def __init__(self):
# 建立两个 graph
self.g1 = tf.Graph()
self.g2 = tf.Graph()
# 为每个 graph 建创建一个 session
self.sess1 = tf.Session(graph=self.g1)
self.sess2 = tf.Session(graph=self.g2)
self.bi_label_list = ['0', '1']
self.bi_num_labels = len(self.bi_label_list)
self.muti_label_list = [str(i) for i in range(40)]
self.muti_num_labels = len(self.muti_label_list)
self.configuration = cf.get_config()
self.max_seq_len = self.configuration['max_seq_length']
self.bi_bert_config_file = self.configuration['bert_config']
self.muti_bert_config_file = self.configuration['bert_config']
self.tokenizer = tokenization.FullTokenizer(
vocab_file=self.configuration['vocab_file'], do_lower_case=True)
self.bi_init_checkpoint = self.configuration['bi_init_checkpoint']
self.muti_init_checkpoint = self.configuration['muti_init_checkpoint']
self._init_session()
def __init__(self):
self.configuration = cf_muti.get_config()
self.bert_config_file = self.configuration['bert_config']
self.do_lower_case = self.configuration['do_lower_case']
self.init_checkpoint = self.configuration['init_checkpoint']
self.max_seq_length = self.configuration['max_seq_length']
self.vocab_file = self.configuration['vocab_file']
self.bert_config = modeling.BertConfig.from_json_file(self.bert_config_file)
self.tokenizer = tokenization.FullTokenizer(
vocab_file=self.vocab_file, do_lower_case=self.do_lower_case)
self.label_list = [str(i) for i in range(40)]
Args = collections.namedtuple("Args", ["init_checkpoint", "max_seq_length", "bert_config", "num_labels"])
self.args = Args(init_checkpoint=self.init_checkpoint, max_seq_length=self.max_seq_length,
bert_config=self.bert_config, num_labels=len(self.label_list))
self.is_training = False
self.use_one_hot_embeddings = False
self.session = None
self.graph = None
self.input_ids_ph = None
self.input_mask_ph = None
self.segment_ids_ph = None
self.init_checkpoint = self.args.init_checkpoint
self.max_seq_length = self.args.max_seq_length
self.bert_config = self.args.bert_config
self.num_labels = self.args.num_labels
self._init_session()
self._build_graph()
def main(_):
tf.logging.set_verbosity(tf.logging.INFO)
processors = {
"cola": ColaProcessor,
"mnli": MnliProcessor,
"mrpc": MrpcProcessor,
"xnli": XnliProcessor,
"muti_attr": MutiattrProcessor,
}
tokenization.validate_case_matches_checkpoint(FLAGS.muti_do_lower_case,
FLAGS.muti_init_checkpoint)
if not FLAGS.muti_do_train and not FLAGS.muti_do_eval and not FLAGS.muti_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.muti_bert_config_file)
if FLAGS.muti_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.muti_max_seq_length, bert_config.max_position_embeddings))
tf.gfile.MakeDirs(FLAGS.muti_output_dir)
task_name = FLAGS.muti_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.muti_vocab_file,
do_lower_case=FLAGS.muti_do_lower_case)
tpu_cluster_resolver = None
if FLAGS.muti_use_tpu and FLAGS.muti_tpu_name:
tpu_cluster_resolver = tf.contrib.cluster_resolver.TPUClusterResolver(
FLAGS.muti_tpu_name,
zone=FLAGS.muti_tpu_zone,
project=FLAGS.muti_gcp_project)
is_per_host = tf.contrib.tpu.InputPipelineConfig.PER_HOST_V2
run_config = tf.contrib.tpu.RunConfig(
cluster=tpu_cluster_resolver,
master=FLAGS.muti_master,
model_dir=FLAGS.muti_output_dir,
save_checkpoints_steps=FLAGS.muti_save_checkpoints_steps,
tpu_config=tf.contrib.tpu.TPUConfig(
iterations_per_loop=FLAGS.muti_iterations_per_loop,
num_shards=FLAGS.muti_num_tpu_cores,
per_host_input_for_training=is_per_host))
train_examples = None
num_train_steps = None
num_warmup_steps = None
if FLAGS.muti_do_train:
train_examples = processor.get_train_examples(FLAGS.muti_data_dir)
num_train_steps = int(
len(train_examples) / FLAGS.muti_train_batch_size *
FLAGS.muti_num_train_epochs)
