def enhance_dataset(bert_path,
path,
table_path,
out_path,
max_n=10,
threshold=1.9):
vocab_file = os.path.join(bert_path, f'vocab.txt')
bert_tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file,
do_lower_case=True)
tables = load_table(table_path)
fout = open(out_path, "w", encoding="utf-8")
cnt = 0
with open(path) as fin:
for line in fin:
if cnt % 1000 == 0:
print(cnt)
cnt += 1
example = json.loads(line.strip())
table = tables[example["table_id"]]
h_aug, q_feature = enhance_example(bert_tokenizer,
example["question_tok"],
table["header"], table["rows"],
max_n, threshold)
example["header_aug"] = h_aug
example["question_feature"] = q_feature
json_str = json.dumps(example,
ensure_ascii=False,
default=json_default_type_checker)
json_str += "\n"
fout.writelines(json_str)
fout.close()
def __init__(self, logger, train_path, eval_path, bert_path, max_length, batch_size, rate, epoch,
loss, tf_config, model_path, summary_path, tag2label=None, encoder_layer=11):
self.logger = logger
self.train_path = train_path
self.eval_path = eval_path
self.bert_path = bert_path
self.max_length = max_length
self.batch_size = batch_size
self.rate = rate
self.epoch = epoch
self.loss = loss
self.encoder_layer = encoder_layer
self.tf_config = tf_config
self.model_path = model_path
self.summary_path = summary_path
vocab_file = os.path.join(self.bert_path, 'vocab.txt')
self.tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file)
self.predictor = None
if tag2label is None:
tag2label = {
'O': 0,
'B-com': 1,
'I-com': 2,
'B-pos': 3,
'I-pos': 4
}
self.tag2label = tag2label
self.label2tag = {}
for key in self.tag2label:
self.label2tag[self.tag2label[key]] = key
def __init__(self,
attribute_name,
init_checkpoint=FLAGS.init_checkpoint,
is_training=False):
"""
checkpoint: Initial checkpoint (usually from a pre-trained BERT model).
is_training: bool. true for training model, false for eval model. Controls
whether dropout will be applied.
"""
self.attribute_name = attribute_name
self.init_checkpoint = init_checkpoint
self.is_training = is_training
self.learning_rate = FLAGS.learning_rate
self.bert_config = modeling.BertConfig.from_json_file(
FLAGS.bert_config_file)
self.tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=False)
self.data_processor = NerProcessor(self.attribute_name)
self.attribute_name = attribute_name
self.attribute_info = attr_ner_label.attribute_infos[attribute_name]
self.attribute_dict = self.attribute_info['dict']
self.attribute_label = self.attribute_info['label']
self.file_prefix = self.attribute_info['file_prefix']
self.labels = self.data_processor.get_labels()
self.num_labels = len(self.labels)
self.graph = tf.Graph()
with self.graph.as_default():
self._model_builder()
self.sess.run(tf.initialize_all_variables())
def load(self, dir):
assert os.path.exists(dir)
with open(os.path.join(dir, "config"), "rb") as fin:
self.config = pickle.load(fin)
vocab_file = os.path.join(self.config['bert_path'], "vocab.txt")
self.tokenizer = tokenization.FullTokenizer(
vocab_file=vocab_file, do_lower_case=self.config['do_lower_case'])
saved_model = sorted(glob.glob(os.path.join(dir, "exported", "*")))[-1]
self.predictor = tf.contrib.predictor.from_saved_model(saved_model)
def load_bert(BERT_PATH):
config_file = os.path.join(BERT_PATH, f'config.json')
vocab_file = os.path.join(BERT_PATH, f'vocab.txt')
init_checkpoint = os.path.join(BERT_PATH, f'pytorch_model.bin')
bert_config = BertConfig.from_json_file(config_file)
bert_config.print_status()
bert_tokenizer = tokenization.FullTokenizer(vocab_file=vocab_file,
do_lower_case=True)
bert_model = BertModel(bert_config)
bert_model.load_state_dict(torch.load(init_checkpoint, map_location='cpu'))
return bert_model, bert_tokenizer, bert_config
def __init__(self, do_train=False):
self.bert_config = modeling.BertConfig.from_json_file(
FLAGS.bert_config_file)
self.tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
self.data_processor = NerProcessor()
