classifier.py

# 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.
"""BERT finetuning runner."""

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

from itertools import islice

from scipy import sparse, io

import collections
import csv
import os
import modeling
import optimization
import tokenization
import tensorflow as tf
import numpy as np 
import networkx as nx

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.")

flags.DEFINE_string(
    "label_vocab", None,
    "The file which contains the class labels.")

flags.DEFINE_string(
    "train_file", None,
    "The file which contains the training snippets.")

flags.DEFINE_string(
    "train_labels", None,
    "The file which contains the training labels.")

flags.DEFINE_string(
    "train_adj", None,
    "The file which contains the training adjacency matrices.")

flags.DEFINE_string(
    "eval_file", None,
    "The file which contains the evaluation snippets.")

flags.DEFINE_string(
    "eval_labels", None,
    "The file which contains the evaluation labels.")

flags.DEFINE_string(
    "eval_adj", None,
    "The file which contains the evaluation adjacency matrices.")

flags.DEFINE_bool("clean_data", False, "Whether to generate a fresh batch of data.")

flags.DEFINE_integer(
    "max_nb_preds", 128,
    "Max number of predictions to generate.")

## 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_bool(
    "do_embed", False,
    "Whether to run the model in inference mode on the test set.")

flags.DEFINE_bool(
    "sparse_adj", False,
    "Whether to run the model in inference mode on the test set.")

tf.flags.DEFINE_string(
    "adj_prefix", 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.")

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.")

flags.DEFINE_bool("shuffle", False, "Whether to run shuffle experiment.")

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.")


class InputExample(object):
  """A single training/test example for simple sequence classification."""

  def __init__(self, guid, text_a, adjacency, 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.
      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.label = label
    self.adjacency = adjacency


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.
  """


class InputFeatures(object):
  """A single set of features of data."""

  def __init__(self,
               input_ids,
               adjacency,
               input_mask,
               segment_ids,
               label_id,
               loss_mask=None,
               is_real_example=True):
    self.input_ids = input_ids
    self.input_mask = input_mask
    self.loss_mask = loss_mask 
    self.adjacency = adjacency
    self.segment_ids = segment_ids
    self.label_id = label_id
    self.is_real_example = is_real_example


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()

  def get_adjacency(self, adj_path, suffix="", fname=None, idx=0):
    if FLAGS.sparse_adj:
      return self._read_sparse_adj(adj_path, suffix=suffix, fname=fname, idx=idx)
    else:
      return self._read_adj(adj_path, from_line=idx*(64+2))

  @classmethod
  def _read_tsv(cls, input_file, delim='\t', quotechar=None):
    """Reads a tab separated value file."""
    with tf.gfile.Open(input_file, "r") as f:
      reader = csv.reader(f, delimiter=delim, quotechar=quotechar)
      lines = []
      for line in reader:
        lines.append(line)
      return lines

  @classmethod
  def _read_adj(cls, input_file, from_line=0):
    with tf.gfile.Open(input_file, "r") as f:
      reader = csv.reader(f, delimiter=',')
      lines = []
      for row in islice(reader, from_line, from_line+FLAGS.max_seq_length):
        lines.append([int(r) for r in row])
      return lines

  @classmethod
  def _read_sparse_adj(cls, path, suffix, fname=None, idx=0):
    if fname==None:
       fname = str(idx)+'_'+FLAGS.adj_prefix+suffix+'.mtx'
    m = io.mmread(os.path.join(path, 'adj', fname))
    return list(m.toarray())

  def get_multi_train_examples(self, paths):
    """See base class."""
    snippets = []; labels = []; prefix = []
    for i,(data_path, label_path) in enumerate(paths):
      snippets.append(self._read_tsv(data_path))
      labels.append(self._read_tsv(label_path))
      prefix.append([str(j)+"_"+FLAGS.adj_prefix.split(',')[i] for j in range(len(labels[-1]))])
    snippets = [_ for s in snippets for _ in s]
    labels   = [_ for l in labels for _ in l]
    prefix   = [_ for p in prefix for _ in p]
    return self._create_examples(snippets, labels, "train", prefix)

  def get_multi_test_examples(self, paths):
    """See base class."""
    snippets = []; labels = []; prefix = []
    for i,(data_path, label_path) in enumerate(paths):
      snippets.append(self._read_tsv(data_path))
      labels.append(self._read_tsv(label_path))
      prefix.append([str(j)+"_"+FLAGS.adj_prefix.split(',')[i] for j in range(len(labels[-1]))])
    snippets = [_ for s in snippets for _ in s]
    labels   = [_ for l in labels for _ in l]
    prefix   = [_ for p in prefix for _ in p]
    return self._create_examples(snippets, labels, "test", prefix)

