def evaluate_line(self, sess, inputs, id_to_tag):
trans = self.trans.eval()
lengths, scores = self.run_step(sess, False, inputs[1:])
lengths = [lengths]
batch_paths = self.decode(scores, lengths, trans)
tags = [id_to_tag[idx] for idx in batch_paths[0]]
return result_to_json(inputs[0], tags)
def evaluate_line(self):
with tf.Session(config=tf_config) as sess:
### 1. create model and load parameters
self.model = Model(self.config)
ckpt = tf.train.get_checkpoint_state(FLAGS.ckpt_path)
if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
logger.info("Reading model parameters from %s" %
ckpt.model_checkpoint_path)
self.model.saver.restore(sess, ckpt.model_checkpoint_path)
else:
logger.info("Created model with fresh parameters.")
sess.run(tf.global_variables_initializer())
### 2. Evaluating
while True:
try:
line = input("请输入测试句子:")
inputs = utils.input_from_line(line, self.ch_2_id)
trans = self.model.trans.eval(session=sess)
lengths, scores = self.model.run_step(sess, False, inputs)
batch_paths = self.model.decode(scores, lengths, trans)
tags = [self.id_2_tag[idx] for idx in batch_paths[0]]
result = utils.result_to_json(inputs[0][0], tags)
print(result)
except Exception as e:
logger.info(e)
def predict_line():
config = load_config(FLAGS.config_file)
logger = get_logger(FLAGS.log_file)
tf_config = tf.ConfigProto()
# 读取词典
with open(FLAGS.map_file, "rb") as f:
char_to_id, id_to_char, tag_to_id, id_to_tag = pickle.load(f)
with tf.Session(config=tf_config) as sess:
# 根据保存的模型读取模型
model = create_model(sess, Model, FLAGS.ckpt_path, load_word2vec,
config, id_to_char, logger)
while True:
# 反复输入句子进行预测
line = input("请输入测试句子:")
inputs, tag = model.evaluate_line(
sess, input_from_line(line, char_to_id), id_to_tag)
if FLAGS.server:
# 如果是命令行测试就打印json数据
result = result_to_json(inputs, tag)
result = js.dumps(result,
ensure_ascii=False,
indent=4,
separators=(',', ': '))
with open('./result/result.json', 'w', encoding='utf-8') as f:
f.write(result)
print("预测结果为：{}".format(result))
def evaluate_line(self, sess, inputs, id_to_tag):
trans = self.trans.eval()
lengths, scores = self.run_step(sess, False, inputs)
batch_paths = self.decode(scores, lengths, trans)
print(id_to_tag)
print(batch_paths)
tags = [id_to_tag[batch_paths[0][idx]] for idx in range(len(batch_paths[0])-1)]
return result_to_json(inputs[0][0], tags)
def evaluate_line(self, sess, inputs, id_to_tag):
#trans条件随机场分类得出的矩阵
trans = self.trans.eval(session=sess)
#score是[句数，字数，label数]
lengths, scores = self.run_step(sess, False, inputs)
#viterbi_decode 由显式序列scores和状态转移阵trans，求隐藏序列的最大概率路径，也即最短路径
batch_paths = self.decode(scores, lengths, trans)
tags = [id_to_tag[idx] for idx in batch_paths[0]]
return result_to_json(inputs[0][0], tags)
def evaluate_line(self, sess, inputs, id_to_tag):
"""
:param sess: session to run the model
:param inputs:
:param id_to_tag: index to tag name
:return:
"""
trans = self.trans.eval()
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 result_to_json(inputs[0][0], tags)
def predict_line(self, sess, sentence, id_to_tag):
"""
一个语句实例进行实体识别测试
:param sess:
:param sentence:
:param id_to_tag:
:return:
"""
trans = self.trans.eval()
lengths, logits = self.run_step(sess, False, sentence)
