def ner():
key = 'text'
val = request.args.get(key)
word_to_id, tag_to_id, id_to_tag, sess, model = Singleton.get_instance()
batch = n_service.get_batch_manager(id_to_tag, tag_to_id, val, word_to_id)
with sess.as_default():
ner_results = model.predict2(sess, batch)
labels = []
datas = []
for i, ret in enumerate(ner_results):
strs = ret.split(" ")
labels.append(strs[1])
datas.append(strs[0])
results = []
for i in range(len(datas)):
if labels[i] == "O":
results.append(datas[i])
else:
results.append(datas[i] + labels[i])
ret2 = " ".join(results)
ret = "key:%s <br> <h4>value:%s</h4> " % (val, ret2)
return ret
def ner_text():
key = 'text'
val = request.args.get(key)
type = request.args.get('type')
word_to_id, tag_to_id, id_to_tag, sess, model = Singleton.get_instance()
batch = n_service.get_batch_manager(id_to_tag, tag_to_id, val, word_to_id)
with sess.as_default():
ner_results = model.predict2(sess, batch)
datas = []
idx = 0
is_flag = False
item = {}
for i, ret in enumerate(ner_results):
strs = ret.split(" ")
label = strs[1]
data = strs[0]
# print "label", label
if label.startswith("B"):
item = Item()
item.type = label[2:]
item.start = idx
is_flag = True
idx += len(data)
elif label.startswith("I"):
idx += len(data)
elif is_flag and label.startswith("O"): # 遇到O结束了
idx += len(data)
item.end = idx
is_flag = False
text = val[item.start:idx - 1]
item.text = text
# print "start", item.start, "end", idx, "text", text
datas.append(item.to_string())
else:
idx += len(data)
if is_flag:
item.end = idx
text = val[item.start:idx]
# print "start2", item.start, "end", idx, "text", text
item.text = text
# print json.dumps(item)
datas.append(item.to_string())
# d = {}
# d["dict"] = datas
# ret = json.dumps(datas)
ret = ",".join(datas)
ret = "{\"docs\":[%s]}" % (ret)
print "ret", ret
return ret
def process(self, req):
res = NerResponse()
val = req.query
#print "get query", val
word_to_id, tag_to_id, id_to_tag, sess, model = Singleton.get_instance(
)
batch = n_service.get_batch_manager(id_to_tag, tag_to_id, val,
word_to_id)
with sess.as_default():
ner_results = model.predict2(sess, batch)
datas = []
val = val.encode('utf-8')
# print "val-len", len(val)
idx = 0
is_flag = False
item = {}
for i, ret in enumerate(ner_results):
strs = ret.split(" ")
label = strs[1]
data = strs[0].encode('utf-8') # 中文3个，英文1个
# print "label", label
if label.startswith("B"):
item = Item()
item.type = dict[label[2:]]
item.start_idx = idx
is_flag = True
idx += len(data)
elif label.startswith("I"):
idx += len(data)
elif is_flag and label.startswith("O"): # 遇到O结束了
# idx += len(data)
item.end_idx = idx
is_flag = False
text = val[item.start_idx:idx]
item.text = text.encode('utf-8')
datas.append(item)
idx += len(data)
else:
idx += len(data)
if is_flag:
item.end_idx = idx
text = val[item.start_idx:idx]
item.text = text.encode('utf-8')
datas.append(item)
res.docs = datas
res.status = 200
model.logger.info("req:{},res:{}".format(req.query, datas))
return res
def ner_text_prob():
key = 'text'
val = request.args.get(key)
print "val-repr", repr(val)
word_to_id, tag_to_id, id_to_tag, sess, model = Singleton.get_instance()
batch = n_service.get_batch_manager(id_to_tag, tag_to_id, val, word_to_id)
with sess.as_default():
ner_results = model.predict_probility(sess, batch)
datas = []
val = val.encode('utf-8')
# print "val-len", len(val)
idx = 0
is_flag = False
item = {}
for i, ret in enumerate(ner_results):
strs = ret.split(" ")
label = strs[1]
data = strs[0].encode('utf-8') #中文3个，英文1个
# print "label", label
if label.startswith("B"):
item = Item()
item.type = dict[label[2:]]
item.start = idx
item.score = float(strs[2])
is_flag = True
idx += len(data)
elif label.startswith("I"):
idx += len(data)
item.score *= float(strs[2])
elif is_flag and label.startswith("O"): # 遇到O结束了
#idx += len(data)
item.end = idx
is_flag = False
text = val[item.start:idx]
item.text = text
item.score = np.power(item.score,
float(1.0 / len(text.decode('utf-8'))))
# print "text-mid-size", len(text.decode('utf-8'))
# print "start", item.start, "end", idx, "text", text
datas.append(item.to_string())
idx += len(data)
else:
idx += len(data)
if is_flag: # 每一个score都计算了的
item.end = idx
text = val[item.start:idx]
# print "start2", item.start, "end", idx, "text", text
item.text = text
item.score = np.power(item.score,
float(1.0 / len(text.decode('utf-8'))))
datas.append(item.to_string())
# d = {}
# d["dict"] = datas
# ret = json.dumps(datas)
ret = ",".join(datas)
ret = "{\"docs\":[%s]}" % (ret)
print "ret", ret
return ret