to_orig_map[len(
tokens)] = tok_to_orig_index[split_token_index]
is_max_context = _check_is_max_context(doc_spans,
doc_span_index,
split_token_index)
token_is_max_context[len(tokens)] = is_max_context
tokens.append(all_doc_tokens[split_token_index])
segment_ids.append(1)
tokens.append("[SEP]")
segment_ids.append(1)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
start_position = None
end_position = None
if is_training:
# For training, if our document chunk does not contain an annotation
# we throw it out, since there is nothing to predict.
doc_start = doc_span.start
doc_end = doc_span.start + doc_span.length - 1
if (example.start_position < doc_start
or example.end_position < doc_start
or example.start_position > doc_end
or example.end_position > doc_end):
continue
doc_offset = len(query_tokens) + 2
start_position = tok_start_position - doc_start + doc_offset
end_position = tok_end_position - doc_start + doc_offset
if example_index < 20:
logger.info("*** Example ***")
logger.info("unique_id: %s" % (unique_id))
logger.info("example_index: %s" % (example_index))
logger.info("doc_span_index: %s" % (doc_span_index))
logger.info("tokens: %s" %
" ".join([printable_text(x) for x in tokens]))
logger.info("token_to_orig_map: %s" % " ".join(
["%d:%d" % (x, y)
for (x, y) in token_to_orig_map.items()]))
logger.info("token_is_max_context: %s" % " ".join([
"%d:%s" % (x, y)
for (x, y) in token_is_max_context.items()
]))
logger.info("input_ids: %s" %
" ".join([str(x) for x in input_ids]))
logger.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
logger.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
if is_training:
answer_text = " ".join(
tokens[start_position:(end_position + 1)])
logger.info("start_position: %d" % (start_position))
logger.info("end_position: %d" % (end_position))
logger.info("answer: %s" % (printable_text(answer_text)))
features.append(
InputFeatures(unique_id=unique_id,
example_index=example_index,
doc_span_index=doc_span_index,
tokens=tokens,
token_to_orig_map=token_to_orig_map,
token_is_max_context=token_is_max_context,
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
start_position=start_position,
end_position=end_position))
unique_id += 1
return features
def convert_examples_to_features_mc(examples, label_list, max_seq_length,
tokenizer):
"""Loads a data file into a list of `InputBatch`s."""
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
features = []
for (ex_index, example) in tqdm(enumerate(examples),
desc="Converting examples"):
inputs = []
tokens_a = [tokenizer.tokenize(t) for t in example.text_a]
inputs.append(tokens_a)
tokens_b = None
if example.text_b:
tokens_b = [tokenizer.tokenize(t) for t in example.text_b]
inputs.append(tokens_b)
tokens_c = None
if example.text_c:
tokens_c = [tokenizer.tokenize(t) for t in example.text_c]
inputs.append(tokens_c)
tokens_d = None
if example.text_d:
tokens_d = [tokenizer.tokenize(t) for t in example.text_d]
inputs.append(tokens_d)
if len(inputs) > 1:
# Modifies `tokens_a` and `tokens_b` in place so that the total
# length is less than the specified length.
# Account for [CLS], [SEP], [SEP] with "- 3"
adjusted_len = max_seq_length - len(inputs) - 1
_truncate_sequences(adjusted_len, inputs)
else:
# Account for [CLS] and [SEP] with "- 2"
for idx, ta in enumerate(tokens_a):
tokens_a[idx] = tokens_a[idx][0:(max_seq_length - 2)]
all_tokens = []
all_token_ids = []
all_segments = []
all_masks = []
for zipped_tokens in zip(*inputs):
tokens = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for idx, field in enumerate(zipped_tokens):
for token in field:
tokens.append(token)
segment_ids.append(idx)
tokens.append("[SEP]")
segment_ids.append(idx)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
if len(input_ids) != max_seq_length:
print("FOUND")
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
all_tokens.append(tokens)
all_token_ids.append(input_ids)
all_segments.append(segment_ids)
all_masks.append(input_mask)
label_id = label_map[example.label]
if ex_index < 5:
logger.info("\n\n")
logger.info("*** Example {} ***\n".format(ex_index))
logger.info("guid: %s" % (example.guid))
_ts = all_tokens
_ids = all_token_ids
_masks = all_masks
_segs = all_segments
logger.info("\n")
for idx, (_t, _id, _mask,
_seg) in enumerate(zip(_ts, _ids, _masks, _segs)):
logger.info("\tOption {}".format(idx))
logger.info("\ttokens: %s" %
" ".join([printable_text(x) for x in _t]))
logger.info("\tinput_ids: %s" % " ".join([str(x)
for x in _id]))
logger.info("\tinput_mask: %s" %
" ".join([str(x) for x in _mask]))
logger.info("\tsegment_ids: %s" %
" ".join([str(x) for x in _seg]))
logger.info("label: %s (id = %d)" % (example.label, label_id))
features.append(
InputFeatures(input_ids=all_token_ids,
input_mask=all_masks,
segment_ids=all_segments,
label_id=label_id))
return features
def convert_examples_to_features(examples, tokenizer, max_seq_length,
doc_stride, max_query_length, is_training):
unique_id = 1000000000
long_input_features = []
for (example_index, example) in enumerate(tqdm(examples, desc="Example")):
input_features = []
query_tokens = tokenizer.tokenize(example.question_text)
if len(query_tokens) > max_query_length:
query_tokens = query_tokens[0:max_query_length]
all_doc_tokens = []
for (i, token) in enumerate(example.doc_tokens):
sub_tokens = tokenizer.tokenize(token)
for sub_token in sub_tokens:
all_doc_tokens.append(sub_token)
