def create_tokenizer_from_hub_module(bert_model_hub, do_lower=True):
"""Get the vocab file and casing info from the Hub module."""
with tf.Graph().as_default():
bert_module = hub.Module(bert_model_hub)
tokenization_info = bert_module(signature="tokenization_info",
as_dict=True)
with tf.compat.v1.Session() as sess:
vocab_file, do_lower_case = sess.run([
tokenization_info["vocab_file"],
tokenization_info["do_lower_case"]
])
return tokenization.FullTokenizer(vocab_file=vocab_file,
do_lower_case=(do_lower_case
and do_lower))
try:
del_file(args.output_dir)
except Exception as e:
print(e)
print('pleace remove the files of output dir and data.conf')
exit(-1)
# check output dir exists
if not os.path.exists(args.output_dir):
os.mkdir(args.output_dir)
# 创建ner dataprocessor对象
processor = processors[args.rc](args.output_dir)
label_list = processor.get_labels(labels=args.label_list)
print(len(label_list))
tokenizer = tokenization.FullTokenizer(
vocab_file=args.vocab_file, do_lower_case=args.do_lower_case)
session_config = tf.ConfigProto(
log_device_placement=False,
inter_op_parallelism_threads=0,
intra_op_parallelism_threads=0,
allow_soft_placement=True)
if args.do_train and args.do_eval:
train_and_eval(args=args, processor=processor, tokenizer=tokenizer,
bert_config=bert_config, sess_config=session_config, label_list=label_list)
# if args.filter_adam_var:
# adam_filter(os.path.join(args.output_dir, 'model'))
if args.do_predict:
predict(args=args, processor=processor, tokenizer=tokenizer,
def main(args):
"""
主函数
"""
os.environ['CUDA_VISIBLE_DEVICES'] = args.device_map
tf.logging.set_verbosity(tf.logging.INFO)
processors = {"RC": RCProcessor}
bert_config = modeling.BertConfig.from_json_file(args.bert_config_file)
if args.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" %
(args.max_seq_length, bert_config.max_position_embeddings))
# 在re train 的时候，才删除上一轮产出的文件，在predicted 的时候不做clean
if args.clean and args.do_train:
if os.path.exists(args.output_dir):
def del_file(path):
ls = os.listdir(path)
for i in ls:
c_path = os.path.join(path, i)
if os.path.isdir(c_path):
del_file(c_path)
else:
os.remove(c_path)
try:
del_file(args.output_dir)
except Exception as e:
print(e)
print('pleace remove the files of output dir and data.conf')
exit(-1)
# check output dir exists
if not os.path.exists(args.output_dir):
os.mkdir(args.output_dir)
# 创建ner dataprocessor对象
processor = processors[args.rc](args.output_dir)
tokenizer = tokenization.FullTokenizer(vocab_file=args.vocab_file,
do_lower_case=args.do_lower_case)
session_config = tf.ConfigProto(log_device_placement=False,
inter_op_parallelism_threads=0,
intra_op_parallelism_threads=0,
allow_soft_placement=True)
# estimator配置，需要研究一下
run_config = tf.estimator.RunConfig(
model_dir=args.output_dir,
save_summary_steps=args.save_summary_steps,
save_checkpoints_steps=args.save_checkpoints_steps,
session_config=session_config)
train_examples = processor.get_train_examples(args.data_dir)
eval_examples = processor.get_dev_examples(args.data_dir)
num_train_steps = None
num_warmup_steps = None
# 训练与验证
if args.do_train and args.do_eval:
# 加载训练数据
num_train_steps = int(
len(train_examples) * 1.0 / args.batch_size *
args.num_train_epochs)
if num_train_steps < 1:
raise AttributeError('training data is so small...')