  def get_labels(self, label_path):
    """See base class."""
    l = self._read_tsv(label_path)
    labels = [item for sublist in l for item in sublist]
    return labels

  def get_train_examples(self, data_path, label_path):
    """See base class."""
    snippets = self._read_tsv(data_path, delim=' ')
    labels = self._read_tsv(label_path, delim=' ')
    return self._create_examples(snippets, labels, "train")

  def get_test_examples(self, data_path, label_path):
    """See base class."""
    snippets = self._read_tsv(data_path, delim=' ')
    labels = self._read_tsv(label_path, delim=' ')
    #snippets = self._read_tsv(data_path)
    #labels = self._read_tsv(label_path)
    return self._create_examples(snippets, labels, "test")

class MethodNamingProcessor(DataProcessor):
  """Processor for the Method Naming data set."""
  def _create_examples(self, lines, labels, set_type, prefix=None):
    """Creates examples for the training and dev sets."""
    examples = []
    tokenizer = tokenization.FullTokenizer(
        vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
    
    for (i, (line, label)) in enumerate(zip(lines, labels)):
      guid = "%s-%s" % (set_type, str(i))
      if len(line) > 0:
        text = tokenization.convert_to_unicode(' '.join(line))
      else:
        text  = tokenization.convert_to_unicode(line[0])
      
      label = tokenization.convert_to_unicode(label[0])
      suffix = "_adj" if set_type=="train" else "_adj_val"
      adj_file = FLAGS.train_adj if set_type=="train" else FLAGS.eval_adj
      if prefix == None:
        adj = self.get_adjacency(adj_file, suffix=suffix, idx=i)
      else:
        fname = prefix[i]+suffix+'.mtx'
        adj = self.get_adjacency(adj_file, suffix=suffix, fname=fname)

      if FLAGS.shuffle:
        split_text = np.asarray(text.split(' '))
        shuffle_idx = np.random.permutation(range(1,len(split_text)))
        shuffle_idx = np.insert(shuffle_idx,0,0)
        text = ' '.join(split_text[shuffle_idx])
        adj = np.asarray(adj)
        G = nx.from_numpy_matrix(adj)
        shuffled_adj = nx.adjacency_matrix(G, nodelist=shuffle_idx).todense()
        padded_adj = np.eye(64)
        padded_adj[:len(shuffle_idx), :len(shuffle_idx)] = shuffled_adj

      examples.append(
          InputExample(guid=guid, text_a=text, adjacency=adj, label=label))
    return examples

class VarNamingProcessor(DataProcessor):
  """Processor for the Method Naming data set."""

  def _create_examples(self, lines, labels, set_type, prefix=None):
    """Creates examples for the training and dev sets."""
    examples = []
    tokenizer = tokenization.FullTokenizer(
        vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)

    for (i, (line, label)) in enumerate(zip(lines, labels)):
      if len(label) > 0:
        guid = "%s-%s" % (set_type, str(i))

        text  = tokenization.convert_to_unicode(line[0])
        label = tokenization.convert_to_unicode(label[0])
        
        suffix = "_adj" if set_type=="train" else "_adj_val"
        adj_file = FLAGS.train_adj if set_type=="train" else FLAGS.eval_adj
        if prefix == None:
          adj = self.get_adjacency(adj_file, suffix=suffix, idx=i) 
        else:
          fname = prefix[i]+suffix+'.mtx'
          adj = self.get_adjacency(adj_file, suffix=suffix, fname=fname)

        examples.append(
            InputExample(guid=guid, text_a=text, adjacency=adj, label=label))
    return examples


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,
        adjacency=[[0] * max_seq_length] * 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)

