run_classifier.py逐行注释
- 本文为BERT的run_classifier.py模块,即分类模块,进行逐行注释。
run_classifier.py
头文件
"""BERT finetuning runner."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import csv
import os
import modeling
import optimization
import tokenization
import tensorflow as tf
参数定义
flags = tf.flags
FLAGS = flags.FLAGS
## Required parameters
flags.DEFINE_string(
"data_dir", None,
"The input data dir. Should contain the .tsv files (or other data files) "
"for the task.")
flags.DEFINE_string(
"bert_config_file", None,
"The config json file corresponding to the pre-trained BERT model. "
"This specifies the model architecture.")
flags.DEFINE_string("task_name", None, "The name of the task to train.")
flags.DEFINE_string("vocab_file", None,
"The vocabulary file that the BERT model was trained on.")
flags.DEFINE_string(
"output_dir", None,
"The output directory where the model checkpoints will be written.")
## Other parameters
flags.DEFINE_string(
"init_checkpoint", None,
"Initial checkpoint (usually from a pre-trained BERT model).")
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_integer(
"max_seq_length", 128,
"The maximum total input sequence length after WordPiece tokenization. "
"Sequences longer than this will be truncated, and sequences shorter "
"than this will be padded.")
flags.DEFINE_bool("do_train", False, "Whether to run training.")
flags.DEFINE_bool("do_eval", False, "Whether to run eval on the dev set.")
flags.DEFINE_bool(
"do_predict", False,
"Whether to run the model in inference mode on the test set.")
flags.DEFINE_integer("train_batch_size", 32, "Total batch size for training.")
flags.DEFINE_integer("eval_batch_size", 8, "Total batch size for eval.")
flags.DEFINE_integer("predict_batch_size", 8, "Total batch size for predict.")
flags.DEFINE_float("learning_rate", 5e-5, "The initial learning rate for Adam.")
flags.DEFINE_float("num_train_epochs", 3.0,
"Total number of training epochs to perform.")
flags.DEFINE_float(
"warmup_proportion", 0.1,
"Proportion of training to perform linear learning rate warmup for. "
"E.g., 0.1 = 10% of training.")
flags.DEFINE_integer("save_checkpoints_steps", 1000,
"How often to save the model checkpoint.")
flags.DEFINE_integer("iterations_per_loop", 1000,
"How many steps to make in each estimator call.")
flags.DEFINE_bool("use_tpu", False, "Whether to use TPU or GPU/CPU.")
tf.flags.DEFINE_string(
"tpu_name", None,
"The Cloud TPU to use for training. This should be either the name "
"used when creating the Cloud TPU, or a grpc://ip.address.of.tpu:8470 "
"url.")
tf.flags.DEFINE_string(
"tpu_zone", None,
"[Optional] GCE zone where the Cloud TPU is located in. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
tf.flags.DEFINE_string(
"gcp_project", None,
"[Optional] Project name for the Cloud TPU-enabled project. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
tf.flags.DEFINE_string("master", None, "[Optional] TensorFlow master URL.")
flags.DEFINE_integer(
"num_tpu_cores", 8,
"Only used if `use_tpu` is True. Total number of TPU cores to use.")
tf.flags是tf.app.flags的新版,使用方法见 https://abhisheksaurabh1985.github.io/2017-12-30-flags-in-python-tf 。flags的引入可以让使用者在cmd中直接操作自定义参数,比如:
python run_classifier.py --do_train True
flags的引入还可以在程序全局使用FLAGS.do_train返回该参数的值。
可自定义的参数如下:
- 输入文件目录(训练,验证,测试用)
- BERT配置文件位置(json格式)
- 任务名称(对应不同的测试数据集)
- 词汇文件位置(txt格式)
- 输出结果目录(将会包括一些checkpoint)
- 初始checkpoint(可继承自预训练BERT模型)
- 单样本序列长度最大值定义(默认128,超出截断,不足补足)
- 训练轮赋能
- 验证轮赋能
- 预测轮赋能
- 训练轮并行样本数(默认32)
- 验证轮并行样本数(默认8)
- 预测轮并行样本数(默认8)
- 学习率(默认5e-5)
- 训练轮数(默认3)
- 热身比例(默认0.1)
- checkpoints保存频率(默认1000步保存一次)
- iterations_per_loop指使用TPU时每轮循环的迭代步数
- 6个TPU相关的配置
InputExample
class InputExample(object):
"""A single training/test example for simple sequence classification."""
def __init__(self, guid, text_a, text_b=None, label=None):
"""Constructs a InputExample.
