def convert_single_example(ex_index, example, label_map, max_seq_length,
tokenizer, mode):
textlist = example.text.split(' ')
labellist = example.label.split(' ')
tokens = []
labels = []
for i, word in enumerate(textlist):
print("word= ", word)
token = tokenizer.tokenize(word)
if '▁' in token:
token.remove('▁')
print("token= ", token)
tokens.extend(token)
label_1 = labellist[i]
for m in range(len(token)):
if m == 0:
labels.append(label_1)
else:
labels.append("X")
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:(max_seq_length - 2)]
labels = labels[0:(max_seq_length - 2)]
ntokens = []
segment_ids = []
label_ids = []
ntokens.append("[CLS]")
segment_ids.append(0)
# append("O") or append("[CLS]") not sure!
label_ids.append(label_map["[CLS]"])
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
ntokens.append("[SEP]")
segment_ids.append(0)
# append("O") or append("[SEP]") not sure!
label_ids.append(label_map["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(ntokens)
input_mask = [1] * len(input_ids)
# label_mask = [1] * len(input_ids)
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
# we don't concerned about it!
label_ids.append(0)
ntokens.append("[PAD]")
print(len(input_ids))
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label_ids: %s" % " ".join([str(x)
for x in label_ids]))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids,
)
write_tokens(ntokens, mode)
return feature
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
label_map = {}
for (i, label) in enumerate(label_list, 1):
label_map[label] = i
#print('sample example: {} {}'.format(example.text, example.label))
textlist = example.text.split(' ')
labellist = example.label.split(' ')
tokens = []
labels = []
for i, word in enumerate(textlist):
token = tokenizer.tokenize(word)
tokens.extend(token)
label_1 = labellist[i]
for m in range(len(token)):
if m == 0:
labels.append(label_1)
else:
labels.append("X")
# tokens = tokenizer.tokenize(example.text)
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:(max_seq_length - 2)]
labels = labels[0:(max_seq_length - 2)]
ntokens = []
segment_ids = []
label_ids = []
ntokens.append("[CLS]")
segment_ids.append(0)
label_ids.append(0)
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
ntokens.append("[SEP]")
segment_ids.append(0)
label_ids.append(0)
input_ids = tokenizer.convert_tokens_to_ids(ntokens)
input_mask = [1] * len(input_ids)
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
label_ids.append(0)
# print(len(input_ids))
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label_ids: %s" % " ".join([str(x)
for x in label_ids]))
feature = InputFeatures(input_ids=input_ids,
input_tokens=ntokens,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids)
return feature
def convert_examples_to_features(examples, seq_length, tokenizer):
"""Loads a data file into a list of `InputBatch`s."""
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, seq_length - 3)
else:
# Account for [CLS] and [SEP] with "- 2"
if len(tokens_a) > seq_length - 2:
tokens_a = tokens_a[0:(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 unambiguously 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 = []
input_type_ids = []
tokens.append("[CLS]")
input_type_ids.append(0)
for token in tokens_a:
tokens.append(token)
input_type_ids.append(0)
tokens.append("[SEP]")
input_type_ids.append(0)
if tokens_b:
for token in tokens_b:
tokens.append(token)
input_type_ids.append(1)
tokens.append("[SEP]")
input_type_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) < seq_length:
input_ids.append(0)
input_mask.append(0)
input_type_ids.append(0)
assert len(input_ids) == seq_length
assert len(input_mask) == seq_length
assert len(input_type_ids) == seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("unique_id: %s" % (example.unique_id))
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
tf.logging.info(
"input_type_ids: %s" % " ".join([str(x) for x in input_type_ids]))
features.append(
InputFeatures(
unique_id=example.unique_id,
tokens=tokens,
input_ids=input_ids,
input_mask=input_mask,
input_type_ids=input_type_ids))
return features
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer, mode):
"""
将一个样本进行分析，然后将字转化为id, 标签转化为id,然后结构化到InputFeatures对象中
:param ex_index: index
:param example: 一个样本
:param label_list: 标签列表
:param max_seq_length:
:param tokenizer:
:param mode:
:return:
"""
label_map = {}
# 1表示从1开始对label进行index化
for (i, label) in enumerate(label_list, 1):
label_map[label] = i
# 保存label->index 的map
if not os.path.exists(os.path.join(FLAGS.output_dir, 'label2id.pkl')):
with codecs.open(os.path.join(FLAGS.output_dir, 'label2id.pkl'),
'wb') as w:
pickle.dump(label_map, w)
textlist = example.text.split(' ')
labellist = example.label.split(' ')
tokens = []
labels = []
for i, word in enumerate(textlist):
# 分词，如果是中文，就是分字
token = tokenizer.tokenize(word)
tokens.extend(token)
label_1 = labellist[i]
for m in range(len(token)):
if m == 0:
labels.append(label_1)
else: # 一般不会出现else
labels.append("X")
# tokens = tokenizer.tokenize(example.text)
# 序列截断
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:(max_seq_length - 2)] # -2 的原因是因为序列需要加一个句首和句尾标志
labels = labels[0:(max_seq_length - 2)]
ntokens = []
segment_ids = []
label_ids = []
ntokens.append("[CLS]") # 句子开始设置CLS 标志
segment_ids.append(0)
# append("O") or append("[CLS]") not sure!
label_ids.append(
label_map["[CLS]"]
) # O OR CLS 没有任何影响，不过我觉得O 会减少标签个数,不过拒收和句尾使用不同的标志来标注，使用LCS 也没毛病
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
ntokens.append("[SEP]") # 句尾添加[SEP] 标志
segment_ids.append(0)
# append("O") or append("[SEP]") not sure!
label_ids.append(label_map["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(
ntokens) # 将序列中的字(ntokens)转化为ID形式
input_mask = [1] * len(input_ids)
