def predict_line(param):
# 初始化日志对象
logger = get_logger(param.test_log_file)
tf_config = tf.ConfigProto()
# 读取字典
mapping_dict = get_dict(param.dict_file)
# 根据保存的模型读取模型
model = Model(param, mapping_dict)
# 开始测试
with tf.Session(config=tf_config) as sess:
# 首先检查模型是否存在
ckpt_path = param.ckpt_path
ckpt = tf.train.get_checkpoint_state(ckpt_path)
# 看是否存在训练好的模型
if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
logger.info("Reading model parameters from {}".format(
ckpt.model_checkpoint_path))
# 如果存在就进行重新加载
model.saver.restore(sess, ckpt.model_checkpoint_path)
else:
logger.info("Cannot find the ckpt files!")
while True:
# 反复输入句子进行预测
line = input("请输入测试句子:")
raw_inputs, model_inputs = input_from_line_with_feature(line)
tag = model.evaluate_line(sess, model_inputs)
result = result_to_json(raw_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 predict_from_pb(self, document):
row = {'content': document}
df = pandas.DataFrame().append(row, ignore_index=True)
filename = "data/{}.csv".format(time.time())
df.to_csv(filename, index=False, escapechar="\\", columns=['content'])
with tf.Graph().as_default():
output_graph_def = tf.GraphDef()
with open(os.path.join(args.model_folder, 'money_model.pb'), "rb") as f:
output_graph_def.ParseFromString(f.read())
tf.import_graph_def(output_graph_def, name="")
with tf.Session() as sess: # config=config
sess.run(tf.global_variables_initializer())
# Get the input placeholders from the graph by name
char_input = sess.graph.get_tensor_by_name('CharInputs:0')
seg_input = sess.graph.get_tensor_by_name('SegInputs:0')
drop_keep_prob = sess.graph.get_tensor_by_name('Dropout:0')
# Tensors we want to evaluate,outputs
lengths = sess.graph.get_tensor_by_name('lengths:0')
logits = sess.graph.get_tensor_by_name("project/logits_outputs:0")
# predictions = sess.graph.get_tensor_by_name("Accuracy/predictions:0")
trans = sess.graph.get_tensor_by_name("crf_loss/transitions:0")
# fo = open(test_data_path, "r", encoding='utf8')
# all_data = fo.readlines()
# fo.close()
# for line in all_data: # 一行行遍历
lines = document.split(r"\n")
lines = [line for line in lines if len(line)>0]
list_original = []
list_amounts = []
for line in lines:
input_batch = input_from_line(line, self.char_to_id) # 处理测试数据格式
feed_dict = create_feed_dict(input_batch, char_input, seg_input, drop_keep_prob) # 创建输入的feed_dict
seq_len, scores = sess.run([lengths, logits], feed_dict)
print('---')
transition_matrix = trans.eval()
batch_paths = decode(scores, seq_len, transition_matrix)
tags = [self.id_to_tag[str(idx)] for idx in batch_paths[0]]
#print(tags)
result = result_to_json(input_batch[0][0], tags)
original = str(result['string'])
entities = result['entities']
if len(entities) != 0:
list_original.extend([original] * len(entities))
for entity in entities:
#是数字金额需要增强逻辑
if digit_regex.match(entity['word']) and len(entity['word']) >= 1:
aug_word = augment(entity['word'], line)
list_amounts.append(aug_word)
else:
list_amounts.append(entity['word'])
#print(entity['word'])
return {"answer":list_amounts,"line":list_original}
def predict(input_str):
with open(config.map_file, "rb") as f:
char_to_id, id_to_char, tag_to_id, id_to_tag = pickle.load(f)
""" 用cpu预测 """
model = torch.load(os.path.join(config.save_dir,"medical_ner_f1_0.976.ckpt"),
map_location="cpu"
)
model.eval()
if not input_str:
input_str = input("请输入文本: ")
_, char_ids, seg_ids, _ = prepare_dataset([input_str], char_to_id, tag_to_id, test=True)[0]
char_tensor = torch.LongTensor(char_ids).view(1,-1)
seg_tensor = torch.LongTensor(seg_ids).view(1,-1)
with torch.no_grad():
""" 得到维特比解码后的路径，并转换为标签 """
paths = model(char_tensor,seg_tensor)
tags = [id_to_tag[idx] for idx in paths[0]]
pprint(result_to_json(input_str, tags))
def cpu_predict(input_str):