Args:
guid: Unique id for the example.
text_a: string. The untokenized text of the first sequence. For single
sequence tasks, only this sequence must be specified.
text_b: (Optional) string. The untokenized text of the second sequence.
Only must be specified for sequence pair tasks.
label: (Optional) string. The label of the example. This should be
specified for train and dev examples, but not for test examples.
"""
self.guid = guid
self.text_a = text_a
self.text_b = text_b
self.label = label
该类用于初始化输入样本。 其中,__init__的四个参数:
- guid:样本的唯一识别id
- text_a:sentence-pair的首句
- text_b:sentence-pair的次句(可以没有)
- label:该样本的标签/真值。
PaddingInputExample
class PaddingInputExample(object):
"""Fake example so the num input examples is a multiple of the batch size.
When running eval/predict on the TPU, we need to pad the number of examples
to be a multiple of the batch size, because the TPU requires a fixed batch
size. The alternative is to drop the last batch, which is bad because it means
the entire output data won't be generated.
We use this class instead of `None` because treating `None` as padding
battches could cause silent errors.
"""
当样本数不够一个batch时,该类用来充数(padding)。
InputFeatures
class InputFeatures(object):
"""A single set of features of data."""
def __init__(self,
input_ids,
input_mask,
segment_ids,
label_id,
is_real_example=True):
self.input_ids = input_ids
self.input_mask = input_mask
self.segment_ids = segment_ids
self.label_id = label_id
self.is_real_example = is_real_example
该类用于初始化输入特征,将输入的raw数据转换成BERT可用的数据。 其中__init__的5个参数:
- input_ids:输入id,模型会在之后将token通过vocab.txt转化成整数id。
- input_mask:输入掩码,用一个包含1,0的列表指示/真token/伪token(padding的token)/。
- segment_ids:段落id,用一个包含0,1的列表指示/首句/次句/。(BERT中的E_a、E_b)
- label_id:标签id,用[0, n-1]编号n种标签。
- is_real_example:指示是否为真样本的布尔值(存在充数的样本)。
DataProcessor
class DataProcessor(object):
"""Base class for data converters for sequence classification data sets."""
def get_train_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the train set."""
raise NotImplementedError()
def get_dev_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the dev set."""
raise NotImplementedError()
def get_test_examples(self, data_dir):
"""Gets a collection of `InputExample`s for prediction."""
raise NotImplementedError()
def get_labels(self):
"""Gets the list of labels for this data set."""
raise NotImplementedError()
@classmethod
def _read_tsv(cls, input_file, quotechar=None):
"""Reads a tab separated value file."""
with tf.gfile.Open(input_file, "r") as f:
reader = csv.reader(f, delimiter="\t", quotechar=quotechar)
lines = []
for line in reader:
lines.append(line)
return lines
该类是之后出现的多个特化数据处理器的父类。 定义的几个方法:
- get_train_examples:用于特化数据处理器获取训练集样本。
- get_dev_examples:用于特化数据处理器获取验证集样本。
- get_test_examples:用于特化数据处理器获取测试集样本。
- get_labels:用于特化数据处理器获取标签。
- _read_tsv:用于特化数据处理器将tsv/csv类型文件转换为列表。
MrpcProcessor
class MrpcProcessor(DataProcessor):
"""Processor for the MRPC data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev.tsv")), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue #跳过第一行,第一行包含表头信息,不包含样本。
guid = "%s-%s" % (set_type, i)
text_a = tokenization.convert_to_unicode(line[3])
text_b = tokenization.convert_to_unicode(line[4])
if set_type == "test":
label = "0"
else:
label = tokenization.convert_to_unicode(line[0])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
该类用于文件输入并将数据转化为class InputExample(object),并写入列表examples。该类是DataProcessor的子类,继承DataProcessor的相关方法。
类似的,还有: class XnliProcessor(DataProcessor) class MnliProcessor(DataProcessor) class ColaProcessor(DataProcessor)
在我们需要处理自己的数据集时,需要自行创建一个自己的数据处理器,并根据数据集的格式对数据进行处理得到InputExample类需要的四个参数guid=guid, text_a=text_a, text_b=text_b, label=label。
convert_single_example
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
# 若输入的example为充数的样本,
# 则将InputFeatures的前4个参数设为相应长度的0的列表,将is_real_example设为假。
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)
# 建立一个{[标签]: 标签序号}的字典
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
# 将首句和次句Tokenize并截短
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"
# 超过阈值则首句次句轮流pop一个token
_truncate_seq_pair(tokens_a, tokens_b, max_seq_length - 3)
else:
# Account for [CLS] and [SEP] with "- 2"
# 超过阈值则直接截短取前端
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 = []
# 为token添加[CLS]以及[SEP]
# 若存在句子对,则将两个句子级联为tokens。
# 为segment_ids根据句子的前后添加指示,0表示首句,1表示次句。
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)
# 将tokens通过vocab.txt转换为整数input_ids
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.