self.label_list = self.data_processor.get_labels()
self.id2label = {i: label for i, label in enumerate(self.label_list)}
self.train_examples = None
self.num_train_steps = None
self.num_warmup_steps = None
self.do_train = do_train
if do_train:
self.train_examples = self.data_processor.get_train_examples(
FLAGS.data_dir)
self.num_train_steps = int(
len(self.train_examples) / FLAGS.train_batch_size *
FLAGS.num_train_epochs)
self.num_warmup_steps = int(self.num_train_steps *
FLAGS.warmup_proportion)
self.model_fn = model_fn_builder(
bert_config=self.bert_config,
num_labels=len(self.label_list),
init_checkpoint=FLAGS.init_checkpoint,
learning_rate=FLAGS.learning_rate,
num_train_steps=self.num_train_steps,
num_warmup_steps=self.num_warmup_steps,
use_tpu=FLAGS.use_tpu,
use_one_hot_embeddings=FLAGS.use_tpu)
is_per_host = tf.contrib.tpu.InputPipelineConfig.PER_HOST_V2
run_config = tf.contrib.tpu.RunConfig(
cluster=None,
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))
self.estimator = tf.contrib.tpu.TPUEstimator(
use_tpu=FLAGS.use_tpu,
model_fn=self.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)
def fit(self):
tf.gfile.MakeDirs(self.save_path)
processor = Processor()
train_examples, labels = processor.get_train_examples(self.train_path)
num_train_steps = int(
len(train_examples) / self.batch_size * self.epoch)
num_warmup_steps = 0 if self.model else int(num_train_steps *
self.warmup_ratio)
if not self.model:
self.labels = labels
self.config['labels'] = self.labels
init_checkpoint = os.path.join(self.bert_path, "bert_model.ckpt")
bert_config_file = os.path.join(self.bert_path, "bert_config.json")
bert_config = modeling.BertConfig.from_json_file(bert_config_file)
model_fn = model_fn_builder(bert_config=bert_config,
num_labels=len(labels),
init_checkpoint=init_checkpoint,
filters=self.filters,
kernel_size=self.kernel_size,
strides=self.strides,
pool_size=self.pool_size,
learning_rate=self.learning_rate,
num_train_steps=num_train_steps,
num_warmup_steps=num_warmup_steps)
run_config = tf.estimator.RunConfig(
model_dir=self.save_path,
save_checkpoints_steps=self.save_checkpoints_steps,
session_config=self.tf_config)
self.model = tf.estimator.Estimator(model_fn=model_fn,
config=run_config)
vocab_file = os.path.join(self.bert_path, "vocab.txt")
self.tokenizer = tokenization.FullTokenizer(
vocab_file=vocab_file, do_lower_case=self.do_lower_case)
train_file = os.path.join(self.save_path, "train.tf_record")
file_based_convert_examples_to_features(train_examples, labels,
self.max_length,
self.tokenizer, train_file)
train_input_fn = file_based_input_fn_builder(
input_file=train_file,
seq_length=self.max_length,
is_training=True,
drop_remainder=True,
batch_size=self.batch_size)
self.model.train(input_fn=train_input_fn, max_steps=num_train_steps)
with open(os.path.join(self.save_path, "config"), "wb") as out:
pickle.dump(self.config, out)
def __init__(self, init_checkpoint=FLAGS.init_checkpoint, is_training=False):
"""
checkpoint: Initial checkpoint (usually from a pre-trained BERT model).
is_training: bool. true for training model, false for eval model. Controls
whether dropout will be applied.
"""
self.init_checkpoint = init_checkpoint
self.is_training = is_training
self.learning_rate = FLAGS.learning_rate
self.bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)
self.tokenizer = tokenization.FullTokenizer(vocab_file=FLAGS.vocab_file)
self.data_processor = DataProcessor()
self.labels = self.data_processor.get_labels()
self.num_labels = len(self.labels)
self.graph = tf.Graph()
self.summary_writer = tf.summary.FileWriter(f'{DIR}/logs', tf.get_default_graph())
with self.graph.as_default():
self._model_builder()
self.sess.run(tf.initialize_all_variables())
def __init__(self, init_checkpoint, is_training=False):
"""
checkpoint: Initial checkpoint (usually from a pre-trained BERT model).
is_training: bool. true for training model, false for eval model. Controls
whether dropout will be applied.
