bert模型源码(tensorflow版本)——create_pretraining_data.py完全解读，详细解析

         下面的代码为create_pretraining_data.py的源码，其中使用“#*”开头的中文注释是我对于代码的解释，穿插在代码中间，请读者从该代码的main函数开始阅读，其中用到的每个函数都做了较为详细的解读

# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Create masked LM/next sentence masked_lm TF examples for BERT."""

# 和BERT原论文代码完全一致
# 此处定义了如何将普通文本转换成可用于预训练BERT模型的tfrecord文件的方法。
# 产生可以用于预训练模型的数据，其中包含MLM和NSP任务

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import collections
import random
import tokenization
import tensorflow as tf
import warnings
warnings.filterwarnings("ignore")

flags = tf.flags

FLAGS = flags.FLAGS

'''
input_file：输入文件路径
output_file：输出文件路径
vocab_file：谷歌提供的词典，值为词典的路径
do_lower_case：当值为True时，则忽略大小写
max_seq_length：每一条训练数据（两句话）相加后的最大长度限制
max_predictions_per_seq：每一条训练数据mask的最大数量
random_seed：一个随机种子
dupe_factor：对文档多次重复随机产生训练集，随机的次数
masked_lm_prob：一条训练数据产生mask的概率，即每条训练数据随机产生max_predictions_per_seq乘以masked_lm_prob数量的mask
short_seq_prob：为了缩小预训练和微调过程的差距，以此概率产生小于max_seq_length的训练数据
'''

flags.DEFINE_string("input_file", None,
                    "Input raw text file (or comma-separated list of files).")

flags.DEFINE_string(
    "output_file", None,
    "Output TF example file (or comma-separated list of files).")

flags.DEFINE_string("vocab_file", None,
                    "The vocabulary file that the BERT model was trained on.")

flags.DEFINE_bool(
    "do_lower_case", True,
    "Whether to lower case the input text. Should be True for uncased "
    "models and False for cased models.")

flags.DEFINE_bool(
    "do_whole_word_mask", False,
    "Whether to use whole word masking rather than per-WordPiece masking.")

flags.DEFINE_integer("max_seq_length", 128, "Maximum sequence length.")

flags.DEFINE_integer("max_predictions_per_seq", 20,
                     "Maximum number of masked LM predictions per sequence.")

flags.DEFINE_integer("random_seed", 12345, "Random seed for data generation.")

flags.DEFINE_integer(
    "dupe_factor", 10,
    "Number of times to duplicate the input data (with different masks).")

flags.DEFINE_float("masked_lm_prob", 0.15, "Masked LM probability.")

flags.DEFINE_float(
    "short_seq_prob", 0.1,
    "Probability of creating sequences which are shorter than the "
    "maximum length.")


class TrainingInstance(object):
  """A single training instance (sentence pair)."""

  def __init__(self, tokens, segment_ids, masked_lm_positions, masked_lm_labels,
               is_random_next):
    self.tokens = tokens
    self.segment_ids = segment_ids
    self.is_random_next = is_random_next
    self.masked_lm_positions = masked_lm_positions
    self.masked_lm_labels = masked_lm_labels

  def __str__(self):
    s = ""
    s += "tokens: %s\n" % (" ".join(
        [tokenization.printable_text(x) for x in self.tokens]))
    s += "segment_ids: %s\n" % (" ".join([str(x) for x in self.segment_ids]))
    s += "is_random_next: %s\n" % self.is_random_next
    s += "masked_lm_positions: %s\n" % (" ".join(
        [str(x) for x in self.masked_lm_positions]))
    s += "masked_lm_labels: %s\n" % (" ".join(
        [tokenization.printable_text(x) for x in self.masked_lm_labels]))
    s += "\n"
    return s

  def __repr__(self):
    return self.__str__()


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)
    #* 保证已经padding了足够多了
    assert len(input_ids) == max_seq_length
    assert len(input_mask) == 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)
    #* 如果小于一个句子最大的mask预测量，则
    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
    