num_warmup_steps = int(num_train_steps * args.warmup_proportion)
# 如何将日志信息打印到文件？
tf.logging.info("***** Running training *****")
tf.logging.info(" Num examples = %d", len(train_examples))
tf.logging.info(" Batch size = %d", args.batch_size)
tf.logging.info(" Num steps = %d", num_train_steps)
tf.logging.info("***** Running evaluation *****")
tf.logging.info(" Num examples = %d", len(eval_examples))
tf.logging.info(" Batch size = %d", args.batch_size)
label_list = processor.get_labels(labels=args.label_list)
# 返回的model_dn 是一个函数，其定义了模型，训练，评测方法，并且使用钩子参数，加载了BERT模型的参数进行了自己模型的参数初始化过程
# tf 新的架构方法，通过定义model_fn 函数，定义模型，然后通过EstimatorAPI进行模型的其他工作，Es就可以控制模型的训练，预测，评估工作等。
model_fn = model_fn_builder(bert_config=bert_config,
num_labels=len(label_list),
init_checkpoint=args.init_checkpoint,
learning_rate=args.learning_rate,
num_train_steps=num_train_steps,
num_warmup_steps=num_warmup_steps,
args=args)
params = {'batch_size': args.batch_size}
estimator = tf.estimator.Estimator(model_fn,
params=params,
config=run_config)
if args.do_train and args.do_eval:
# 1. 将数据转化为tf_record 数据
train_file = os.path.join(args.output_dir, "train.tf_record")
if not os.path.exists(train_file):
filed_based_convert_examples_to_features(train_examples,
label_list,
args.max_seq_length,
tokenizer, train_file)
# 2.读取record 数据，组成batch
train_input_fn = file_based_input_fn_builder(
input_file=train_file,
seq_length=args.max_seq_length,
is_training=True,
drop_remainder=True)
# estimator.train(input_fn=train_input_fn, max_steps=num_train_steps)
eval_file = os.path.join(args.output_dir, "eval.tf_record")
if not os.path.exists(eval_file):
filed_based_convert_examples_to_features(eval_examples, label_list,
args.max_seq_length,
tokenizer, eval_file)
eval_input_fn = file_based_input_fn_builder(
input_file=eval_file,
seq_length=args.max_seq_length,
is_training=False,
drop_remainder=False)
# train and eval togither
# early stop hook
early_stopping_hook = tf.contrib.estimator.stop_if_no_decrease_hook(
estimator=estimator,
metric_name='loss',
max_steps_without_decrease=num_train_steps,
eval_dir=None,
min_steps=0,
run_every_secs=None,
run_every_steps=args.save_checkpoints_steps)
train_spec = tf.estimator.TrainSpec(input_fn=train_input_fn,
max_steps=num_train_steps,
hooks=[early_stopping_hook])
eval_spec = tf.estimator.EvalSpec(input_fn=eval_input_fn)
tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
# 预测结果
if args.do_predict:
predict_examples = processor.get_test_examples(args.data_dir)
predict_file = os.path.join(args.output_dir, "predict.tf_record")
filed_based_convert_examples_to_features(predict_examples, label_list,
args.max_seq_length,
tokenizer, predict_file)
tf.logging.info("***** Running prediction*****")
tf.logging.info(" Num examples = %d", len(predict_examples))
tf.logging.info(" Batch size = %d", args.batch_size)
predict_input_fn = file_based_input_fn_builder(
input_file=predict_file,
seq_length=args.max_seq_length,
is_training=False,
drop_remainder=False)
result = estimator.predict(input_fn=predict_input_fn)
output_predict_file = os.path.join(args.output_dir,
"prediction_test.txt")
def result_to_pair(writer, data_file, result):
f = open(data_file, 'r')
for line, prediction in zip(f, result):
line = line.strip()
label = label_list[np.argmax(prediction['probabilities'])]
writer.write(line + '\t' + str(label) + '\n')
with codecs.open(output_predict_file, 'w', encoding='utf-8') as writer:
result_to_pair(writer, os.path.join(args.data_dir, 'test.txt'),
result)
# 在predict的时候同时需输出dev集的结果，因为需要后处理，再线上评估
eval_file = os.path.join(args.output_dir, "eval.tf_record")