# The -3 accounts for [CLS], [SEP] and [SEP]
max_tokens_for_doc = max_seq_length - len(query_tokens) - 3
# We can have documents that are longer than the maximum sequence length.
# To deal with this we do a sliding window approach, where we take chunks
# of the up to our max length with a stride of `doc_stride`.
_DocSpan = collections.namedtuple( # pylint: disable=invalid-name
"DocSpan", ["start", "length"])
doc_spans = []
start_offset = 0
while start_offset < len(all_doc_tokens):
length = len(all_doc_tokens) - start_offset
if length > max_tokens_for_doc:
length = max_tokens_for_doc
doc_spans.append(_DocSpan(start=start_offset, length=length))
if start_offset + length == len(all_doc_tokens):
break
start_offset += min(length, doc_stride)
for (doc_span_index, doc_span) in enumerate(doc_spans):
tokens = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for token in query_tokens:
tokens.append(token)
segment_ids.append(0)
tokens.append("[SEP]")
segment_ids.append(0)
for i in range(doc_span.length):
split_token_index = doc_span.start + i
tokens.append(all_doc_tokens[split_token_index])
segment_ids.append(1)
tokens.append("[SEP]")
segment_ids.append(1)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
if example_index < 20:
logger.info("*** Example ***")
logger.info("unique_id: %s" % (unique_id))
logger.info("example_index: %s" % (example_index))
logger.info("doc_span_index: %s" % (doc_span_index))
logger.info("tokens: %s" %
" ".join([printable_text(x) for x in tokens]))
logger.info("input_ids: %s" %
" ".join([str(x) for x in input_ids]))
logger.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
logger.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
input_features.append(
InputFeature(input_ids=input_ids,
token_type_ids=segment_ids,
attention_mask=input_mask))
long_input_features.append(
LongInputFeature(
example.qas_id, input_features,
int(example.answerable)
if example.answerable is not None else None))
unique_id += 1
return long_input_features
def convert_examples_to_features(examples, label_list, max_seq_length,
tokenizer):
"""Loads a data file into a list of `InputBatch`s."""
print("#examples", len(examples))
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
features = [[]]
for (ex_index, example) in enumerate(examples):
tokens_a = tokenizer.tokenize(example.text_a)
tokens_b = None
tokens_c = None
if example.text_b:
tokens_b = tokenizer.tokenize(example.text_b)
if example.text_c:
tokens_c = tokenizer.tokenize(example.text_c)
if tokens_c:
_truncate_seq_tuple(tokens_a, tokens_b, tokens_c,
max_seq_length - 4)
tokens_b = tokens_c + ["[SEP]"] + tokens_b
elif tokens_b:
_truncate_seq_pair(tokens_a, tokens_b, max_seq_length - 3)
else:
if len(tokens_a) > max_seq_length - 2:
tokens_a = tokens_a[0:(max_seq_length - 2)]
tokens = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for token in tokens_a:
tokens.append(token)
segment_ids.append(0)
tokens.append("[SEP]")
segment_ids.append(0)
if tokens_b:
for token in tokens_b:
tokens.append(token)
segment_ids.append(1)
tokens.append("[SEP]")
segment_ids.append(1)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
label_id = label_map[example.label]
if ex_index < 5:
logger.info("*** Example ***")
logger.info("guid: %s" % (example.guid))
logger.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
logger.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
logger.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
logger.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
logger.info("label: %s (id = %d)" % (example.label, label_id))
features[-1].append(
InputFeatures(input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id))
if len(features[-1]) == n_class:
features.append([])
if len(features[-1]) == 0:
features = features[:-1]
print('#features', len(features))
return features