    '''
    input_ids是把tokens转换成词汇表对应的索引(ids)
    input_mask表示哪些位置是有真实字符的(用1表示)，哪些位置是padding，是1维列表
    segment_ids表示A句和B句的对应关系
    masked_lm_positions表示哪些位置被掩码了，记录token中的索引，格式为[20,32,......]
    masked_lm_ids表示被掩码的索引对应的label，将该label转换为词汇表中的ids记录
    masked_lm_weights表示有多少位置被掩码了，列表刚开始部分为1的值证明是有效掩码信息
    next_sentence_label表示B句是否是正常的A句后面的句子
    '''
    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 create_int_feature(values):
  feature = tf.train.Feature(int64_list=tf.train.Int64List(value=list(values)))
  return feature


def create_float_feature(values):
  feature = tf.train.Feature(float_list=tf.train.FloatList(value=list(values)))
  return feature


def create_training_instances(input_files, tokenizer, max_seq_length,
                              dupe_factor, short_seq_prob, masked_lm_prob,
                              max_predictions_per_seq, rng):
  """Create `TrainingInstance`s from raw text."""
  all_documents = [[]]

  # Input file format:
  # (1) One sentence per line. These should ideally be actual sentences, not
  # entire paragraphs or arbitrary spans of text. (Because we use the
  # sentence boundaries for the "next sentence prediction" task).
  # (2) Blank lines between documents. Document boundaries are needed so
  # that the "next sentence prediction" task doesn't span between documents.
  #* 假设传入了多个文件，每个文件中有多行的句子，则all_documents为二维列表
  #* 第一个维度是文档(根据传入的路径)，第二个维度为该文档第几行，第三维度为每行的token
  for input_file in input_files:
    with tf.gfile.GFile(input_file, "r") as reader:
      while True:
        line = tokenization.convert_to_unicode(reader.readline())
        if not line:
          break
        line = line.strip()

        # Empty lines are used as document delimiters
        if not line:
          all_documents.append([])
        tokens = tokenizer.tokenize(line)
        if tokens:
          #* append单个列表
          all_documents[-1].append(tokens)

  # Remove empty documents
  #* 删除空列表
  all_documents = [x for x in all_documents if x]
  rng.shuffle(all_documents)

  #* 对于每篇文章(document)认为是(dupe_factor)篇文章，一篇文章会重复生成样本(由于有随机数的存在，保证训练内容不会重复)
  #* 利用函数create_instances_from_document生成训练实例
  vocab_words = list(tokenizer.vocab.keys())
  instances = []
  for _ in range(dupe_factor):
    for document_index in range(len(all_documents)):
      instances.extend(
          create_instances_from_document(
              all_documents, document_index, max_seq_length, short_seq_prob,
              masked_lm_prob, max_predictions_per_seq, vocab_words, rng))

  rng.shuffle(instances)
  return instances

#* 从每个文档中生成训练数据instance，实例中包括以下信息：
'''
tokens：词，其中包括A句和B句，存在掩码[MASK]
segement_ids：句子编码，第一句为0，第二句为1
is_random_next：第二句是随机查找，还是为第一句的下文
masked_lm_positions：tokens中被mask的位置，[20,32,...]
masked_lm_labels：tokens中被mask的原来的词
返回值：instances
'''
def create_instances_from_document(
    all_documents, document_index, max_seq_length, short_seq_prob,
    masked_lm_prob, max_predictions_per_seq, vocab_words, rng):
  """Creates `TrainingInstance`s for a single document."""
  document = all_documents[document_index]

  #* Account for [CLS], [SEP], [SEP]，计算每行的最大token数量
  max_num_tokens = max_seq_length - 3

  # We *usually* want to fill up the entire sequence since we are padding
  # to `max_seq_length` anyways, so short sequences are generally wasted
  # computation. However, we *sometimes*
  # (i.e., short_seq_prob == 0.1 == 10% of the time) want to use shorter
  # sequences to minimize the mismatch between pre-training and fine-tuning.
  # The `target_seq_length` is just a rough target however, whereas
  # `max_seq_length` is a hard limit.
  #* short_seq_prob是为了缩小预训练和微调过程的差距，以此概率产生小于max_seq_length的训练数据
  #* target_seq_length只是一个粗略的目标，max_seq_length才是硬性的指标上限。但是后文都用target_seq_length判断
  target_seq_length = max_num_tokens
  #* 进行随机数的判断，如果判断小于该概率了，则该篇文章的所有句子的长度上限都会随机一个(2,max_length)的值
  if rng.random() < short_seq_prob:
    target_seq_length = rng.randint(2, max_num_tokens)