# label_mask = [1] * len(input_ids)
# padding, 使用
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
# we don't concerned about it!
label_ids.append(0)
ntokens.append("**NULL**")
# label_mask.append(0)
# print(len(input_ids))
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
# assert len(label_mask) == max_seq_length
# 打印部分样本数据信息
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label_ids: %s" % " ".join([str(x)
for x in label_ids]))
# tf.logging.info("label_mask: %s" % " ".join([str(x) for x in label_mask]))
# 结构化为一个类
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids,
# label_mask = label_mask
)
# mode='test'的时候才有效
write_tokens(ntokens, mode)
return feature
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
if isinstance(example, PaddingInputExample):
return InputFeatures(
input_ids=[0] * max_seq_length,
input_mask=[0] * max_seq_length,
segment_ids=[0] * max_seq_length,
label_id=[0] * max_seq_length,
is_real_example=False)
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
labelfile = os.path.join(FLAGS.output_dir, "nerlabel2id.pkl")
with open(labelfile, 'wb') as fw:
pickle.dump(label_map, fw)
tokens = example.text.split("\t")
# tokens = tokenizer.tokenize(example.text)
labels = example.label.split("\t")
# Account for [CLS] and [SEP] with "- 2"
if len(tokens) > max_seq_length - 2:
tokens = tokens[0:(max_seq_length - 2)]
labels = labels[0:(max_seq_length - 2)]
ntokens = []
segment_ids = []
label_id = []
ntokens.append("[CLS]")
segment_ids.append(0)
label_id.append(label_map["[CLS]"])
# for token in tokens:
for i in range(len(tokens)):
ntokens.append(tokens[i])
segment_ids.append(0)
label_id.append(label_map[labels[i]])
ntokens.append("[SEP]")
segment_ids.append(0)
label_id.append(label_map["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(ntokens)
# 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)
label_id.append(label_map["[PAD]"])
ntokens.append("[PAD]")
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_id) == max_seq_length
assert len(ntokens) == max_seq_length
# label_id = label_map[example.label]
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in ntokens]))
tf.logging.info("input_ids: %s" %
" ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label_id: %s" % " ".join([str(x) for x in label_id]))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id,
is_real_example=True)
return feature
def convert_single_example(ex_index, example, max_seq_length, tokenizer, mode):
"""
将一个样本进行分析，然后将字转化为id, 标签转化为id,然后结构化到InputFeatures对象中
:param ex_index: 样本在examples列表中的index
:param example: 一个InputExample样本对象，包含了source和target句子
:param max_seq_length: 序列的最大长度
:param tokenizer: tokenizer对象
:param mode: 模式，训练，验证，预测
:return:
"""
# 在这里采用wordpiece算法对source和target进行分词
tokens_source = tokenizer.tokenize(example.source)
tokens_target = tokenizer.tokenize(example.target)
# 序列截断，在这里 -2 的原因是因为序列需要加一个句首[CLS]和句尾[SEP]标志
if len(tokens_source) > max_seq_length - 2:
tokens_source = tokens_source[0:(max_seq_length - 2)]
if len(tokens_target) > max_seq_length - 2:
tokens_target = tokens_target[0:(max_seq_length - 2)]
# 对输入的token首尾分别加上[CLS]和[SEP]
ntokens_source = ["[CLS]"] + tokens_source + ["[SEP]"]
ntokens_target = ["[CLS]"] + tokens_target + ["[SEP]"]
# 将ntokens进行index映射，转化为index的形式, 映射的vocab是传入的vocab.txt
source_ids = tokenizer.convert_tokens_to_ids(ntokens_source)
target_ids = tokenizer.convert_tokens_to_ids(ntokens_target)
# 初始化句子向量
source_segment_ids = [0] * len(ntokens_source)
target_segment_ids = [0] * len(ntokens_target)
# mask，真实的token用1表示，pad用0表示，只有真实的token会被attention
source_mask = [1] * len(source_ids)
target_mask = [1] * len(target_ids)
# 按照最大长度补全
while len(source_ids) < max_seq_length:
source_ids.append(0)
source_mask.append(0)
source_segment_ids.append(0)
while len(target_ids) < max_seq_length:
target_ids.append(0)
target_mask.append(0)
target_segment_ids.append(0)
assert len(source_ids) == max_seq_length
assert len(source_mask) == max_seq_length
assert len(source_segment_ids) == max_seq_length
assert len(target_ids) == max_seq_length
assert len(target_mask) == max_seq_length
assert len(target_segment_ids) == max_seq_length
# 打印部分样本数据信息
if ex_index < 1:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens_source]))
tf.logging.info("source_ids: %s" %
" ".join([str(x) for x in source_ids]))
tf.logging.info("source_mask: %s" %
" ".join([str(x) for x in source_mask]))
tf.logging.info("source_segment_ids: %s" %
" ".join([str(x) for x in source_segment_ids]))
tf.logging.info("target_ids: %s" %
" ".join([str(x) for x in target_ids]))
tf.logging.info("target_mask: %s" %
" ".join([str(x) for x in target_mask]))
tf.logging.info("target_segment_ids: %s" %
" ".join([str(x) for x in target_segment_ids]))
# 实例化成一个InputFeatures对象
feature = InputFeatures(source_ids=source_ids,
source_mask=source_mask,
source_segment_ids=source_segment_ids,
target_ids=target_ids,
target_mask=target_mask,
target_segment_ids=target_segment_ids)
return feature
def convert_single_example_to_feature(ex_index, example, label_map,
max_seq_length, tokenizer, mode):
textlist = example.tokens
poslist = example.poss
chunklist = example.chunks
labellist = example.labels
tokens = []
poss = []
chunks = []
labels = []
for i, word in enumerate(textlist):
token = tokenizer.tokenize(word)
tokens.extend(token)
pos_1 = poslist[i]
chunk_1 = chunklist[i]
label_1 = labellist[i]
for m in range(len(token)):
if m == 0:
poss.append(pos_1)
chunks.append(chunk_1)
labels.append(label_1)
else:
poss.append("X")
chunks.append("X")
labels.append("X")
# tokens = tokenizer.tokenize(example.text)
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:(max_seq_length - 2)]
poss = poss[0:(max_seq_length - 2)]
chunks = chunks[0:(max_seq_length - 2)]
labels = labels[0:(max_seq_length - 2)]
# save tokens, poss, chunks, labels back to example
example.tokenized_tokens = tokens
example.tokenized_poss = poss
example.tokenized_chunks = chunks
example.tokenized_labels = labels
ntokens = []
segment_ids = []
label_ids = []
ntokens.append("[CLS]")
segment_ids.append(0)
# label_ids.append(label_map["[CLS]"])
label_ids.append(0)
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
ntokens.append("[SEP]")
segment_ids.append(0)
# label_ids.append(label_map["[SEP]"])
label_ids.append(0)
input_ids = tokenizer.convert_tokens_to_ids(ntokens)
input_mask = [1] * len(input_ids)