# 用掩码1表示真token,0表示充数的token。
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
# 若input_ids长度小于max_seq_length,则用0充数。
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
# 断言函数判断是否相等,否则raise错误
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
# 从字典label_map中找到example.label的序号id
label_id = label_map[example.label]
# 打印所有样本中前5个样本的信息
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: %s (id = %d)" % (example.label, label_id))
# 返回包含5个变量的InputFeatures类
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id,
is_real_example=True)
return feature
该类用于将数据处理器返回的多个InputExample类中的单个InputExample转换为InputFeatures类。
file_based_convert_examples_to_features
def file_based_convert_examples_to_features(
examples, label_list, max_seq_length, tokenizer, output_file):
"""Convert a set of `InputExample`s to a TFRecord file."""
# TFRecordWriter初始化
writer = tf.python_io.TFRecordWriter(output_file)
# 循环写入每个样本的features
for (ex_index, example) in enumerate(examples):
# 进度条(每10000提示一次)
if ex_index % 10000 == 0:
tf.logging.info("Writing example %d of %d" % (ex_index, len(examples)))
# 将InputExample转换为InputFeatures
feature = convert_single_example(ex_index, example, label_list,
max_seq_length, tokenizer)
# 建立整型的特征集,特征集以字典形式存在:
# int64_list {
# value: 0
# value: 1
# value: ...
# }
def create_int_feature(values):
f = tf.train.Feature(int64_list=tf.train.Int64List(value=list(values)))
return f
# 建立一个有序字典
features = collections.OrderedDict()
# 建立5个整型特征集
features["input_ids"] = create_int_feature(feature.input_ids)
features["input_mask"] = create_int_feature(feature.input_mask)
features["segment_ids"] = create_int_feature(feature.segment_ids)
features["label_ids"] = create_int_feature([feature.label_id])
features["is_real_example"] = create_int_feature(
[int(feature.is_real_example)])
# 建立嵌套字典存储features
# features {
# feature {
# key: "input_ids"
# value {
# int64_list {
# value: 0
# value: 1
# }
# }
# }
# feature {
# key: "input_mask"
# value {
# int64_list {
# value: 0
# value: 0
# }
# }
# }
# }
tf_example = tf.train.Example(features=tf.train.Features(feature=features))
#序列化之后(转换为二进制流)写入TFRecord
writer.write(tf_example.SerializeToString())
writer.close()
该函数用于将每个样本的特征写入TFRecord留底。 关于TFRecord,我找到了一篇比较简洁明了的博文 https://www.cnblogs.com/yanshw/p/12419616.html 。
file_based_input_fn_builder
def file_based_input_fn_builder(input_file, seq_length, is_training,
drop_remainder):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
name_to_features = {
"input_ids": tf.FixedLenFeature([seq_length], tf.int64),
"input_mask": tf.FixedLenFeature([seq_length], tf.int64),
"segment_ids": tf.FixedLenFeature([seq_length], tf.int64),
"label_ids": tf.FixedLenFeature([], tf.int64),
"is_real_example": tf.FixedLenFeature([], tf.int64),
}
def _decode_record(record, name_to_features):
"""Decodes a record to a TensorFlow example."""