  # We DON'T just concatenate all of the tokens from a document into a long
  # sequence and choose an arbitrary split point because this would make the
  # next sentence prediction task too easy. Instead, we split the input into
  # segments "A" and "B" based on the actual "sentences" provided by the user
  # input.
  instances = []
  current_chunk = []
  current_length = 0
  i = 0
  #* document是二维列表，第一维度为哪行，第二维度为该行对应token
  #* 循环document中的每一行(每个句子)
  while i < len(document):
    segment = document[i]
    #* current_chunk也是二位列表，第一维度为行，第二维度为该行对应token，其记录当前处理的所有行信息(可能包含不止一行，由下面294的if决定)
    current_chunk.append(segment)
    current_length += len(segment)
    #* 当循环到了该文章的最后一行或已经扩展的token数大于了当前每句的最大长度，进行截断区分句A和句B
    #* 注意，这里的i、a_end等都表示的是行索引，表示哪一行句子要怎样怎样，每个句子中还有多个token，这里考虑的是将句子直接加入某一列表中
    if i == len(document) - 1 or current_length >= target_seq_length:
      if current_chunk:
        # `a_end` is how many segments from `current_chunk` go into the `A`
        # (first) sentence.
        #* a_end表示NSP任务中句A的末尾位置索引(这里是行索引)
        a_end = 1
        #* 如果现在的current_chunk中包含行数≥2，那么进行随机截断
        if len(current_chunk) >= 2:
          a_end = rng.randint(1, len(current_chunk) - 1)
        
        #* 获取句A的全部token，以一维列表的形式记录
        tokens_a = []
        for j in range(a_end):
          #* extend直接将current_chunk[j]列表中元素添加到tokens_a列表中，所以tokens_a为一维列表
          tokens_a.extend(current_chunk[j])

        #* 获取句B的全部token，可能是真的句A的后面，或者是其他文章的某个长度的句子，具体看is_random_next的值
        #* 将对应的句B结果(token)放tokens_b中
        tokens_b = []
        # Random next
        is_random_next = False
        #* 如果是其他文章的句子
        if len(current_chunk) == 1 or rng.random() < 0.5:
          is_random_next = True
          #* 首先获得句B的最大长度是多少，表示为target_b_length
          target_b_length = target_seq_length - len(tokens_a)

          # This should rarely go for more than one iteration for large
          # corpora. However, just to be careful, we try to make sure that
          # the random document is not the same as the document
          # we're processing.
          #* 随机获得句B在哪篇文章中，循环十次，只要不是和当前文章的索引一致就可以跳出循环
          for _ in range(10):
            random_document_index = rng.randint(0, len(all_documents) - 1)
            if random_document_index != document_index:
              break

          #* 从对应的文章中获得句B，句B的起始位置(从哪一行开始)也是随机的，根据target_b_length获得一段token。
          random_document = all_documents[random_document_index]
          random_start = rng.randint(0, len(random_document) - 1)
          for j in range(random_start, len(random_document)):
            tokens_b.extend(random_document[j])
            if len(tokens_b) >= target_b_length:
              break
          # We didn't actually use these segments so we "put them back" so
          # they don't go to waste.
          #* 挺讲究，因为当前句B是随机找的，所以之前句A的正经后续并没有用到，将索引i向前退，不浪费数据
          #* 这里向前退是按照行来退的，看a_end代表哪行，然后如果判定句B为随机句，则current_chunk中后面的句子都浪费了，要回退
          num_unused_segments = len(current_chunk) - a_end
          i -= num_unused_segments
        # Actual next
        else:
          #* 如果是真实的句A的后面，根据之前的current_chunk将句A之后的每行的token按行依次放入句B(tokens_b)中
          is_random_next = False
          for j in range(a_end, len(current_chunk)):
            tokens_b.extend(current_chunk[j])
        
        #* 观察句A和句B的长度和是否大于最大约定长度，如果是则逐个删除A或B的首或尾。
        truncate_seq_pair(tokens_a, tokens_b, max_num_tokens, rng)

        assert len(tokens_a) >= 1
        assert len(tokens_b) >= 1

        #* 下面按照句A和句B产生完整的tokens(包含分隔符)以及分割id(segment_ids)
        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)