# padding
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
# we don't concerned about it!
label_ids.append(0)
ntokens.append("**NULL**")
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
if ex_index < 5:
print("*** Example ***")
print("guid: %s" % (example.guid))
print("tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
print("input_ids: %s" % " ".join([str(x) for x in input_ids]))
print("input_mask: %s" % " ".join([str(x) for x in input_mask]))
print("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
print("label_ids: %s" % " ".join([str(x) for x in label_ids]))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids,
)
return feature
def write_instance_to_example_files(instances, tokenizer, max_seq_length,
max_predictions_per_seq, output_files):
"""Create TF example files from `TrainingInstance`s."""
writers = []
for output_file in output_files:
writers.append(tf.python_io.TFRecordWriter(output_file))
writer_index = 0
total_written = 0
for (inst_index, instance) in enumerate(instances):
input_ids = tokenizer.convert_tokens_to_ids(instance.tokens)
input_mask = [1] * len(input_ids)
segment_ids = list(instance.segment_ids)
assert len(input_ids) <= max_seq_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
# print("length of segment_ids:",len(segment_ids),"max_seq_length:", max_seq_length)
assert len(segment_ids) == max_seq_length
masked_lm_positions = list(instance.masked_lm_positions)
masked_lm_ids = tokenizer.convert_tokens_to_ids(
instance.masked_lm_labels)
masked_lm_weights = [1.0] * len(masked_lm_ids)
while len(masked_lm_positions) < max_predictions_per_seq:
masked_lm_positions.append(0)
masked_lm_ids.append(0)
masked_lm_weights.append(0.0)
next_sentence_label = 1 if instance.is_random_next else 0
features = collections.OrderedDict()
features["input_ids"] = create_int_feature(input_ids)
features["input_mask"] = create_int_feature(input_mask)
features["segment_ids"] = create_int_feature(segment_ids)
features["masked_lm_positions"] = create_int_feature(
masked_lm_positions)
features["masked_lm_ids"] = create_int_feature(masked_lm_ids)
features["masked_lm_weights"] = create_float_feature(masked_lm_weights)
features["next_sentence_labels"] = create_int_feature(
[next_sentence_label])
tf_example = tf.train.Example(features=tf.train.Features(
feature=features))
writers[writer_index].write(tf_example.SerializeToString())
writer_index = (writer_index + 1) % len(writers)
total_written += 1
if inst_index < 20:
tf.logging.info("*** Example ***")
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in instance.tokens]))
for feature_name in features.keys():
feature = features[feature_name]
values = []
if feature.int64_list.value:
values = feature.int64_list.value
elif feature.float_list.value:
values = feature.float_list.value
tf.logging.info(
"%s: %s" %
(feature_name, " ".join([str(x) for x in values])))
for writer in writers:
writer.close()
tf.logging.info("Wrote %d total instances", total_written)
def convert_examples_to_features(examples,
tokenizer,
max_seq_length,
doc_stride,
max_query_length,
is_training,
verbose_logging=False,
logger=None):
"""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
out_of_span = False
if not (tok_start_position >= doc_start
and tok_end_position <= doc_end):
out_of_span = True
if out_of_span:
start_position = 0
end_position = 0
else:
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 < 2 and verbose_logging:
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([tokenization.printable_text(x) for x in tokens]))
logger.info("token_to_orig_map: %s" % " ".join([
"%d:%d" % (x, y)
for (x, y) in six.iteritems(token_to_orig_map)
]))
logger.info("token_is_max_context: %s" % " ".join([
"%d:%s" % (x, y)
for (x, y) in six.iteritems(token_is_max_context)
]))
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" %
(tokenization.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
if len(features) % 5000 == 0:
logger.info("Processing features: %d" % (len(features)))
return features
def convert_examples_to_features(examples, label_list, max_seq_length,
tokenizer):
"""Loads a data file into a list of `InputBatch`s.
Convert examples into token form as input of BERT model.
"""
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
input_data = []
guids = []
for (ex_index, example) in enumerate(examples):
guid = example.guid
guids.append(guid)
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)]
if ex_index % 10000 == 0:
tf.logging.info("Writing example %d of %d" %
(ex_index, len(examples)))
# 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)
input_mask = [1] * len(input_ids)
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 < 3:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" %
" ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label: %s (id = %d)" % (example.label, label_id))
features = collections.OrderedDict()
features["input_ids"] = input_ids
features["input_mask"] = input_mask
features["segment_ids"] = segment_ids
features["label_ids"] = label_id
features["guid"] = guids
print(guids)
input_data.append(features)
return input_data
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer, mode):
"""
:param ex_index: example num
:param example:
:param label_list: all labels
:param max_seq_length:
:param tokenizer: WordPiece tokenization
:param mode:
:return: feature
IN this part we should rebuild input sentences to the following format.