  # 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 = []
  MSK = 4
  #mask_ids = []
  for idx, token in enumerate(tokens_a):
    tokens.append(token)
   # if token == "[MASK]":
   #   mask_ids.append(True)
   # else:
   #   mask_ids.append(False)
    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

  split_label = tokenizer.tokenize(' '.join(example.label.split(',')))

  is_multilabel = len(split_label) > 1 
  if is_multilabel:
    split_label = tokenizer.convert_tokens_to_ids(split_label)
    label_id = []
    msk_count = 0
    for idx, token in enumerate(tokens_a):
      if token == "[MASK]":
        label_id.append(split_label[msk_count])
        msk_count += 1
      else:
        label_id.append(input_ids[idx])
    while len(label_id) < max_seq_length:
      label_id.append(0)
  else:
    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]))
    if is_multilabel:
      tf.logging.info("label: {} (ids = {})".format(example.label, label_id))
      #tf.logging.info("masked - idx: {})".format(mask_ids))
    else:
      tf.logging.info("label: %s (id = %d)" % (example.label, label_id))
    for v in example.adjacency:
      tf.logging.info("adjacency: %s" % " ".join([str(x) for x in v]))

  feature = InputFeatures(
      input_ids=input_ids,
      adjacency=example.adjacency,
      input_mask=input_mask,
      segment_ids=segment_ids,
      label_id=label_id,
      #loss_mask=mask_ids,
      is_real_example=True)
  return feature

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."""

  writer = tf.python_io.TFRecordWriter(output_file)

  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)

    def create_int_feature(values):
      f = tf.train.Feature(int64_list=tf.train.Int64List(value=list(values)))
      return f

    adjacency = feature.adjacency

    features = collections.OrderedDict()
    features["input_ids"] = create_int_feature(feature.input_ids)
    features["adjacency"] = create_int_feature([item for sublist in adjacency for item in sublist])
    features["input_mask"] = create_int_feature(feature.input_mask)
    features["segment_ids"] = create_int_feature(feature.segment_ids)
    if FLAGS.task_name == 'varname':
      features["label_ids"] = create_int_feature(feature.label_id)
    else: 
      features["label_ids"] = create_int_feature([feature.label_id])
    features["is_real_example"] = create_int_feature(
        [int(feature.is_real_example)])
   # features["loss_mask"] = create_int_feature(feature.loss_mask)
    tf_example = tf.train.Example(features=tf.train.Features(feature=features))
    writer.write(tf_example.SerializeToString())
  writer.close()


def file_based_input_fn_builder(input_file, seq_length, is_training,
                                drop_remainder):
  """Creates an `input_fn` closure to be passed to TPUEstimator."""

  if FLAGS.task_name=='varname':
    lids = tf.FixedLenFeature([seq_length], tf.int64)
  else:
    lids = tf.FixedLenFeature([], tf.int64)
  name_to_features = {
      "input_ids": tf.FixedLenFeature([seq_length], tf.int64),
      "adjacency": tf.FixedLenFeature([seq_length*seq_length], tf.int64),
      "input_mask": tf.FixedLenFeature([seq_length], tf.int64),
      "segment_ids": tf.FixedLenFeature([seq_length], tf.int64),
  #    "loss_mask": tf.FixedLenFeature([seq_length], tf.int64), 
      "label_ids": lids, 
      "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


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)
    if total_length <= max_length:
      break
    if len(tokens_a) > len(tokens_b):
      tokens_a.pop()
    else:
      tokens_b.pop()


def create_embedding_model(bert_config, is_training, input_ids, input_mask, adj_mask, segment_ids,
                 labels, num_labels, use_one_hot_embeddings, loss_mask=None):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      adj_mask=adj_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  encoder_output = model.get_sequence_output()

  with tf.variable_scope("loss"):
    attention_output = model.get_attention_output()

    return (encoder_output, attention_output)


def create_model(bert_config, is_training, input_ids, input_mask, adj_mask, segment_ids,
                 labels, num_labels, use_one_hot_embeddings, loss_mask=None):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      adj_mask=adj_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  is_multilabel = FLAGS.task_name == 'varname'

  if is_multilabel:
    output_layer = model.get_sequence_output()
    final_hidden_shape = modeling.get_shape_list(output_layer, expected_rank=3)
    batch_size = final_hidden_shape[0]
    seq_length = final_hidden_shape[1]
    hidden_size = final_hidden_shape[2]
  else:
    output_layer = model.get_pooled_output()
    hidden_size = output_layer.shape[-1].value
 
  output_weights = tf.get_variable(
      "output_weights", [num_labels, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  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
      output_layer = tf.nn.dropout(output_layer, keep_prob=0.9)
    if is_multilabel:
      output_layer = tf.reshape(output_layer, [batch_size*seq_length, hidden_size])
    logits = tf.matmul(output_layer, output_weights, transpose_b=True)
    logits = tf.nn.bias_add(logits, output_bias)