        #* 根据tokens输出掩码之后的tokens、掩码位置masked_lm_positions、掩码位置的真实词masked_lm_labels。(都是列表)
        #* masked_lm_prob为产生mask的概率，max_predictions_per_seq为每个数据产生mask的最大值，一条数据的mask数量为样本token数×掩码概率与最大掩码数量中选一个较小的
        (tokens, masked_lm_positions,
         masked_lm_labels) = create_masked_lm_predictions(
             tokens, masked_lm_prob, max_predictions_per_seq, vocab_words, rng)
        #* 将掩码后的token， AB句对应的id，B句是否是A的正常后续还是随机后续，掩码的token索引以及掩码对应位置的label传入训练实例中
        instance = TrainingInstance(
            tokens=tokens,
            segment_ids=segment_ids,
            is_random_next=is_random_next,
            masked_lm_positions=masked_lm_positions,
            masked_lm_labels=masked_lm_labels)
        instances.append(instance)
      current_chunk = []
      current_length = 0
    i += 1

  return instances


MaskedLmInstance = collections.namedtuple("MaskedLmInstance",
                                          ["index", "label"])

'''
每个输入序列，只有最多15%的token被mask，而其中80%的机会被替换成[MASK]，
10%的机会保持原词不变，10%的机会随机替换为字典中的任意词。代码如何实现呢？
先获取每个token的索引位置，然后随机打乱索引位置，接着取前15%的token进行替换即可。
在替换中，再次利用随机函数，实现80%替换为[MASK]等，代码层面利用random函数还是比较巧妙的。
'''
#* 该函数输入当前一个样本的token，掩码概率，每个样本最大掩码的数量(当前样本掩码数量是从token×掩码概率与最大掩码数量中选一个较小的)，词汇表，随机数
#* 输出为掩码后原样本的tokens(列表),对应掩码的位置索引masked_lm_positions(列表)，以及对应掩码的真实值masked_lm_labels(列表)
def create_masked_lm_predictions(tokens, masked_lm_prob,
                                 max_predictions_per_seq, vocab_words, rng):
  """Creates the predictions for the masked LM objective."""

  cand_indexes = []
  for (i, token) in enumerate(tokens):
    #* 不掩盖分割符
    if token == "[CLS]" or token == "[SEP]":
      continue
    # Whole Word Masking means that if we mask all of the wordpieces
    # corresponding to an original word. When a word has been split into
    # WordPieces, the first token does not have any marker and any subsequence
    # tokens are prefixed with ##. So whenever we see the ## token, we
    # append it to the previous set of word indexes.
    #
    # Note that Whole Word Masking does *not* change the training code
    # at all -- we still predict each WordPiece independently, softmaxed
    # over the entire vocabulary.
    if (FLAGS.do_whole_word_mask and len(cand_indexes) >= 1 and
        token.startswith("##")):
      cand_indexes[-1].append(i)
    else:
      cand_indexes.append([i])
      
  #* cand_indexes表示tokens列表中所有token的索引(不包含起始符和分隔符),故意弄成了二维列表(感觉没啥意义)
  #* 将索引顺序随机打乱
  rng.shuffle(cand_indexes)
  #* output_token为最终输出的token列表，其中包括[MASK]
  output_tokens = list(tokens)
  #* masked token数量，从最大mask配置数和seq长度*mask比例中取一个最小数，作为这个seq最终的mask数量
  num_to_predict = min(max_predictions_per_seq,
                       max(1, int(round(len(tokens) * masked_lm_prob))))

  masked_lms = []
  #* covered_indexes存放被mask token的索引位置
  covered_indexes = set()
  #* 因为被打乱顺序了，所以只要取前num_to_predict个索引进行mask就好
  for index_set in cand_indexes:
    if len(masked_lms) >= num_to_predict:
      break
    # If adding a whole-word mask would exceed the maximum number of
    # predictions, then just skip this candidate.
    if len(masked_lms) + len(index_set) > num_to_predict:
      continue
    #* 下面考虑当前要mask的索引是否已经被mask过了，如果是的话则continue，不是则将新的mask索引记录到covered_indexes中
    is_any_index_covered = False
    for index in index_set:
      if index in covered_indexes:
        is_any_index_covered = True
        break
    if is_any_index_covered:
      continue
    