example:[Jim,Hen,##son,was,a,puppet,##eer]
labels: [B,I,X,O,O,O,X]
"""
label_map = {}
# here start with zero which means that "[PAD]" is zero
# start with 1, 0 for paddding
for i, label in enumerate(label_list, 1):
label_map[label] = i
with open(FLAGS.output_dir + '/label2id.pkl', 'wb+') as wf:
pickle.dump(label_map, wf)
textlist = example.text.split(' ')
labellist = example.label.split(' ')
tokens = []
labels = []
for i, (word, label) in enumerate(zip(textlist, labellist)):
token = tokenizer.tokenize(word)
tokens.extend(token)
for m in range(len(token)):
if m == 0:
labels.append(label)
else:
labels.append('X')
# only Account for [CLS] with "- 1".
# account for ending signal [SEP], with "- 2"
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:max_seq_length - 2]
labels = labels[0:max_seq_length - 2]
ntokens = []
segment_ids = []
label_ids = []
# begining signal [CLS]
ntokens.append("[CLS]")
segment_ids.append(0)
label_ids.append(label_map["[CLS]"])
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
'''
after that we don't add "[SEP]" because we want a sentence don't have
stop tag, because i think its not very necessary.
or if add "[SEP]" the model even will cause problem, special the crf layer was used.
'''
# endinging signal [SEP]
ntokens.append("[SEP]")
segment_ids.append(0)
label_ids.append(label_map["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(ntokens)
input_mask = [1] * len(input_ids)
#use zero to padding sequence
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
label_ids.append(0)
ntokens.append("**NULL**")
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
if ex_index < 5:
logging.info("*** Example ***")
logging.info("guid: %s" % (example.guid))
logging.info("tokens: %s" %
" ".join([tokenization.printable_text(x)
for x in tokens]))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids,
)
# we need ntokens because if we do predict it can help us return to original token.
return feature, ntokens, label_ids
def convert_single_example(ex_index, example, max_seq_length, tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
tokens = tokenizer.tokenize(example.text)
# Account for [CLS] and [SEP] with "- 2"
if len(tokens) > max_seq_length - 2:
tokens = tokens[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 unambiguously 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 the "sentence vector". Note that this only makes sense because
# the entire model is fine-tuned.
tokens = ["[CLS]"] + tokens + ["[SEP]"]
segment_ids = [0] * len(tokens)
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 ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info("tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
tf.logging.info("labels: %s" % " ".join([str(x) for x in example.labels]))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=example.labels)
return feature
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
if isinstance(example, PaddingInputExample):
return InputFeatures(input_ids=[0] * max_seq_length,
input_mask=[0] * max_seq_length,
segment_ids=[0] * max_seq_length,
label_id=0,
is_real_example=False)
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
with open(FLAGS.middle_output + "/label2id.pkl", 'wb') as w:
pickle.dump(label_map, w)
tokens_a = tokenizer.tokenize(example.text_a)
if len(tokens_a) > max_seq_length - 1:
tokens_a = tokens_a[0:(max_seq_length - 1)]
tokens = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for token in tokens_a:
tokens.append(token)
segment_ids.append(0)
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 < 3:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label: %s (id = %d)" % (example.label, label_id))
feature = InputFeatures(input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id,
is_real_example=True)
return feature, label_id
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer, mode):
label_map = {}
for (i, label) in enumerate(label_list, 1):
label_map[label] = i
with open('./output/label2id.pkl', 'wb') as w:
pickle.dump(label_map, w)
textlist = example.text.split(' ')
labellist = example.label.split(' ')
tokens = []
labels = []
for i, word in enumerate(textlist):
token = tokenizer.tokenize(word)
tokens.extend(token)
label_1 = labellist[i]
for m in range(len(token)):
if m == 0:
labels.append(label_1)
else:
labels.append("X")
# tokens = tokenizer.tokenize(example.text)
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:(max_seq_length - 2)]
labels = labels[0:(max_seq_length - 2)]
ntokens = []
segment_ids = []
label_ids = []
ntokens.append("[CLS]")
segment_ids.append(0)
# append("O") or append("[CLS]") not sure!
label_ids.append(label_map["[CLS]"])
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
ntokens.append("[SEP]")
segment_ids.append(0)
# append("O") or append("[SEP]") not sure!
label_ids.append(label_map["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(ntokens)
input_mask = [1] * len(input_ids)
# label_mask = [1] * len(input_ids)
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
# we don't concerned about it!
label_ids.append(0)
ntokens.append("**NULL**")
# label_mask.append(0)
# print(len(input_ids))
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
# assert len(label_mask) == max_seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label_ids: %s" % " ".join([str(x)
for x in label_ids]))
# tf.logging.info("label_mask: %s" % " ".join([str(x) for x in label_mask]))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids,
# label_mask = label_mask
)
write_tokens(ntokens, mode)
write_labels(labels, mode)
return feature
def convert_single_example(ex_index, example, token_label_list, predicate_label_list, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
if isinstance(example, PaddingInputExample):
return InputFeatures(
input_ids=[0] * max_seq_length,
input_mask=[0] * max_seq_length,
segment_ids=[0] * max_seq_length,
token_label_ids=[0] * max_seq_length,
predicate_label_id = [0],
is_real_example=False)
token_label_map = {}
for (i, label) in enumerate(token_label_list):
token_label_map[label] = i
predicate_label_map = {}
for (i, label) in enumerate(predicate_label_list):
predicate_label_map[label] = i
text_token = example.text_token.split("\t")[0].split(" ")
if example.token_label is not None:
token_label = example.token_label.split("\t")[0].split(" ")
else:
token_label = ["O"] * len(text_token)
assert len(text_token) == len(token_label)
text_predicate = example.text_token.split("\t")[1]
if example.token_label is not None:
token_predicate = example.token_label.split("\t")[1]
else:
token_predicate = text_predicate
assert text_predicate == token_predicate
tokens_b = [text_predicate] * len(text_token)
predicate_id = predicate_label_map[text_predicate]
_truncate_seq_pair(text_token, tokens_b, max_seq_length - 3)
tokens = []
token_label_ids = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
token_label_ids.append(token_label_map["[CLS]"])
for token, label in zip(text_token, token_label):
tokens.append(token)
segment_ids.append(0)
token_label_ids.append(token_label_map[label])
tokens.append("[SEP]")
segment_ids.append(0)
token_label_ids.append(token_label_map["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(tokens)
#bert_tokenizer.convert_tokens_to_ids(["[SEP]"]) --->[102]
bias = 1 #1-100 dict index not used
for token in tokens_b:
input_ids.append(predicate_id + bias) #add bias for different from word dict
segment_ids.append(1)
token_label_ids.append(token_label_map["[category]"])
input_ids.append(tokenizer.convert_tokens_to_ids(["[SEP]"])[0]) #102
segment_ids.append(1)
token_label_ids.append(token_label_map["[SEP]"])
# 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)
token_label_ids.append(0)
tokens.append("[Padding]")
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(token_label_ids) == max_seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
tf.logging.info("token_label_ids: %s" % " ".join([str(x) for x in token_label_ids]))
tf.logging.info("predicate_id: %s" % str(predicate_id))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
token_label_ids=token_label_ids,
predicate_label_id=[predicate_id],
is_real_example=True)
return feature
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
'''Converts a single `InputClass` into a single `InputFeatures`.'''