example = tf.parse_single_example(record, name_to_features)
# tf.Example only supports tf.int64, but the TPU only supports tf.int32.
# So cast all int64 to int32.
for name in list(example.keys()):
t = example[name]
if t.dtype == tf.int64:
t = tf.to_int32(t)
example[name] = t
return example
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
# For training, we want a lot of parallel reading and shuffling.
# For eval, we want no shuffling and parallel reading doesn't matter.
d = tf.data.TFRecordDataset(input_file)
if is_training:
d = d.repeat()
d = d.shuffle(buffer_size=100)
d = d.apply(
tf.contrib.data.map_and_batch(
lambda record: _decode_record(record, name_to_features),
batch_size=batch_size,
drop_remainder=drop_remainder))
return d
return input_fn
* 该函数用于基于文件将输入函数传入TPU,有关TPU的部分先搁置。
_truncate_seq_pair
def _truncate_seq_pair(tokens_a, tokens_b, max_length):
"""Truncates a sequence pair in place to the maximum length."""
# This is a simple heuristic which will always truncate the longer sequence
# one token at a time. This makes more sense than truncating an equal percent
# of tokens from each, since if one sequence is very short then each token
# that's truncated likely contains more information than a longer sequence.
while True:
total_length = len(tokens_a) + len(tokens_b)
# 若总长小于等于最大长度则跳出循环
if total_length <= max_length:
break
# 若总长大于最大长度,
# 则根据句子对的长短依次pop超出的部分直到符合要求。
if len(tokens_a) > len(tokens_b):
tokens_a.pop()
else:
tokens_b.pop()
该函数用于处理句子对输入时,总长超出最大长度的情况。
create_model
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
labels, num_labels, use_one_hot_embeddings):
"""Creates a classification model."""
# 初始化一个BERT模型
model = modeling.BertModel(
config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
# In the demo, we are doing a simple classification task on the entire
# segment.
#
# If you want to use the token-level output, use model.get_sequence_output()
# instead.
# 输出定义为pooling后的输出
output_layer = model.get_pooled_output()
hidden_size = output_layer.shape[-1].value
# 生成一个对应输出层的截断正态分布的参数矩阵(标准差为0.02)
output_weights = tf.get_variable(
"output_weights", [num_labels, hidden_size],
initializer=tf.truncated_normal_initializer(stddev=0.02))
# 偏置设为0
output_bias = tf.get_variable(
"output_bias", [num_labels], initializer=tf.zeros_initializer())
with tf.variable_scope("loss"):
if is_training:
# I.e., 0.1 dropout
# 为了减弱过拟合,将dropout的保持概率设为0.9
output_layer = tf.nn.dropout(output_layer, keep_prob=0.9)
# W_t * x + b 输出未归一化的“概率”logits,对应num_labels个label
logits = tf.matmul(output_layer, output_weights, transpose_b=True)
logits = tf.nn.bias_add(logits, output_bias)
# 将digits经过softmax后得到对应num_labels个label的归一化输出
probabilities = tf.nn.softmax(logits, axis=-1)
log_probs = tf.nn.log_softmax(logits, axis=-1)
# 为了计算样本误差,将label进行one-hot编码
one_hot_labels = tf.one_hot(labels, depth=num_labels, dtype=tf.float32)
# 计算每个样本对应标签的误差
per_example_loss = -tf.reduce_sum(one_hot_labels * log_probs, axis=-1)
# 得到平均误差(应该是交叉熵)
loss = tf.reduce_mean(per_example_loss)
# 返回平均误差loss,包含每个对应label误差的样本误差per_example_loss,未归一化的概率logits,归一化后的概率probabilities。
return (loss, per_example_loss, logits, probabilities)
该函数借助model.py的一些参数和方法建立了一个基于BERT模型的encoder + 单层神经网络分类器,是run_classifier.py算法实现的核心框架部分。 关于dropout的源码解释,参考 https://blog.csdn.net/qq_20412595/article/details/82824830 。 关于logits和softmax的理解,参考 https://www.zhihu.com/question/60751553 。
model_fn_builder
def model_fn_builder(bert_config, num_labels, init_checkpoint, learning_rate,
num_train_steps, num_warmup_steps, use_tpu,
use_one_hot_embeddings):
"""Returns `model_fn` closure for TPUEstimator."""