    if is_multilabel:
      logits = tf.reshape(logits, [batch_size, seq_length, num_labels])
      #logits = tf.transpose(logits, [2, 0, 1])
      #logits = tf.reshape(logits, [16, FLAGS.max_seq_length, num_labels])
    probabilities = tf.nn.softmax(logits, axis=-1)
    log_probs = tf.nn.log_softmax(logits, axis=-1)

    one_hot_labels = tf.one_hot(labels, depth=num_labels, dtype=tf.float32)
    per_example_loss = -tf.reduce_sum(one_hot_labels * log_probs, axis=-1)
    #loss_mask = tf.cast(loss_mask, tf.bool)

    if is_multilabel:
      #loss = tf.reduce_sum(tf.boolean_mask(per_example_loss, loss_mask))
      loss = tf.reduce_mean(per_example_loss)
    else:
      loss = tf.reduce_mean(per_example_loss)

    attention_output = model.get_attention_output()

    return (loss, per_example_loss, logits, probabilities, attention_output)


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"]
    #loss_mask = features["loss_mask"] if FLAGS.task_name == "varname" else None
    adjacency = features["adjacency"]
    
    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, attention_output) = create_model(
    #    bert_config, is_training, input_ids, input_mask, adjacency, segment_ids, label_ids,
    #    num_labels, use_one_hot_embeddings, loss_mask)
    if FLAGS.do_embed:
      (encoder_output, attention_output) = create_embedding_model(
          bert_config, is_training, input_ids, input_mask, adjacency, segment_ids, label_ids,
          num_labels, use_one_hot_embeddings)
    else:
      (total_loss, per_example_loss, logits, probabilities, attention_output) = create_model(
          bert_config, is_training, input_ids, input_mask, adjacency, 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)
    elif FLAGS.do_embed:
       output_spec = tf.contrib.tpu.TPUEstimatorSpec(
          mode=mode,
          predictions={
          "encoder_output": encoder_output,
          "attention_output": attention_output
       },
       scaffold_fn=scaffold_fn)
       return output_spec
    else:
      output_spec = tf.contrib.tpu.TPUEstimatorSpec(
          mode=mode,
          predictions={
            "probabilities": probabilities,
            "attention_output": attention_output
          },
          scaffold_fn=scaffold_fn)
    return output_spec

  return model_fn


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

  processors = {
      "methodname": MethodNamingProcessor,
      "varname": VarNamingProcessor,
  }

  tokenization.validate_case_matches_checkpoint(FLAGS.do_lower_case,
                                                FLAGS.init_checkpoint)

  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.")

  bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)

  tf.gfile.MakeDirs(FLAGS.output_dir)

  task_name = FLAGS.task_name.lower()

  if task_name not in processors:
    raise ValueError("Task not found: %s" % (task_name))

  processor = processors[task_name]()

  label_list = processor.get_labels(FLAGS.label_vocab)

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

  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
  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
  if FLAGS.do_train:
    train_file = os.path.join(FLAGS.output_dir, "train.tf_record")
    if FLAGS.clean_data:
      all_train = FLAGS.train_file.split(',')
      all_label = FLAGS.train_labels.split(',')
      nb_files = len(all_train)
      assert len(all_train) == len(all_label)
      if nb_files > 1:
        files = zip(all_train, all_label)
        print(files)
        train_examples = processor.get_multi_train_examples(files)
        print(len(train_examples))
      else:
        train_examples = processor.get_train_examples(FLAGS.train_file, FLAGS.train_labels)
      if FLAGS.num_train_epochs == 0:
         num_train_steps = 1
      else:
        num_train_steps = int(
            len(train_examples) / FLAGS.train_batch_size * FLAGS.num_train_epochs)
      file_based_convert_examples_to_features(
          train_examples, label_list, FLAGS.max_seq_length, tokenizer, train_file)
    else:
      num_train_steps = FLAGS.num_train_epochs
    num_warmup_steps = int(num_train_steps * FLAGS.warmup_proportion)

  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.
  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)

  if FLAGS.do_train:

    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)
    train_input_fn = file_based_input_fn_builder(
        input_file=train_file,
        seq_length=FLAGS.max_seq_length,
        is_training=True,
        drop_remainder=True)
    estimator.train(input_fn=train_input_fn, max_steps=num_train_steps)
 
  if FLAGS.do_eval:
    all_eval = FLAGS.eval_file.split(',')
    all_label = FLAGS.eval_labels.split(',')
    nb_files = len(all_eval)
    assert len(all_eval) == len(all_label)
    if nb_files > 1:
      files = zip(all_eval, all_label)
      print(files)
      eval_examples = processor.get_multi_test_examples(files)
    else:
      eval_examples = processor.get_test_examples(FLAGS.eval_file, FLAGS.eval_labels)
    num_actual_eval_examples = len(eval_examples)
    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).
      while len(eval_examples) % FLAGS.eval_batch_size != 0:
        eval_examples.append(PaddingInputExample())

    eval_file = os.path.join(FLAGS.output_dir, "eval.tf_record")
    file_based_convert_examples_to_features(
        eval_examples, label_list, FLAGS.max_seq_length, tokenizer, eval_file)