if isinstance(example, PaddingClass):
return InputFeatures(input_ids=[0] * max_seq_length,
input_mask=[0] * max_seq_length,
segment_ids=[0] * max_seq_length,
label_id=0,
is_real_example=False)
label_map = {label: i for (i, label) in enumerate(label_list)}
tokens_a = tokenizer.tokenize(example.text_a)
tokens_b = tokenizer.tokenize(example.text_b) if example.text_b else None
if tokens_b: # opt. for no take_b, - 3 = [CLS] + [SEP] * 2
_truncate_seq_pair(tokens_a, tokens_b, max_seq_length - 3)
else: # - 2 = [CLS] + [SEP]
if len(tokens_a) > max_seq_length - 2:
tokens_a = tokens_a[0:(max_seq_length - 2)]
# (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
tokens = ['[CLS]', *[ii for ii in tokens_a], '[SEP]']
segment_ids = [0] * (len(tokens_a) + 1) # No.X sentences
if tokens_b:
tokens = [*tokens, *[ii for ii in tokens_b], '[SEP]']
segment_ids = [*segment_ids, *[1] * (len(tokens_b) + 1)]
input_ids = tokenizer.convert_tokens_to_ids(tokens)
input_mask = [1] * len(input_ids) # mask 0: pad; 1: real
pad_len = max_seq_length - len(input_ids)
if not pad_len: # zero-padded
input_ids = [*input_ids, *[0] * pad_len]
input_mask = [*input_mask, *[0] * pad_len]
segment_ids = [*segment_ids, *[0] * pad_len]
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:
tf.logging.info(">>>>> Example >>>>>")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label: %s (id = %d)" % (example.label, label_id))
feature = InputFeatures(input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id,
is_real_example=True)
return feature
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
if isinstance(example, PaddingInputExample):
return InputFeatures(input_ids=[0] * max_seq_length,
input_mask=[0] * max_seq_length,
segment_ids=[0] * max_seq_length,
label_id=0,
is_real_example=False)
tokens_a = tokenizer.tokenize(example.text_a)
tokens_b = tokenizer.tokenize(example.text_b)
while True:
total_length = len(tokens_a) + len(tokens_b)
if total_length <= max_seq_length - 3:
break
if len(tokens_a) > len(tokens_b):
tokens_a.pop()
else:
tokens_b.pop()
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)
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
if example.label in label_list:
label_id = int(example.label)
else:
label_id = -1
if ex_index < 3:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label: %s (id = %d)" % (example.label, label_id))
feature = InputFeatures(input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id,
is_real_example=True)
raw_tokens = example.text_a + '[SEP]' + example.text_b
return feature, raw_tokens, label_id
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
if isinstance(example, PaddingInputExample):
return InputFeatures(input_ids=[0] * max_seq_length,
input_mask=[0] * max_seq_length,
segment_ids=[0] * max_seq_length,
label_id=0,
is_real_example=False)
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
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 unambiguously 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 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:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label: %s (id = %d)" % (example.label, label_id))
feature = InputFeatures(input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id,
is_real_example=True)
return feature
def convert_single_example(ex_index, example, label_list, max_seq_length, tokenizer, forward, backward, mode):
"""
:param ex_index: example num
:param example:
:param label_list: all labels
:param max_seq_length:
:param tokenizer: WordPiece tokenization
:param mode:
:return: feature
IN this part we should rebuild input sentences to the following format.
example:[Jim,Hen,##son,was,a,puppet,##eer]
labels: [B,I,X,O,O,O,X]
"""
label_map = dict()
# here start with zero which means that "[PAD]" is zero
# start with 1, 0 for paddding
for i, label in enumerate(label_list, 1):
label_map[label] = i
with open('./data/{}/label2id.pkl'.format(FLAGS.dataset), 'wb+') as wf:
pickle.dump(label_map, wf)
pos2id = dict()
with open('./data/{}/pos2id.pkl'.format(FLAGS.dataset), 'rb') as rf:
pos2id = pickle.load(rf)
with open('./data/{}/mark2id.pkl'.format(FLAGS.dataset), 'rb') as rf2:
mark2id = pickle.load(rf2)
textlist = example.text.split(' ')
labellist = example.label.split(' ')
# featurelist = [list(map(int, features.split('-'))) for features in example.feature.split(' ')]
poslist = example.pos.split(' ')
marklist = example.mark.split(' ')
tokens = []
labels = []
poses = []
marks = []
for i, (word, label, pos, mark) in enumerate(zip(textlist, labellist, poslist, marklist)):
token = tokenizer.tokenize(word)
tokens.extend(token)
for m in range(len(token)):
if m == 0:
labels.append(label)
poses.append(pos)
marks.append(mark)
else:
labels.append('X')
poses.append(0)
marks.append(0)