predict_input_fn_d = file_based_input_fn_builder(
input_file=eval_file,
seq_length=args.max_seq_length,
is_training=False,
drop_remainder=False)
result_d = estimator.predict(input_fn=predict_input_fn_d)
output_d_file = os.path.join(args.output_dir, "prediction_dev.txt")
with codecs.open(output_d_file, 'w', encoding='utf-8') as writer:
result_to_pair(writer, os.path.join(args.data_dir, 'dev.txt'),
result_d)
# filter model
if args.filter_adam_var:
adam_filter(args.output_dir)
def main(_):
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.INFO)
processors = {
"cola": ColaProcessor,
"mnli": MnliProcessor,
"mrpc": MrpcProcessor,
"xnli": XnliProcessor,
}
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)
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))
tf.compat.v1.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()
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)
# 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:
train_file = os.path.join(FLAGS.output_dir, "train.tf_record")
file_based_convert_examples_to_features(
train_examples, label_list, FLAGS.max_seq_length, tokenizer, train_file)
tf.compat.v1.logging.info("***** Running training *****")
tf.compat.v1.logging.info(" Num examples = %d", len(train_examples))
tf.compat.v1.logging.info(" Batch size = %d", FLAGS.train_batch_size)
tf.compat.v1.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)
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.compat.v1.logging.info("***** Running evaluation *****")
tf.compat.v1.logging.info(" Num examples = %d (%d actual, %d padding)",
len(eval_examples), num_actual_eval_examples,
len(eval_examples) - num_actual_eval_examples)
tf.compat.v1.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.
if FLAGS.use_tpu:
assert len(eval_examples) % FLAGS.eval_batch_size == 0
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=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.compat.v1.gfile.GFile(output_eval_file, "w") as writer:
tf.compat.v1.logging.info("***** Eval results *****")
for key in sorted(result.keys()):
tf.compat.v1.logging.info(" %s = %s", key, str(result[key]))
writer.write("%s = %s\n" % (key, str(result[key])))
if FLAGS.do_predict:
predict_examples = processor.get_test_examples(FLAGS.data_dir)
num_actual_predict_examples = len(predict_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.
while len(predict_examples) % FLAGS.predict_batch_size != 0:
predict_examples.append(PaddingInputExample())
predict_file = os.path.join(FLAGS.output_dir, "predict.tf_record")
file_based_convert_examples_to_features(predict_examples, label_list,
FLAGS.max_seq_length, tokenizer,
predict_file)
tf.compat.v1.logging.info("***** Running prediction*****")
tf.compat.v1.logging.info(" Num examples = %d (%d actual, %d padding)",
len(predict_examples), num_actual_predict_examples,
len(predict_examples) - num_actual_predict_examples)
tf.compat.v1.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)
output_predict_file = os.path.join(FLAGS.output_dir, "test_results.tsv")
with tf.compat.v1.gfile.GFile(output_predict_file, "w") as writer:
num_written_lines = 0
tf.compat.v1.logging.info("***** Predict results *****")
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
assert num_written_lines == num_actual_predict_examples
def main(args):
"""
主函数
"""
os.environ['CUDA_VISIBLE_DEVICES'] = args.device_map
tf.logging.set_verbosity(tf.logging.INFO)
processors = {
"ner": NerProcessor
}
bert_config = modeling.BertConfig.from_json_file(args.bert_config_file)