def model_fn(features, labels, mode, params): # pylint: disable=unused-argument
"""The `model_fn` for TPUEstimator."""
tf.logging.info("*** Features ***")
for name in sorted(features.keys()):
tf.logging.info(" name = %s, shape = %s" % (name, features[name].shape))
input_ids = features["input_ids"]
input_mask = features["input_mask"]
segment_ids = features["segment_ids"]
label_ids = features["label_ids"]
is_real_example = None
if "is_real_example" in features:
is_real_example = tf.cast(features["is_real_example"], dtype=tf.float32)
else:
is_real_example = tf.ones(tf.shape(label_ids), dtype=tf.float32)
is_training = (mode == tf.estimator.ModeKeys.TRAIN)
(total_loss, per_example_loss, logits, probabilities) = create_model(
bert_config, is_training, input_ids, input_mask, segment_ids, label_ids,
num_labels, use_one_hot_embeddings)
tvars = tf.trainable_variables()
initialized_variable_names = {}
scaffold_fn = None
if init_checkpoint:
(assignment_map, initialized_variable_names
) = modeling.get_assignment_map_from_checkpoint(tvars, init_checkpoint)
if use_tpu:
def tpu_scaffold():
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
return tf.train.Scaffold()
scaffold_fn = tpu_scaffold
else:
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
tf.logging.info("**** Trainable Variables ****")
for var in tvars:
init_string = ""
if var.name in initialized_variable_names:
init_string = ", *INIT_FROM_CKPT*"
tf.logging.info(" name = %s, shape = %s%s", var.name, var.shape,
init_string)
output_spec = None
if mode == tf.estimator.ModeKeys.TRAIN:
train_op = optimization.create_optimizer(
total_loss, learning_rate, num_train_steps, num_warmup_steps, use_tpu)
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode,
loss=total_loss,
train_op=train_op,
scaffold_fn=scaffold_fn)
elif mode == tf.estimator.ModeKeys.EVAL:
def metric_fn(per_example_loss, label_ids, logits, is_real_example):
predictions = tf.argmax(logits, axis=-1, output_type=tf.int32)
accuracy = tf.metrics.accuracy(
labels=label_ids, predictions=predictions, weights=is_real_example)
loss = tf.metrics.mean(values=per_example_loss, weights=is_real_example)
return {
"eval_accuracy": accuracy,
"eval_loss": loss,
}
eval_metrics = (metric_fn,
[per_example_loss, label_ids, logits, is_real_example])
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode,
loss=total_loss,
eval_metrics=eval_metrics,
scaffold_fn=scaffold_fn)
else:
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode,
predictions={"probabilities": probabilities},
scaffold_fn=scaffold_fn)
return output_spec
return model_fn
* 该函数用于建立BERT分类模型并将参数传入TPU,有关TPU的部分先搁置。
input_fn_builder
def input_fn_builder(features, seq_length, is_training, drop_remainder):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
all_input_ids = []
all_input_mask = []
all_segment_ids = []
all_label_ids = []
for feature in features:
all_input_ids.append(feature.input_ids)
all_input_mask.append(feature.input_mask)
all_segment_ids.append(feature.segment_ids)
all_label_ids.append(feature.label_id)
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
num_examples = len(features)
# This is for demo purposes and does NOT scale to large data sets. We do
# not use Dataset.from_generator() because that uses tf.py_func which is
# not TPU compatible. The right way to load data is with TFRecordReader.
d = tf.data.Dataset.from_tensor_slices({
"input_ids":
tf.constant(
all_input_ids, shape=[num_examples, seq_length],
dtype=tf.int32),
"input_mask":
tf.constant(
all_input_mask,
shape=[num_examples, seq_length],
dtype=tf.int32),
"segment_ids":
tf.constant(
all_segment_ids,
shape=[num_examples, seq_length],
dtype=tf.int32),
"label_ids":
tf.constant(all_label_ids, shape=[num_examples], dtype=tf.int32),
})
if is_training:
d = d.repeat()
d = d.shuffle(buffer_size=100)
d = d.batch(batch_size=batch_size, drop_remainder=drop_remainder)
return d
return input_fn
* 该函数用于不基于文件将输入函数传入TPU,有关TPU的部分先搁置。
convert_examples_to_features
def convert_examples_to_features(examples, label_list, max_seq_length,
tokenizer):
"""Convert a set of `InputExample`s to a list of `InputFeatures`."""