# only Account for [CLS] with "- 1".
# account for ending signal [SEP], with "- 2"
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0: max_seq_length - 2]
labels = labels[0: max_seq_length - 2]
poses = poses[0: max_seq_length - 2]
marks = marks[0: max_seq_length - 2]
ntokens = []
segment_ids = []
label_ids = []
pos_ids = []
mark_ids = []
# begin signal [CLS]
ntokens.append("[CLS]")
segment_ids.append(0)
label_ids.append(label_map["[CLS]"])
pos_ids.append(pos2id["[CLS]"])
mark_ids.append(mark2id["[CLS]"])
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
pos_ids.append(int(poses[i]))
mark_ids.append(int(marks[i]))
# ending signal [SEP]
ntokens.append("[SEP]")
segment_ids.append(0)
label_ids.append(label_map["[SEP]"])
pos_ids.append(pos2id["[SEP]"])
mark_ids.append(pos2id["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(ntokens)
input_mask = [1] * len(input_ids)
# use zero to padding sequence
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
label_ids.append(0)
pos_ids.append(0)
mark_ids.append(0)
ntokens.append("**NULL**")
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
assert len(pos_ids) == max_seq_length
assert len(mark_ids) == max_seq_length
if ex_index < 5:
logging.info("*** Example ***")
logging.info("guid: %s" % example.guid)
logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in tokens]))
logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
logging.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
logging.info("label_ids: %s" % " ".join([str(x) for x in label_ids]))
logging.info("pos_ids: %s" % " ".join([str(x) for x in pos_ids]))
logging.info("mark_ids: %s" % " ".join([str(x) for x in mark_ids]))
logging.info("ntokens: %s" % " ".join([str(x) for x in ntokens]))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids,
pos_ids=pos_ids,
mark_ids=mark_ids,
forward=forward,
backward=backward,
)
# we need ntokens because if we do predict it can help us return to original token.
return feature, ntokens, label_ids
def convert_single_example(ex_index, example, entity_label_list,
class_label_list, max_seq_length, tokenizer, mode):
class_label_map = {}
for (i, label) in enumerate(class_label_list):
class_label_map[label] = i
label_map = {}
label_map_reference = {}
for (i, label) in enumerate(entity_label_list, 1):
label_map[label] = i
label_map_reference[label] = i
label_map_reference['x'] = len(entity_label_list) + 1
label2idpath = os.path.join(FLAGS.output_dir, 'label2id.json')
label2idpath_reference = os.path.join(FLAGS.output_dir,
'label2id_reference.json')
if not os.path.exists(label2idpath_reference):
with open(label2idpath_reference, 'w', encoding='utf8') as w:
json.dump(label_map_reference, w)
textlist = example.text.split()
labellist = list(example.entity_label)
tokens = []
labels = []
reference_labels = []
unknow_index = []
for i, word in enumerate(textlist):
token = tokenizer.tokenize(word)
tokens.extend(token)
label_1 = labellist[i]
for m in range(len(token)):
if m == 0:
labels.append(label_1)
reference_labels.append(label_1)
else:
if label_1 == 'o':
labels.append('o')
else:
labels.append('m')
reference_labels.append('x')
if token[m] == "[UNK]":
unknow_index.append(i)
assert len(tokens) == len(labels) == len(reference_labels)
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:(max_seq_length - 2)]
labels = labels[0:(max_seq_length - 2)]
reference_labels = reference_labels[0:(max_seq_length - 2)]
ntokens = []
segment_ids = []
label_ids = []
reference_label_ids = []
ntokens.append("[CLS]")
segment_ids.append(0)
label_ids.append(label_map['o'])
reference_label_ids.append(label_map_reference['o'])
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
reference_label_ids.append(label_map_reference[reference_labels[i]])
ntokens.append("[SEP]")
segment_ids.append(0)
label_ids.append(label_map['o'])
reference_label_ids.append(label_map_reference['o'])
input_ids = tokenizer.convert_tokens_to_ids(ntokens)
input_mask = [1] * len(input_ids)
class_label = class_label_map[example.class_label]
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
label_ids.append(0)
reference_label_ids.append(0)
ntokens.append("**NULL**")
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
assert len(reference_label_ids) == max_seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label_ids: %s" % " ".join([str(x)
for x in label_ids]))
feature = InputFeatures(input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids,
reference_label_ids=reference_label_ids,
class_label=class_label)
output_tokens = []
for i, each in enumerate(ntokens):
if each != "[UNK]":
output_tokens.append(each)
else:
index = unknow_index[0]
output_tokens.append(textlist[index])
unknow_index = unknow_index[1:]
write_tokens(output_tokens, mode)
return feature
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""
将一个样本进行分析，然后将字转化为id, 标签转化为id,然后结构化到InputFeatures对象中
:param ex_index: index
:param example: 一个样本
:param label_list: 标签列表
:param max_seq_length:
:param tokenizer:
:return:
"""
label_map = {}
# 1表示从1开始对label进行index化
for (i, label) in enumerate(label_list, 1):
label_map[label] = i
tokens_a = tokenizer.tokenize(example.text_a)
label_a = example.label.split(' ')
if len(tokens_a) > max_seq_length:
tokens_a = tokens_a[0:max_seq_length]
label_a = label_a[0:max_seq_length]
tokens = []
label_ids = []
segment_ids = []
for index, token in enumerate(tokens_a):
tokens.append(token)
label_ids.append(label_map[label_a[index]])
segment_ids.append(0)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
input_mask = [1] * len(input_ids)
while len(input_ids) < max_seq_length:
input_ids.append(0)
label_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
assert len(label_ids) == max_seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label_ids: %s" % " ".join([str(x)
for x in label_ids]))
feature = InputFeatures(input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids)
return feature
def convert_single_example_train(ex_index, example, label_list, max_seq_length, tokenizer, output_dir, mode):
label_map = {}
for (i, label) in enumerate(label_list, 1):
label_map[label] = i
if not os.path.exists(os.path.join(output_dir, 'label2id.pkl')):
with codecs.open(os.path.join(output_dir, 'label2id.pkl'), 'wb') as w:
pickle.dump(label_map, w)
textlist = example.text.split(' ')
labellist = example.label.split(' ')
tokens = []
labels = []
for i, word in enumerate(textlist):
# 分词，如果是中文，就是分字,但是对于一些不在BERT的vocab.txt中得字符会被进行WordPice处理
token = tokenizer.tokenize(word)
tokens.extend(token)
label_1 = labellist[i]
for m in range(len(token)):
if m == 0:
labels.append(label_1)
else: # 一般不会出现else
labels.append("X")
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:(max_seq_length - 2)]
labels = labels[0:(max_seq_length - 2)]
ntokens = ["[CLS]"]
segment_ids = [0]
label_ids = [label_map["[CLS]"]]
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
ntokens.append("[SEP]")
segment_ids.append(0)
label_ids.append(label_map["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(ntokens)
input_mask = [1] * len(input_ids)
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
label_ids.append(0)
ntokens.append("**NULL**")
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
tf.logging.info("label_ids: %s" % " ".join([str(x) for x in label_ids]))
feature = InputFeatures(input_ids=input_ids, input_mask=input_mask,
segment_ids=segment_ids, label_ids=label_ids)
# mode='test'的时候才有效
write_tokens(ntokens, output_dir, mode)
return feature
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""
:param ex_index: example num
:param example:
:param label_list: all labels
:param max_seq_length:
:param tokenizer: WordPiece tokenization
:param mode:
:return: feature
IN this part we should rebuild input sentences to the following format.
