def demo(model, config, id_to_tag, tag_to_id):
logger = get_logger(config.log_file)
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = True
with tf.Session(config=tf_config) as sess:
ckpt = tf.train.get_checkpoint_state(config.ckpt_path)
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
saver = tf.train.Saver()
if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
logger.info("Reading model parameters from %s" %
ckpt.model_checkpoint_path)
# saver = tf.train.import_meta_graph('ckpt/ner.ckpt.meta')
# saver.restore(session, tf.train.latest_checkpoint("ckpt/"))
saver.restore(sess, ckpt.model_checkpoint_path)
while True:
line = input("input sentence, please:")
inputs = input_from_line(line, config.max_seq_len, tag_to_id)
trans = model.trans.eval(sess)
feed_dict = get_feed_dict(model, False, inputs, config)
lengths, scores = sess.run([model.lengths, model.logits],
feed_dict)
batch_paths = decode(scores, lengths, trans, config)
tags = [id_to_tag[idx] for idx in batch_paths[0]]
result = bio_to_json(inputs[0], tags[1:-1])
print(result['entities'])
def bert_ner_infer():
params = json.loads(request.get_data(), encoding="utf-8")
text = params["text"]
url = params["url"]
x, len_list = create_infer_inputs(text, max_len, tokenizer)
print("len_list: ", len_list)
input_ids = x[0].tolist()
token_type_ids = x[1].tolist()
attention_mask = x[2].tolist()
data = json.dumps({
"signature_name": "serving_default",
"inputs": {
"input_ids": input_ids,
"token_type_ids": token_type_ids,
"attention_mask": attention_mask
}
})
headers = {"content-type": "application/json"}
result = requests.post(url, data=data, headers=headers)
result = json.loads(result.text)
pred_logits = result["outputs"][0]
pred = np.argmax(pred_logits, axis=1).tolist()
print("pred: ", pred)
predict_label = []
for j in range(min(len_list[0], max_len)):
predict_label.append(id2tag[pred[j]])
return_result = bio_to_json(text, predict_label)
return jsonify(return_result)
def evaluate_line(self, sess, inputs, id_to_tag):
trans = self.trans.eval(sess)
lengths, scores = self.run_step(sess, False, inputs)
batch_paths = self.decode(scores, lengths, trans)
tags = [id_to_tag[idx] for idx in batch_paths[0]]
return bio_to_json(inputs[0], tags[1:-1])
def bert_ner_model_infer(sentence, tag_to_id, id_to_tag, max_length,
docker_url):
final_result = []
return_result = {"code": 200, "message": "success"}
# sentence 去掉空格
sentence_ = sentence.replace('\r\n', '✈').replace(' ', '✈')
sentence_ = sentence_.replace('\u3000', '✈') # todo 新增处理
sentence_ = sentence_.replace('\xa0', '✈')
# 统计空格和韩文的信息
null_index_list, korean_index_list, null_korean_dict, m1 = count_korean_null_info(
sentence_)
sentence_ = ''.join(sentence_.split())
# sentence 去掉韩文
sentence_ = re.sub('[\uac00-\ud7ff]+', '', sentence_)
try:
# for sentence in text:
# _, segment_ids, word_ids, word_mask, label_ids
token_result = input_from_line(sentence_, max_length,
tag_to_id) # FLAGS.max_seq_len
word_ids = token_result[2].tolist()
word_mask = token_result[3].tolist()
seg_ids = token_result[1].tolist()
data = json.dumps({
"signature_name": "serving_default",
"inputs": {
"input_ids": word_ids,
"input_mask": word_mask,
"segment_ids": seg_ids,
"dropout": 1.0
}
})
headers = {"content-type": "application/json"}
json_response = requests.post(docker_url, data=data, headers=headers)
if json_response.status_code == 200:
result = json.loads(json_response.text)
if result == '':
temp = {'content': sentence, "Entity": []}
logger.info("实体识别最终结果:" + str(temp))
return temp
else:
pred = result["outputs"][0]
pred = np.array(pred)
label_list = pred.argmax(axis=1).tolist()[1:-1]
# 还原空格和韩语
if len(null_korean_dict) > 0:
label_list = preprocessing_korean_null(
korean_index_list, null_korean_dict, m1, label_list,
tag_to_id)
pred_label = []
for i in range(min(len(sentence),
max_length - 2)): # FLAGS.max_seq_len
pred_label.append(id_to_tag[label_list[i]])
pred_label = processing_general_format(pred_label)
logger.info("模型预测的结果: {}".format(pred_label))
# pred_label = pred_label[1:-1]
res = bio_to_json(sentence, pred_label)
if len(res['entities']) != 0:
for i in range(len(res['entities'])):
if res['entities'][i]["name"] != " ":
final_result.append(res['entities'][i])
else:
final_result = []
temp = {'content': sentence, "Entity": final_result}
logger.info("实体识别抽取结果：" + str(temp))
return temp
else:
temp = {'content': sentence, "Entity": []}
logger.info("实体识别抽取最终结果:" + str(temp))
return temp
except Exception as e:
logger.error(traceback.format_exc())
return_result["code"] = 400
return_result["message"] = traceback.format_exc()
return_result['content'] = ''
return_result["Entity"] = []
return return_result
def evaluate_line(self, sess, inputs, id_to_tag):
lengths, batch_paths = self.run_step(sess, False, inputs)
print(batch_paths)
tags = [id_to_tag[idx] for idx in batch_paths[0][2:lengths[0]]]
return bio_to_json(inputs[0], tags)