"""
self.init_checkpoint = init_checkpoint
self.is_training = is_training
self.bert_config = modeling.BertConfig.from_json_file(
f"{DIR}/model/pretrained_model/bert_config.json")
self.tokenizer = tokenization.FullTokenizer(
vocab_file=f"{DIR}/model/pretrained_model/vocab.txt")
self.data_processor = DataProcessor()
self.labels = self.data_processor.get_labels()
self.num_labels = len(self.labels)
self.max_seq_length = 256
self.graph = tf.Graph()
with self.graph.as_default():
self._model_builder()
self.sess.run(tf.initialize_all_variables())
def main(_):
logging.set_verbosity(logging.INFO)
processors = {"ner": NerProcessor}
# if not FLAGS.do_train and not FLAGS.do_eval:
# raise ValueError("At least one of `do_train` or `do_eval` must be True.")
bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)
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))
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()
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_examples = processor.get_train_examples(FLAGS.data_dir)
num_train_steps = int(
len(train_examples) / FLAGS.train_batch_size *
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)
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:
train_file = os.path.join(FLAGS.output_dir, "train.tf_record")
_, _ = filed_based_convert_examples_to_features(
train_examples, label_list, FLAGS.max_seq_length, tokenizer,
train_file)
logging.info("***** Running training *****")
logging.info(" Num examples = %d", len(train_examples))
logging.info(" Batch size = %d", FLAGS.train_batch_size)
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:
eval_examples = processor.get_dev_examples(FLAGS.data_dir)
eval_file = os.path.join(FLAGS.output_dir, "eval.tf_record")
batch_tokens, batch_labels = filed_based_convert_examples_to_features(
eval_examples, label_list, FLAGS.max_seq_length, tokenizer,
eval_file)
logging.info("***** Running evaluation *****")
logging.info(" Num examples = %d", len(eval_examples))
logging.info(" Batch size = %d", FLAGS.eval_batch_size)
# if FLAGS.use_tpu:
# eval_steps = int(len(eval_examples) / FLAGS.eval_batch_size)
# 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=False)
result = estimator.evaluate(input_fn=eval_input_fn)
output_eval_file = os.path.join(FLAGS.output_dir, "eval_results.txt")
with open(output_eval_file, "w") as wf:
logging.info("***** Eval results *****")
confusion_matrix = result["confusion_matrix"]
p, r, f = metrics.calculate(confusion_matrix, len(label_list) - 1)
logging.info("***********************************************")
logging.info("********************P = %s*********************",
str(p))
logging.info("********************R = %s*********************",
str(r))
logging.info("********************F = %s*********************",
str(f))
logging.info("***********************************************")
if FLAGS.do_predict:
with open(FLAGS.data_dir + '/label2id.pkl', 'rb') as rf:
label2id = pickle.load(rf)
id2label = {value: key for key, value in label2id.items()}
predict_examples = processor.get_test_examples(FLAGS.data_dir)
predict_file = os.path.join(FLAGS.output_dir, "predict.tf_record")
batch_tokens, batch_labels = filed_based_convert_examples_to_features(
predict_examples, label_list, FLAGS.max_seq_length, tokenizer,
predict_file)
logging.info("***** Running prediction*****")
logging.info(" Num examples = %d", len(predict_examples))
logging.info(" Batch size = %d", FLAGS.predict_batch_size)
predict_input_fn = file_based_input_fn_builder(
input_file=predict_file,
seq_length=FLAGS.max_seq_length,
is_training=False,
drop_remainder=False)
result = estimator.predict(input_fn=predict_input_fn)
output_predict_file = os.path.join(FLAGS.output_dir, "label_test.txt")
# here if the tag is "X" means it belong to its before token, here for convenient evaluate use
# conlleval.pl we discarding it directly
Writer(output_predict_file, result, batch_tokens, batch_labels,
id2label)