example:[Jim,Hen,##son,was,a,puppet,##eer]
labels: [I-PER,I-PER,X,O,O,O,X]
"""
label_map = {}
# # here start with zero this means that "[PAD]" is zero
for (i, label) in enumerate(label_list):
label_map[label] = i
with open(info.middle_output + "/label2id.pkl", 'wb') as w:
pickle.dump(label_map, w)
textlist = example.text.split(' ')
labellist = example.label.split(' ')
tokens = []
labels = []
for i, (word, label) in enumerate(zip(textlist, labellist)):
token = tokenizer.tokenize(word)
tokens.extend(token)
for i, _ in enumerate(token):
if i == 0:
labels.append(label)
else:
labels.append("X")
# only Account for [CLS] with "- 1".
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:(max_seq_length - 1)]
labels = labels[0:(max_seq_length - 1)]
ntokens = []
segment_ids = []
label_ids = []
ntokens.append("[CLS]")
segment_ids.append(0)
label_ids.append(label_map["[CLS]"])
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
# after that we don't add "[SEP]" because we want a sentence don't have
# stop tag, because i think its not very necessary.
# or if add "[SEP]" the model even will cause problem, special the crf layer was used.
input_ids = tokenizer.convert_tokens_to_ids(ntokens)
mask = [1] * len(input_ids)
# use zero to padding and you should
while len(input_ids) < max_seq_length:
input_ids.append(0)
mask.append(0)
segment_ids.append(0)
label_ids.append(0)
ntokens.append("[PAD]")
assert len(input_ids) == max_seq_length
assert len(mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
assert len(ntokens) == max_seq_length
if ex_index < 3:
logging.info("*** Example ***")
logging.info("guid: %s" % (example.guid))
logging.info("tokens: %s" %
" ".join([tokenization.printable_text(x)
for x in tokens]))
logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
logging.info("input_mask: %s" % " ".join([str(x) for x in mask]))
logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
logging.info("label_ids: %s" % " ".join([str(x) for x in label_ids]))
feature = InputFeatures(
input_ids=input_ids,
mask=mask,
segment_ids=segment_ids,
label_ids=label_ids,
)
# we need ntokens because if we do predict it can help us return to original token.
# return feature, ntokens, label_ids
return feature
def convert_single_example_dev(ex_index, text, label, label2id, max_seq_length,
tokenizer):
"""
将一个样本进行分析，然后将字转化为id, 标签转化为id,然后结构化到InputFeatures对象中
:param ex_index: index
:param example: 一个样本
:param label_list: 标签列表
:param max_seq_length:
:param tokenizer:
:param mode:
:return:
"""
O_index = label2id["O"]
# L: ['B-ORG', 'M-ORG', 'M-ORG', 'M-ORG']
# W: ['中', '共', '中', '央']
textlist = text.split(' ')
labellist = label.split(' ')
tokens = []
labels = []
for i, word in enumerate(textlist):
# 对每个字进行tokenize，返回list
token = tokenizer.tokenize(word)
tokens.extend(token)
label_1 = labellist[i]
for m in range(len(token)):
if m == 0:
labels.append(label_1)
else: # 一般不会出现else
labels.append("X")
# 序列截断
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:(max_seq_length - 2)] # -2 的原因是因为序列需要加一个句首和句尾标志
labels = labels[0:(max_seq_length - 2)]
ntokens = []
segment_ids = []
label_ids = []
ntokens.append("[CLS]") # 句子开始设置CLS 标志
segment_ids.append(0)
label_ids.append(label2id["[CLS]"]) #
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label2id[labels[i]])
ntokens.append("[SEP]") # 句尾添加[SEP] 标志
segment_ids.append(0)
label_ids.append(label2id["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(
ntokens) # 将序列中的字(ntokens)转化为ID形式
input_mask = [1] * len(input_ids)
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
# label用O去padding
label_ids.append(O_index)
ntokens.append("[PAD]")
# label_mask.append(0)
# print(len(input_ids))
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
# assert len(label_mask) == max_seq_length
# 打印部分样本数据信息
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label_ids: %s" % " ".join([str(x)
for x in label_ids]))
# tf.logging.info("label_mask: %s" % " ".join([str(x) for x in label_mask]))
# 结构化为一个类
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids,
# label_mask = label_mask
)
return feature
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
if isinstance(example, PaddingInputExample):
return InputFeatures(input_ids=[0] * max_seq_length,
input_mask=[0] * max_seq_length,
segment_ids=[0] * max_seq_length,
label_id=0,
is_real_example=False)
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
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)]
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:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info(
"tokens: %s" %
" ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x)
for x in input_ids]))
tf.logging.info("input_mask: %s" %
" ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" %
" ".join([str(x) for x in segment_ids]))
tf.logging.info("label: %s (id = %d)" % (example.label, label_id))
feature = InputFeatures(input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id,
is_real_example=True)
return feature
def convert_single_example(ex_index, example, label_list, max_seq_length, tokenizer, mode):
"""