with open(config.data_proc_file, "rb") as f:
train_data, dev_data, test_data = pickle.load(f)
char_to_id, id_to_char, tag_to_id, id_to_tag = pickle.load(f)
emb_matrix = pickle.load(f)
with open(config.map_file, "rb") as f:
char_to_id, id_to_char, tag_to_id, id_to_tag = pickle.load(f)
device = torch.device("cuda" if None else "cpu")
model = NERLSTM_CRF(config, char_to_id, tag_to_id, emb_matrix, device)
state_dict = torch.load(os.path.join(config.save_dir, "medical_ner.ckpt"),
map_location="cpu")
model.load_state_dict(state_dict)
""" 用cpu预测 """
model.eval()
if not input_str:
input_str = input("请输入文本: ")
_, char_ids, seg_ids, _ = prepare_dataset([input_str],
char_to_id,
tag_to_id,
test=True)[0]
char_tensor = torch.LongTensor(char_ids).view(1, -1)
seg_tensor = torch.LongTensor(seg_ids).view(1, -1)
with torch.no_grad():
""" 得到维特比解码后的路径，并转换为标签 """
paths = model(char_tensor, seg_tensor)
tags = [id_to_tag[idx] for idx in paths[0]]
pprint(result_to_json(input_str, tags))
def predict(input_str):
with open(config.map_file, "rb") as f:
char_to_id, id_to_char, tag_to_id, id_to_tag = pickle.load(f)
model = torch.load(os.path.join(config.save_dir,
"medical_ner_0.9723.ckpt"),
map_location="cpu")
if not input_str:
input_str = input("请输入文本: ")
char_ids, seg_ids, _ = prepare_dataset([input_str],
char_to_id,
tag_to_id,
test=True)[0]
char_tensor = torch.LongTensor(char_ids).view(1, -1)
seg_tensor = torch.LongTensor(seg_ids).view(1, -1)
""" 得到维特比解码后的路径，并转换为标签 """
paths = model(char_tensor, seg_tensor)
tags = [id_to_tag[idx] for idx in paths[0]]
""" 把开头和结尾的<start>,<end>标签去掉 """
tags.pop(0)
tags.pop(-1)
pprint(result_to_json(input_str, tags))
def predict_from_pb():
with open(FLAGS.json_file,
"r") as f: # with open(FLAGS.map_file, "rb") as f:
char_to_id, id_to_char, tag_to_id, id_to_tag = json.load(
f) # pickle.load(f)
print('json file loaded')
with tf.Graph().as_default():
output_graph_def = tf.GraphDef()
with open(pb_path, "rb") as f:
output_graph_def.ParseFromString(f.read())
tf.import_graph_def(output_graph_def, name="")
with tf.Session() as sess: # config=config
sess.run(tf.global_variables_initializer())
# Get the input placeholders from the graph by name
char_input = sess.graph.get_tensor_by_name('CharInputs:0')
seg_input = sess.graph.get_tensor_by_name('SegInputs:0')
drop_keep_prob = sess.graph.get_tensor_by_name('Dropout:0')
# Tensors we want to evaluate,outputs
lengths = sess.graph.get_tensor_by_name('lengths:0')
logits = sess.graph.get_tensor_by_name("project/logits_outputs:0")
# predictions = sess.graph.get_tensor_by_name("Accuracy/predictions:0")
trans = sess.graph.get_tensor_by_name("crf_loss/transitions:0")
fo = open(test_data_path, "r", encoding='utf8')
all_data = fo.readlines()
fo.close()
for line in all_data: # 一行行遍历
input_batch = input_from_line(line, char_to_id) # 处理测试数据格式
feed_dict = create_feed_dict(input_batch, char_input,
seg_input,
drop_keep_prob) # 创建输入的feed_dict
seq_len, scores = sess.run([lengths, logits], feed_dict)
print('---')
transition_matrix = trans.eval()
batch_paths = decode(scores, seq_len, transition_matrix)
tags = [id_to_tag[str(idx)] for idx in batch_paths[0]]
print(tags)
result = result_to_json(input_batch[0][0], tags)
original = str(result['string'])
entities = result['entities']
if len(entities) != 0:
for entity in entities:
print(entity['word'])
def predict_text(model, input_str):
if not input_str:
input_str = input("请输入文本: ")
_, char_ids, seg_ids, _ = prepare_dataset([input_str],
char_to_id,
tag_to_id,
test=True)[0]
char_tensor = torch.LongTensor(char_ids).view(1, -1)
seg_tensor = torch.LongTensor(seg_ids).view(1, -1)
with torch.no_grad():
""" 得到维特比解码后的路径，并转换为标签 """
paths = model(char_tensor, seg_tensor)
tags = [id_to_tag[idx] for idx in paths[0]]
return result_to_json(input_str, tags)