if args.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" %
(args.max_seq_length, bert_config.max_position_embeddings))
# 在re train 的时候，才删除上一轮产出的文件，在predicted 的时候不做clean
if args.clean and args.do_train:
print('hahaha')
if os.path.exists(args.output_dir):
def del_file(path):
ls = os.listdir(path)
for i in ls:
c_path = os.path.join(path, i)
if os.path.isdir(c_path):
del_file(c_path)
else:
os.remove(c_path)
try:
del_file(args.output_dir)
except Exception as e:
print(e)
print('pleace remove the files of output dir and data.conf')
exit(-1)
# check output dir exists
if not os.path.exists(args.output_dir):
os.mkdir(args.output_dir)
# 创建ner dataprocessor对象
processor = processors[args.ner](args.output_dir)
tokenizer = tokenization.FullTokenizer(
vocab_file=args.vocab_file, do_lower_case=args.do_lower_case)
session_config = tf.ConfigProto(
log_device_placement=False,
inter_op_parallelism_threads=0,
intra_op_parallelism_threads=0,
allow_soft_placement=True)
# estimator配置，需要研究一下
run_config = tf.estimator.RunConfig(
model_dir=args.output_dir,
save_summary_steps=500,
save_checkpoints_steps=500,
session_config=session_config
)
train_examples = processor.get_train_examples(args.data_dir)
eval_examples = processor.get_dev_examples(args.data_dir)
num_train_steps = None
num_warmup_steps = None
# 训练与验证
if args.do_train and args.do_eval:
# 加载训练数据
num_train_steps = int(
len(train_examples) * 1.0 / args.batch_size * args.num_train_epochs)
if num_train_steps < 1:
raise AttributeError('training data is so small...')
num_warmup_steps = int(num_train_steps * args.warmup_proportion)
# 如何将日志信息打印到文件？
tf.logging.info("***** Running training *****")
tf.logging.info(" Num examples = %d", len(train_examples))
tf.logging.info(" Batch size = %d", args.batch_size)
tf.logging.info(" Num steps = %d", num_train_steps)
tf.logging.info("***** Running evaluation *****")
tf.logging.info(" Num examples = %d", len(eval_examples))
tf.logging.info(" Batch size = %d", args.batch_size)
label_list = processor.get_labels()
# 返回的model_dn 是一个函数，其定义了模型，训练，评测方法，并且使用钩子参数，加载了BERT模型的参数进行了自己模型的参数初始化过程
# tf 新的架构方法，通过定义model_fn 函数，定义模型，然后通过EstimatorAPI进行模型的其他工作，Es就可以控制模型的训练，预测，评估工作等。
model_fn = model_fn_builder(
bert_config=bert_config,
num_labels=len(label_list) + 1,
init_checkpoint=args.init_checkpoint,
learning_rate=args.learning_rate,
num_train_steps=num_train_steps,
num_warmup_steps=num_warmup_steps,
args=args) # 这个numlabels为什么要加1，要研究一下
params = {
'batch_size': args.batch_size
}
estimator = tf.estimator.Estimator(
model_fn,
params=params,
config=run_config)
if args.do_train and args.do_eval:
# 1. 将数据转化为tf_record 数据
train_file = os.path.join(args.output_dir, "train.tf_record")
if not os.path.exists(train_file):
filed_based_convert_examples_to_features(
train_examples, label_list, args.max_seq_length, tokenizer, train_file, args.output_dir)
# 2.读取record 数据，组成batch
train_input_fn = file_based_input_fn_builder(
input_file=train_file,
seq_length=args.max_seq_length,
is_training=True,
drop_remainder=True)
# estimator.train(input_fn=train_input_fn, max_steps=num_train_steps)
eval_file = os.path.join(args.output_dir, "eval.tf_record")
if not os.path.exists(eval_file):
filed_based_convert_examples_to_features(
eval_examples, label_list, args.max_seq_length, tokenizer, eval_file, args.output_dir)
eval_input_fn = file_based_input_fn_builder(
input_file=eval_file,
seq_length=args.max_seq_length,
is_training=False,
drop_remainder=False)
# train and eval togither
# early stop hook
early_stopping_hook = tf.contrib.estimator.stop_if_no_decrease_hook(
estimator=estimator,