    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

    eval_drop_remainder = True if FLAGS.use_tpu else False
    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)

    result = estimator.evaluate(input_fn=eval_input_fn, steps=eval_steps)

    output_eval_file = os.path.join(FLAGS.output_dir, "eval_results.txt")
    with tf.gfile.GFile(output_eval_file, "w") as writer:
      tf.logging.info("***** Eval results *****")
      for key in sorted(result.keys()):
        tf.logging.info("  %s = %s", key, str(result[key]))
        writer.write("%s = %s\n" % (key, str(result[key])))

  if FLAGS.do_predict:
    predict_file = os.path.join(FLAGS.output_dir, "predict.tf_record")
    if FLAGS.clean_data:
      all_eval = FLAGS.eval_file.split(',')
      all_label = FLAGS.eval_labels.split(',')
      nb_files = len(all_eval)
      assert len(all_eval) == len(all_label)
      if nb_files > 1:
        files = zip(all_eval, all_label)
        print(files)
        predict_examples = processor.get_multi_test_examples(files)
      else:
        predict_examples = processor.get_test_examples(FLAGS.eval_file, FLAGS.eval_labels)

     # predict_examples = processor.get_test_examples(FLAGS.eval_file, FLAGS.eval_labels)
      num_actual_predict_examples = len(predict_examples)
      file_based_convert_examples_to_features(predict_examples, label_list,
                                              FLAGS.max_seq_length, tokenizer,
                                              predict_file)
    else:
      num_actual_predict_examples = FLAGS.max_nb_preds
    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.
      while len(predict_examples) % FLAGS.predict_batch_size != 0:
        predict_examples.append(PaddingInputExample())


    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)

    predict_drop_remainder = True if FLAGS.use_tpu else False
    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)

    result = estimator.predict(input_fn=predict_input_fn)

    #assert num_written_lines == num_actual_predict_examples

    if FLAGS.do_embed:
      output_predict_file = os.path.join(FLAGS.output_dir, "encoder_results.tsv")
      with tf.gfile.GFile(output_predict_file, "w") as writer:
        num_written_lines = 0
        tf.logging.info("***** Extracting encoder results *****")
        for (i, prediction) in enumerate(result):
          embeddings = prediction["encoder_output"]
          for emb in embeddings: 
            out_line = [] 
            for e in emb:
              out_line.append(str(e))
            writer.write('\t'.join(out_line)+"\n")
          if i >= num_actual_predict_examples:
            break
          num_written_lines += 1

    else:
      output_predict_file = os.path.join(FLAGS.output_dir, "test_results.tsv")
      with tf.gfile.GFile(output_predict_file, "w") as writer:
        num_written_lines = 0
        tf.logging.info("***** Predict results *****")
        for (i, prediction) in enumerate(result):
          probabilities = prediction["probabilities"]
          if task_name=='varname':
            for class_prob in probabilities: 
              out_line = [] 
              for prob in class_prob:
                out_line.append(str(prob))
              writer.write('\t'.join(out_line)+"\n")
            #if i >= num_actual_predict_examples:
            if i >= FLAGS.max_nb_preds:
              break
          else:        
            output_line = "\t".join(
                 str(class_probability)
                 for class_probability in probabilities) + "\n"
            writer.write(output_line)
          num_written_lines += 1

    output_predict_file = os.path.join(FLAGS.output_dir, "attention_results.tsv")
    with tf.gfile.GFile(output_predict_file, "w") as writer:
      num_written_lines = 0
      tf.logging.info("***** Attention results *****")
      for (i, prediction) in enumerate(result):
        att = prediction["attention_output"]
        for i in range(12):
          for j in range(64):
            for k in range(64):
              writer.write("%s " % str(att[i][j][k]))
          writer.write("\n")
        break
        num_written_lines += 1


if __name__ == "__main__":
  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("label_vocab")
  flags.mark_flag_as_required("bert_config_file")
  flags.mark_flag_as_required("output_dir")
  tf.app.run()