    #* 下面进行mask操作，80%词替换为[MASK],10%替换为原始值，10%替换为词典中随机词
    #* 将掩码后token的结果记录在output_token列表中，并且将掩码的token原索引以及真实label记录在masked_lms中
    for index in index_set:
      covered_indexes.add(index)

      masked_token = None
      # 80% of the time, replace with [MASK]
      if rng.random() < 0.8:
        masked_token = "[MASK]"
      else:
        # 10% of the time, keep original
        if rng.random() < 0.5:
          masked_token = tokens[index]
        # 10% of the time, replace with random word
        else:
          masked_token = vocab_words[rng.randint(0, len(vocab_words) - 1)]

      output_tokens[index] = masked_token
      #* 保存被掩码的token的原索引位置以及对应真实的label
      masked_lms.append(MaskedLmInstance(index=index, label=tokens[index]))
  assert len(masked_lms) <= num_to_predict
  #* 按照下标重排，保证是原来句子中出现的顺序
  masked_lms = sorted(masked_lms, key=lambda x: x.index)

  masked_lm_positions = []
  masked_lm_labels = []
  #* 将被掩码的索引和原来真实的label分别记录
  for p in masked_lms:
    masked_lm_positions.append(p.index)
    masked_lm_labels.append(p.label)

  return (output_tokens, masked_lm_positions, masked_lm_labels)


def truncate_seq_pair(tokens_a, tokens_b, max_num_tokens, rng):
  """Truncates a pair of sequences to a maximum sequence length."""
  #* 如果句A和句B的长度超过了最大长度，那么就需要删除句A或句B的首或尾token
  while True:
    total_length = len(tokens_a) + len(tokens_b)
    if total_length <= max_num_tokens:
      break

    trunc_tokens = tokens_a if len(tokens_a) > len(tokens_b) else tokens_b
    assert len(trunc_tokens) >= 1

    # We want to sometimes truncate from the front and sometimes from the
    # back to add more randomness and avoid biases.
    if rng.random() < 0.5:
      del trunc_tokens[0]
    else:
      trunc_tokens.pop()


def main(_):
  tf.logging.set_verbosity(tf.logging.INFO)

  tokenizer = tokenization.FullTokenizer(
      vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)

  #* input_files就是文件路径组成的列表，表示有多篇文章
  input_files = []
  for input_pattern in FLAGS.input_file.split(","):
    input_files.extend(tf.gfile.Glob(input_pattern))

  tf.logging.info("*** Reading from input files ***")
  for input_file in input_files:
    tf.logging.info("  %s", input_file)

  rng = random.Random(FLAGS.random_seed)
  #* 产生训练实例，方便后面写入，根据相关参数产生掩码的token等信息
  instances = create_training_instances(
      input_files, tokenizer, FLAGS.max_seq_length, FLAGS.dupe_factor,
      FLAGS.short_seq_prob, FLAGS.masked_lm_prob, FLAGS.max_predictions_per_seq,
      rng)

  output_files = FLAGS.output_file.split(",")
  tf.logging.info("*** Writing to output files ***")
  for output_file in output_files:
    tf.logging.info("  %s", output_file)
  '''
  input_ids是把tokens转换成词汇表对应的索引(ids)
  input_mask表示哪些位置是有真实字符的(用1表示)，哪些位置是padding，是1维列表
  segment_ids表示A句和B句的对应关系
  masked_lm_positions表示哪些位置被掩码了，记录token中的索引，格式为[20,32,......]
  masked_lm_ids表示被掩码的索引对应的label，将该label转换为词汇表中的ids记录
  masked_lm_weights表示有多少位置被掩码了，列表刚开始部分为1的值证明是有效掩码信息
  next_sentence_label表示B句是否是正常的A句后面的句子
  '''
  write_instance_to_example_files(instances, tokenizer, FLAGS.max_seq_length,
                                  FLAGS.max_predictions_per_seq, output_files)


if __name__ == "__main__":
  flags.mark_flag_as_required("input_file")
  flags.mark_flag_as_required("output_file")
  flags.mark_flag_as_required("vocab_file")
  tf.app.run()

如果读者在阅读过程中有任何问题，希望在评论区和我进行交流！本人是NLP小白，也希望与各位大佬交流讨论！