基于bert的文本表征向量embedding模型由ckpt转成pb

# -*- coding: utf-8 -*-
"""
Created on Sun Apr 28 10:20:04 2019

@author: wumingshi
"""

#import contextlib
import json
import os
from enum import Enum
from termcolor import colored
import sys
import modeling
import logging
import tensorflow as tf
import argparse
from tokenization import FullTokenizer, validate_case_matches_checkpoint
os.environ['CUDA_VISIBLE_DEVICES']='0'

def set_logger(context, verbose=False):
    if os.name == 'nt':  # for Windows
        return NTLogger(context, verbose)

    logger = logging.getLogger(context)
    logger.setLevel(logging.DEBUG if verbose else logging.INFO)
    formatter = logging.Formatter(
        '%(levelname)-.1s:' + context + ':[%(filename).3s:%(funcName).3s:%(lineno)3d]:%(message)s', datefmt=
        '%m-%d %H:%M:%S')
    console_handler = logging.StreamHandler()
    console_handler.setLevel(logging.DEBUG if verbose else logging.INFO)
    console_handler.setFormatter(formatter)
    logger.handlers = []
    logger.addHandler(console_handler)
    return logger

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
    
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

def convert_single_example(ex_index, example, label_list, max_seq_length,
                           tokenizer):
  """Converts a single `InputExample` into a single `InputFeatures`."""

  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