num_warmup_steps = int(num_train_steps * FLAGS.muti_warmup_proportion)
model_fn = model_fn_builder(bert_config=bert_config,
num_labels=len(label_list),
init_checkpoint=FLAGS.muti_init_checkpoint,
learning_rate=FLAGS.muti_learning_rate,
num_train_steps=num_train_steps,
num_warmup_steps=num_warmup_steps,
use_tpu=FLAGS.muti_use_tpu,
use_one_hot_embeddings=FLAGS.muti_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.muti_use_tpu,
model_fn=model_fn,
config=run_config,
train_batch_size=FLAGS.muti_train_batch_size,
eval_batch_size=FLAGS.muti_eval_batch_size,
predict_batch_size=FLAGS.muti_predict_batch_size)
if FLAGS.muti_do_train:
train_file = os.path.join(FLAGS.muti_output_dir, "train.tf_record")
file_based_convert_examples_to_features(train_examples, label_list,
FLAGS.muti_max_seq_length,
tokenizer, train_file)
tf.logging.info("***** Running training *****")
tf.logging.info(" Num examples = %d", len(train_examples))
tf.logging.info(" Batch size = %d", FLAGS.muti_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.muti_max_seq_length,
is_training=True,
drop_remainder=True)
estimator.train(input_fn=train_input_fn, max_steps=num_train_steps)
if FLAGS.muti_do_eval:
eval_examples = processor.get_dev_examples(FLAGS.muti_data_dir)
num_actual_eval_examples = len(eval_examples)
if FLAGS.muti_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.muti_eval_batch_size != 0:
eval_examples.append(PaddingInputExample())
eval_file = os.path.join(FLAGS.muti_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.muti_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.muti_use_tpu:
assert len(eval_examples) % FLAGS.muti_eval_batch_size == 0
eval_steps = int(len(eval_examples) // FLAGS.muti_eval_batch_size)
eval_drop_remainder = True if FLAGS.muti_use_tpu else False
eval_input_fn = file_based_input_fn_builder(
input_file=eval_file,
seq_length=FLAGS.muti_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.muti_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.muti_do_predict:
predict_examples = processor.get_test_examples(FLAGS.muti_data_dir)
num_actual_predict_examples = len(predict_examples)
if FLAGS.muti_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.muti_predict_batch_size != 0:
predict_examples.append(PaddingInputExample())
predict_file = os.path.join(FLAGS.muti_output_dir, "predict.tf_record")
file_based_convert_examples_to_features(predict_examples, label_list,
FLAGS.muti_max_seq_length,
tokenizer, predict_file)
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.muti_predict_batch_size)
predict_drop_remainder = True if FLAGS.muti_use_tpu else False
predict_input_fn = file_based_input_fn_builder(
input_file=predict_file,
seq_length=FLAGS.muti_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.muti_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 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
saver2 = tf.train.Saver()
saver2.restore(sess2,
tf.train.latest_checkpoint("muti_model_bert3_4_0107/"))
print('Successfully load the model_2!')
from bot_img.run_classifier_bi_attr import InputExample, convert_examples_to_features
bi_label_list = ['0', '1']
bi_num_labels = len(bi_label_list)
muti_label_list = [str(i) for i in range(40)]
muti_num_labels = len(muti_label_list)
tokenizer = tokenization.FullTokenizer(
vocab_file="../bert_3/chinese_L-3_H-576_A-3/vocab_4380.txt",
do_lower_case=True)
def bi_predict(sent):
prediction = None
result = {}
if not sent:
return 0
examples = [InputExample(0, sent, label='0')]
features = convert_examples_to_features(examples, bi_label_list, 128,
tokenizer)
input_ids = [feature.input_ids for feature in features]
input_mask = [feature.input_mask for feature in features]
segment_ids = [feature.segment_ids for feature in features]
label_ids = [feature.label_id for feature in features]