features = []
for (ex_index, example) in enumerate(examples):
if ex_index % 10000 == 0:
tf.logging.info("Writing example %d of %d" % (ex_index, len(examples)))
feature = convert_single_example(ex_index, example, label_list,
max_seq_length, tokenizer)
features.append(feature)
return features
该函数用于批量将InputExample类转换为InputFeatures类。
\main:
def main(_):
# 将日志信息打印
tf.logging.set_verbosity(tf.logging.INFO)
# 建立关于数据处理器的字典
processors = {
"cola": ColaProcessor,
"mnli": MnliProcessor,
"mrpc": MrpcProcessor,
"xnli": XnliProcessor,
}
# 验证do_lower_case的选择是否与checkpoint匹配
tokenization.validate_case_matches_checkpoint(FLAGS.do_lower_case,
FLAGS.init_checkpoint)
# 保证至少在参数部分选择了train/eval/predict三个操作中的一种,否则报错
if not FLAGS.do_train and not FLAGS.do_eval and not FLAGS.do_predict:
raise ValueError(
"At least one of `do_train`, `do_eval` or `do_predict' must be True.")
# 输入BERT的配置文件
bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)
# 如果参数部分设定的最大序列长度大于BERT的最大position embedding长度,则报错
if FLAGS.max_seq_length > bert_config.max_position_embeddings:
raise ValueError(
"Cannot use sequence length %d because the BERT model "
"was only trained up to sequence length %d" %
(FLAGS.max_seq_length, bert_config.max_position_embeddings))
# 创建参数设定的输出目录
tf.gfile.MakeDirs(FLAGS.output_dir)
# 传入任务名称
task_name = FLAGS.task_name.lower()
# 若任务不在数据处理器的字典中,则报错
if task_name not in processors:
raise ValueError("Task not found: %s" % (task_name))
# 选择对应任务的数据处理器
processor = processors[task_name]()
# 传入数据处理器处理后的标签列表
label_list = processor.get_labels()
# 选择使用FullTokenizer分词,选择词典和是否小写
tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
# TPU参数初始化
tpu_cluster_resolver = None
if FLAGS.use_tpu and FLAGS.tpu_name:
tpu_cluster_resolver = tf.contrib.cluster_resolver.TPUClusterResolver(
FLAGS.tpu_name, zone=FLAGS.tpu_zone, project=FLAGS.gcp_project)
is_per_host = tf.contrib.tpu.InputPipelineConfig.PER_HOST_V2
# TPU配置设定
run_config = tf.contrib.tpu.RunConfig(
cluster=tpu_cluster_resolver,
master=FLAGS.master,
model_dir=FLAGS.output_dir,
save_checkpoints_steps=FLAGS.save_checkpoints_steps,
tpu_config=tf.contrib.tpu.TPUConfig(
iterations_per_loop=FLAGS.iterations_per_loop,
num_shards=FLAGS.num_tpu_cores,
per_host_input_for_training=is_per_host))
train_examples = None
num_train_steps = None
num_warmup_steps = None
# 训练部分初始化参数
if FLAGS.do_train:
# 通过数据处理器的函数获得训练样本
train_examples = processor.get_train_examples(FLAGS.data_dir)
# 实际训练步数 = 训练样本总数 / 批大小 * 训练轮数
num_train_steps = int(
len(train_examples) / FLAGS.train_batch_size * FLAGS.num_train_epochs)
# warmup步数 = 实际训练步数 * 热身比例
num_warmup_steps = int(num_train_steps * FLAGS.warmup_proportion)
# BERT建模
model_fn = model_fn_builder(
bert_config=bert_config,
num_labels=len(label_list),
init_checkpoint=FLAGS.init_checkpoint,
learning_rate=FLAGS.learning_rate,
num_train_steps=num_train_steps,
num_warmup_steps=num_warmup_steps,
use_tpu=FLAGS.use_tpu,
use_one_hot_embeddings=FLAGS.use_tpu)