batch_paths = self.decode(logits, lengths, trans)
tags = [id_to_tag[idx] for idx in batch_paths[0]]
return result_to_json(sentence[0][0], tags)
def evaluate_line(self, sess, inputs, id_to_tag):
trans = self.trans.eval()
lengths, scores = self.evaluate_run(sess, False, inputs)
strings = inputs[0]
batch_paths = self.decode(scores, lengths, trans)
result = []
for i in range(len(strings)):
char_input = strings[i][:lengths[i]]
#print(char_input)
tags = [id_to_tag[idx] for idx in batch_paths[i][:lengths[i]]]
json_result = result_to_json(char_input, tags)
print(json_result)
result.append(json_result)
return result
def evaluate_line(self, sess, inputs, id_to_tag):
# print("evaluate_line:{}".format(inputs))
# "[['请调出汇坤园北门2016年4月9日中午11:40的录像,四倍速回放'], " \
# "[[25, 27, 28, 134, 731, 41, 29, 5, 4, 7, 2, 16, 17, 14, 8, 24, 9, 42, 33, 2, 2, 3, 14, 7, 6, 10, 11, 1, 108, 18, 19, 34, 36]], [[0, 1, 3, 1, 2, 3, 1, 3, 1, 2, 2, 3, 0, 0, 0, 0, 0, 1, 3, 1, 3, 0, 1, 3, 0, 1, 3, 0, 1, 3, 0, 1, 3]], " \
# "[[]]]"
trans = self.trans.eval()
# print("trans:{}".format(trans.shape))
lengths, scores = self.run_step(sess, False, inputs)
# print("lengths:{}".format(lengths.shape))
# print("scores:{}".format(scores.shape))
# trans: (35, 35)
# lengths: (1,)
# scores: (1, 33, 34)
batch_paths = self.decode(scores, lengths, trans)
# print("batch_paths:{}".format(batch_paths))
'batch_paths:[[24, 32, 24, 32, 21, 7, 15, 9, 17, 25, 23, 26, 25, 6, 17, 25, 23, 26, 25, 23, 26, 8, 32, 24, 32, 24, 32, 24, 21, 19, 4, 33, 18]]'
tags = [id_to_tag[idx] for idx in batch_paths[0]]
# print("tags:{}".format(tags))
"tags:['B-SPEED', 'S-ELOC', 'B-SPEED', 'S-ELOC', 'B-ETIME', 'B-SLOC', 'B-STIME', 'B-DAY', 'I-SPEED', 'E-SPEED', 'I-DAY', 'I-ROAD', 'E-SPEED', 'E-TYPE', 'I-SPEED', 'E-SPEED', 'I-DAY', 'I-ROAD', 'E-SPEED', 'I-DAY', 'I-ROAD', 'E-SLOC', 'S-ELOC', 'B-SPEED', 'S-ELOC', 'B-SPEED', 'S-ELOC', 'B-SPEED', 'B-ETIME', 'E-YEAR', 'I-ETIME', 'S-SLOC', 'B-YEAR']"
return result_to_json(inputs[0][0], tags)
def evaluate_line(self, sess, inputs, id_to_tag):
'''
inputs: 心脏病、腰部叩击痛阳性和肌无力
lengths: array([15])
scores: shape (1, 15, 51)
trans: 解码用的转移矩阵
batch_paths: [[13, 2, 14, 0, 6, 7, 10, 3, 3, 3, 11, 0, 0, 0, 0]]
tags: ['B-DIS', 'I-DIS', 'E-DIS', 'O', 'B-REG', 'E-REG', 'B-SGN', 'I-SGN', 'I-SGN', 'I-SGN', 'E-SGN', 'O', 'O', 'O', 'O']
return:
{
'entities': [
{
'word': '心脏病',
'type': 'DIS',
'start': 0,
'end': 3
},
{
'word': '腰部',
'type': 'REG',
'start': 4,
'end': 6
},
{
'word': '叩击痛阳性',
'type': 'SGN',
'start': 6,
'end': 11
}
],
'string': '心脏病、腰部叩击痛阳性和肌无力'
}
'''
lengths, scores = self.run_step(sess, False, inputs)
trans = self.trans.eval(session=sess)
batch_paths = self.decode(scores, lengths, trans)
tags = [id_to_tag[idx] for idx in batch_paths[0]]
return result_to_json(inputs[0][0], tags)
def evaluate_line(self, sess, inputs, label_vocab):
trans = self.trans.eval()
lengths, scores = self.run_step(sess, False, inputs)
batch_paths = self.decode(scores, lengths, trans)
tags = [label_vocab[idx] for idx in batch_paths[0]]
return result_to_json(inputs[0][0], tags)
def evaluate_line(self, sess, inputs, id_to_tag):
trans = self.trans.eval()
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 result_to_json(inputs[0][0], tags)