metric_name='loss',
max_steps_without_decrease=1000,
eval_dir=None,
min_steps=0,
run_every_secs=None,
run_every_steps=args.save_checkpoints_steps)
train_spec = tf.estimator.TrainSpec(input_fn=train_input_fn, max_steps=num_train_steps,
hooks=[early_stopping_hook])
eval_spec = tf.estimator.EvalSpec(input_fn=eval_input_fn)
tf.estimator.train_and_evaluate(estimator, train_spec, eval_spec)
# 预测结果
if args.do_predict:
token_path = os.path.join(args.output_dir, "token_test.txt")
if os.path.exists(token_path):
os.remove(token_path)
with codecs.open(os.path.join(args.output_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(args.data_dir)
predict_file = os.path.join(args.output_dir, "predict.tf_record")
filed_based_convert_examples_to_features(predict_examples, label_list,
args.max_seq_length, tokenizer,
predict_file, args.output_dir, mode="test")
tf.logging.info("***** Running prediction*****")
tf.logging.info(" Num examples = %d", len(predict_examples))
tf.logging.info(" Batch size = %d", args.batch_size)
predict_input_fn = file_based_input_fn_builder(
input_file=predict_file,
seq_length=args.max_seq_length,
is_training=False,
drop_remainder=False)
result = estimator.predict(input_fn=predict_input_fn)
output_predict_file = os.path.join(args.output_dir, "label_test.txt")
def result_to_pair(writer, examples, result):
for predict_line, prediction in zip(examples, result):
idx = 0
line = ''
line_token = str(predict_line.text).split(' ')
label_token = str(predict_line.label).split(' ')
len_seq = len(label_token)
if len(line_token) != len(label_token):
tf.logging.info(predict_line.text)
tf.logging.info(predict_line.label)
break
for id in prediction:
if idx >= len_seq:
break
if id == 0:
continue
curr_labels = id2label[id]
if curr_labels in ['[CLS]', '[SEP]']:
continue
try:
line += line_token[idx] + ' ' + \
label_token[idx] + ' ' + curr_labels + '\n'
except Exception as e:
tf.logging.info(e)
tf.logging.info(predict_line.text)
tf.logging.info(predict_line.label)
line = ''
break
idx += 1
writer.write(line + '\n')
with codecs.open(output_predict_file, 'w', encoding='utf-8') as writer:
result_to_pair(writer, predict_examples, result)
eval_result = conlleval.return_report(output_predict_file)
print(''.join(eval_result))
# 写结果到文件中
with codecs.open(os.path.join(args.output_dir, 'predict_score.txt'), 'a', encoding='utf-8') as fd:
fd.write(''.join(eval_result))
# 在predict的时候同时需输出train和dev集的结果，因为需要后一步的生成负样本数据
train_file = os.path.join(args.output_dir, "train.tf_record")
eval_file = os.path.join(args.output_dir, "eval.tf_record")
predict_input_fn_t = file_based_input_fn_builder(
input_file=train_file,
seq_length=args.max_seq_length,
is_training=False,
drop_remainder=False)
predict_input_fn_d = file_based_input_fn_builder(
input_file=eval_file,
seq_length=args.max_seq_length,
is_training=False,
drop_remainder=False)
result_t = estimator.predict(input_fn=predict_input_fn_t)
output_t_file = os.path.join(args.output_dir, "label_train.txt")
result_d = estimator.predict(input_fn=predict_input_fn_d)
output_d_file = os.path.join(args.output_dir, "label_dev.txt")
with codecs.open(output_t_file, 'w', encoding='utf-8') as writer:
result_to_pair(writer, train_examples, result_t)
with codecs.open(output_d_file, 'w', encoding='utf-8') as writer:
result_to_pair(writer, eval_examples, result_d)
eval_result_t = conlleval.return_report(output_t_file)
print(''.join(eval_result_t))
eval_result_d = conlleval.return_report(output_d_file)
print(''.join(eval_result_d))
# filter model
if args.filter_adam_var:
adam_filter(args.output_dir)