将一个样本进行分析，然后将字转化为id, 标签转化为id,然后结构化到InputFeatures对象中
:param ex_index: index
:param example: 一个样本
:param label_list: 标签列表
:param max_seq_length:
:param tokenizer:
:param mode:
:return:
"""
label_map = {}
# 1表示从1开始对label进行index化
for (i, label) in enumerate(label_list, 1):
label_map[label] = i
# 保存label->index 的map
with codecs.open(os.path.join(FLAGS.output_dir, 'label2id.pkl'), 'wb') as w:
pickle.dump(label_map, w)
textlist = example.text.split(' ')
labellist = example.label.split(' ')
tokens = []
labels = []
for i, word in enumerate(textlist):
# 分词，如果是中文，就是分字
token = tokenizer.tokenize(word)
tokens.extend(token)
label_1 = labellist[i]
for m in range(len(token)):
if m == 0:
labels.append(label_1)
else: # 一般不会出现else
labels.append("X")
# tokens = tokenizer.tokenize(example.text)
# 序列截断
if len(tokens) >= max_seq_length - 1:
tokens = tokens[0:(max_seq_length - 2)] # -2 的原因是因为序列需要加一个句首和句尾标志
labels = labels[0:(max_seq_length - 2)]
ntokens = []
segment_ids = []
label_ids = []
ntokens.append("[CLS]") # 句子开始设置CLS 标志
segment_ids.append(0)
# append("O") or append("[CLS]") not sure!
label_ids.append(label_map["[CLS]"]) # O OR CLS 没有任何影响，不过我觉得O 会减少标签个数,不过拒收和句尾使用不同的标志来标注，使用LCS 也没毛病
for i, token in enumerate(tokens):
ntokens.append(token)
segment_ids.append(0)
label_ids.append(label_map[labels[i]])
ntokens.append("[SEP]") # 句尾添加[SEP] 标志
segment_ids.append(0)
# append("O") or append("[SEP]") not sure!
label_ids.append(label_map["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(ntokens) # 将序列中的字(ntokens)转化为ID形式
input_mask = [1] * len(input_ids)
# label_mask = [1] * len(input_ids)
# padding, 使用
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
# we don't concerned about it!
label_ids.append(0)
ntokens.append("**NULL**")
# label_mask.append(0)
# print(len(input_ids))
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(label_ids) == max_seq_length
# assert len(label_mask) == max_seq_length
# 打印部分样本数据信息
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
tf.logging.info("label_ids: %s" % " ".join([str(x) for x in label_ids]))
# tf.logging.info("label_mask: %s" % " ".join([str(x) for x in label_mask]))
# 结构化为一个类
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_ids=label_ids,
# label_mask = label_mask
)
# mode='test'的时候才有效
write_tokens(ntokens, mode)
return feature
def convert_single_example(ex_index,
example,
token_label_list,
sent_label_list,
max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
if isinstance(example, PaddingInputExample):
return InputFeatures(
input_ids=[0] * max_seq_length,
input_masks=[0] * max_seq_length,
segment_ids=[0] * max_seq_length,
token_label_ids=[0] * max_seq_length,
sent_label_id=0)
token_label_map = {}
for (i, token_label) in enumerate(token_label_list):
token_label_map[token_label] = i
sent_label_map = {}
for (i, sent_label) in enumerate(sent_label_list):
sent_label_map[sent_label] = i
token_items = example.text.split(" ")
token_label_items = example.token_label.split(" ")
tokens = []
token_labels = []
for token, token_label in zip(token_items, token_label_items):
token_subitems = tokenizer.tokenize(token)
token_label_subitems = [token_label] + ["X"] * (len(token_subitems) - 1)
tokens.extend(token_subitems)
token_labels.extend(token_label_subitems)
if len(tokens) > max_seq_length - 2:
tokens = tokens[0:(max_seq_length - 2)]
if len(token_labels) > max_seq_length - 2:
token_labels = token_labels[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 unambiguously 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 the "sentence vector". Note that this only makes sense because
# the entire model is fine-tuned.
input_tokens = []
segment_ids = []
token_label_ids = []
sent_label_id = sent_label_map[example.sent_label]
input_tokens.append("[CLS]")
segment_ids.append(0)
token_label_ids.append(token_label_map["[CLS]"])
for i, token in enumerate(tokens):
input_tokens.append(token)
segment_ids.append(0)
token_label_ids.append(token_label_map[token_labels[i]])
input_tokens.append("[SEP]")
segment_ids.append(0)
token_label_ids.append(token_label_map["[SEP]"])
input_ids = tokenizer.convert_tokens_to_ids(input_tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real tokens are attended to.
input_masks = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_masks.append(0)
segment_ids.append(0)
token_label_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_masks) == max_seq_length
assert len(segment_ids) == max_seq_length
assert len(token_label_ids) == max_seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info("tokens: %s" % " ".join([tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
tf.logging.info("input_masks: %s" % " ".join([str(x) for x in input_masks]))
tf.logging.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
tf.logging.info("token_label_ids: %s" % " ".join([str(x) for x in token_label_ids]))
tf.logging.info("sent_label_id: %s" % str(sent_label_id))
feature = InputFeatures(
input_ids=input_ids,
input_masks=input_masks,
segment_ids=segment_ids,
token_label_ids=token_label_ids,
sent_label_id=sent_label_id)
return feature