# If TPU is not available, this will fall back to normal Estimator on CPU
# or GPU.
# 配置TPUEstimator(若没有TPU,则会使用CPU/GPU)
estimator = tf.contrib.tpu.TPUEstimator(
use_tpu=FLAGS.use_tpu,
model_fn=model_fn,
config=run_config,
train_batch_size=FLAGS.train_batch_size,
eval_batch_size=FLAGS.eval_batch_size,
predict_batch_size=FLAGS.predict_batch_size)
# 训练部分主题
if FLAGS.do_train:
# 指定输出训练数据的TFRecord文件的位置
train_file = os.path.join(FLAGS.output_dir, "train.tf_record")
# 批量将样本转化为包含features的TFRecord文件
file_based_convert_examples_to_features(
train_examples, label_list, FLAGS.max_seq_length, tokenizer, train_file)
# 打印样本总数、批大小、实际训练步数
tf.logging.info("***** Running training *****")
tf.logging.info(" Num examples = %d", len(train_examples))
tf.logging.info(" Batch size = %d", FLAGS.train_batch_size)
tf.logging.info(" Num steps = %d", num_train_steps)
# 基于文件解包并生成传入TPUEstimator的包含features的输入函数
train_input_fn = file_based_input_fn_builder(
input_file=train_file,
seq_length=FLAGS.max_seq_length,
is_training=True,
drop_remainder=True)
# 将features和实际训练步数传入TPUEstimator并开始训练
estimator.train(input_fn=train_input_fn, max_steps=num_train_steps)
# 评估阶段
if FLAGS.do_eval:
# 数据处理器获取评估集样本
eval_examples = processor.get_dev_examples(FLAGS.data_dir)
# 评估集样本个数
num_actual_eval_examples = len(eval_examples)
# 如果使用TPU
if FLAGS.use_tpu:
# TPU requires a fixed batch size for all batches, therefore the number
# of examples must be a multiple of the batch size, or else examples
# will get dropped. So we pad with fake examples which are ignored
# later on. These do NOT count towards the metric (all tf.metrics
# support a per-instance weight, and these get a weight of 0.0).
# 如果(样本个数 / 批大小)不整除,则做padding充数使其可被整除
while len(eval_examples) % FLAGS.eval_batch_size != 0:
eval_examples.append(PaddingInputExample())
# 建立评估TFRecord文件的目录
eval_file = os.path.join(FLAGS.output_dir, "eval.tf_record")
# 批量将样本转化为包含features的TFRecord文件
file_based_convert_examples_to_features(
eval_examples, label_list, FLAGS.max_seq_length, tokenizer, eval_file)
# 打印实际样本个数、padding的样本个数、批大小
tf.logging.info("***** Running evaluation *****")
tf.logging.info(" Num examples = %d (%d actual, %d padding)",
len(eval_examples), num_actual_eval_examples,
len(eval_examples) - num_actual_eval_examples)
tf.logging.info(" Batch size = %d", FLAGS.eval_batch_size)
# This tells the estimator to run through the entire set.
eval_steps = None
# However, if running eval on the TPU, you will need to specify the
# number of steps.
# 如果使用TPU,需要确定评估阶段的步数。
if FLAGS.use_tpu:
# 断言函数确保(样本个数 / 批大小)整除,否则报错。
assert len(eval_examples) % FLAGS.eval_batch_size == 0
# 实际评估步数即为整除后的得到的整数
# 可以看到这里写程序的同学十分谨慎,既用了整除“//”,又强制转换了类型为整型
eval_steps = int(len(eval_examples) // FLAGS.eval_batch_size)
# 如果使用TPU则用drop_remainder,否则不用。
eval_drop_remainder = True if FLAGS.use_tpu else False
# 基于文件解包并生成传入TPUEstimator的包含features的输入函数
eval_input_fn = file_based_input_fn_builder(
input_file=eval_file,
seq_length=FLAGS.max_seq_length,
is_training=False,
drop_remainder=eval_drop_remainder)
# 将features和实际评估步数传入TPUEstimator并开始评估
result = estimator.evaluate(input_fn=eval_input_fn, steps=eval_steps)
# 创建评估结果的txt文档
output_eval_file = os.path.join(FLAGS.output_dir, "eval_results.txt")
# 向评估结果的txt文档写入result返回的四个参数
# eval_accuracy、eval_loss、global_step、loss
# *目前不知道这几个参数的具体含义,需要进一步查看estimator.evaluate返回的参数是如何定义的。
with tf.gfile.GFile(output_eval_file, "w") as writer:
tf.logging.info("***** Eval results *****")
for key in sorted(result.keys()):
# 屏幕打印
tf.logging.info(" %s = %s", key, str(result[key]))
# 写入txt文件
writer.write("%s = %s\n" % (key, str(result[key])))
# 预测阶段
if FLAGS.do_predict:
# 传入预测样本
predict_examples = processor.get_test_examples(FLAGS.data_dir)
# 预测样本总数
num_actual_predict_examples = len(predict_examples)
# 如果使用TPU
if FLAGS.use_tpu:
# TPU requires a fixed batch size for all batches, therefore the number
# of examples must be a multiple of the batch size, or else examples
# will get dropped. So we pad with fake examples which are ignored
# later on.
# 若样本总数不能整除batch,则padding样本充数。
while len(predict_examples) % FLAGS.predict_batch_size != 0:
predict_examples.append(PaddingInputExample())
# 创建TFRecord文件目录
predict_file = os.path.join(FLAGS.output_dir, "predict.tf_record")
# 批量将样本转化为包含features的TFRecord文件
file_based_convert_examples_to_features(predict_examples, label_list,
FLAGS.max_seq_length, tokenizer,
predict_file)
# 打印实际样本数、padding样本数、批大小
tf.logging.info("***** Running prediction*****")
tf.logging.info(" Num examples = %d (%d actual, %d padding)",
len(predict_examples), num_actual_predict_examples,
len(predict_examples) - num_actual_predict_examples)
tf.logging.info(" Batch size = %d", FLAGS.predict_batch_size)
# 如果使用TPU则用drop_remainder,否则不用。
predict_drop_remainder = True if FLAGS.use_tpu else False
# 基于文件解包并生成传入TPUEstimator的包含features的输入函数
predict_input_fn = file_based_input_fn_builder(
input_file=predict_file,
seq_length=FLAGS.max_seq_length,
is_training=False,
drop_remainder=predict_drop_remainder)
# 开始预测
result = estimator.predict(input_fn=predict_input_fn)
# 创建预测结果的tsv文件
output_predict_file = os.path.join(FLAGS.output_dir, "test_results.tsv")
# 写入tsv文件
with tf.gfile.GFile(output_predict_file, "w") as writer:
num_written_lines = 0
tf.logging.info("***** Predict results *****")
# 按照标签顺序写入每个预测样本对应标签的概率
# *需要进一步查看estimator.predict返回的数据是如何定义的
for (i, prediction) in enumerate(result):
probabilities = prediction["probabilities"]
if i >= num_actual_predict_examples:
break
output_line = "\t".join(
str(class_probability)
for class_probability in probabilities) + "\n"
writer.write(output_line)
num_written_lines += 1
# 断言函数保证写入tsv的行数与预测样本总数相同,否则报错。
assert num_written_lines == num_actual_predict_examples
主函数部分。
if name == “main”:
# 主函数调用
if __name__ == "__main__":
# 将data_dir、task_name、vocab_file、bert_config_file、output_dir设置为必要参数
flags.mark_flag_as_required("data_dir")
flags.mark_flag_as_required("task_name")
flags.mark_flag_as_required("vocab_file")
flags.mark_flag_as_required("bert_config_file")
flags.mark_flag_as_required("output_dir")
# 该函数会先处理flag之后默认运行main函数
tf.app.run()
flag处理以及主函数调用。 tf.app.run()的源码解释详见 https://blog.csdn.net/helei001/article/details/51859423
def run(main=None):
f = flags.FLAGS
f._parse_flags()
main = main or sys.modules['__main__'].main
sys.exit(main(sys.argv))
后记
run_classifier.py的代码官方给出了982行,我写这篇文章写了两天,前期准备不知道用了多久,因为一直在看这个分类器的代码。说实话准备充分的话看起来还是很轻松的,但几个月之前刚下BERT看这个分类器的代码的时候真的是一头雾水,即便我已经大致了解了BERT的实际工作原理。
还剩下一些关于TPUEstimator参数配置的部分没加注释,还有一些关于评估阶段测试阶段的result返回数据没确认。之后会继续补齐。