def function(self, text):
try:
text = text.replace(' ', '')
x = [_ for _ in text]
# Pre-processor data
processor = utils.load_processor(
model_path='saved_model/time_entity/1')
tensor = processor.process_x_dataset([x])
# only for bert Embedding
tensor = [{
"Input-Token:0": i.tolist(),
"Input-Segment:0": np.zeros(i.shape).tolist()
} for i in tensor]
# predict
r = requests.post(
"http://localhost:8501/v1/models/time_entity:predict",
json={"instances": tensor})
preds = r.json()['predictions']
# Convert result back to labels
labels = processor.reverse_numerize_label_sequences(
np.array(preds).argmax(-1))
entities = get_predict('TIME', text, labels[0])
return entities
except Exception:
self.write(traceback.format_exc().replace('\n', '<br>'))
def predict():
t1 = time.time()
text = '据《新闻联播》报道，9月9日至11日，中央纪委书记赵乐际到河北调研。'
x = [_ for _ in text]
# Pre-processor data
processor = utils.load_processor(model_path='saved_model/time_entity/1')
tensor = processor.process_x_dataset([x])
# only for bert Embedding
tensor = [{
"Input-Token:0": i.tolist(),
"Input-Segment:0": np.zeros(i.shape).tolist()
} for i in tensor]
# predict
r = requests.post("http://localhost:8501/v1/models/time_entity:predict",
json={"instances": tensor})
preds = r.json()['predictions']
# Convert result back to labels
labels = processor.reverse_numerize_label_sequences(
np.array(preds).argmax(-1))
for label in labels:
entities = get_predict('TIME', text, label)
print(entities)
t2 = time.time()
print('cost time: %ss.' % str(round(t2 - t1, 4)))
def test_load(self):
model_path = os.path.join(tempfile.gettempdir(), str(time.time()))
model = BiGRU_Model()
model.fit(class_train_x, class_train_y, epochs=1)
model.save(model_path)
processor = utils.load_processor(model_path)
assert processor.token2idx == model.embedding.processor.token2idx
assert processor.label2idx == model.embedding.processor.label2idx
assert processor.__class__ == model.embedding.processor.__class__
process_x_0 = processor.process_x_dataset(class_train_x[:10])
process_x_1 = model.embedding.process_x_dataset(class_train_x[:10])
assert np.array_equal(process_x_0, process_x_1)
def get_ai_reply(sentence,
model_path='../data/chat_model_serving/1',
url="http://106.54.166.111:8501/v1/models/gru_bert:predict",
answer='../data/answer.csv'):
sentence = list(sentence)
processor = utils.load_processor(model_path=model_path)
tensor = processor.process_x_dataset([sentence])
tensor = [{
"Input-Token:0": i.tolist(),
"Input-Segment:0": np.zeros(i.shape).tolist()
} for i in tensor]
req = requests.post(url, json={"instances": tensor})
preds = req.json()['predictions']
labels = processor.reverse_numerize_label_sequences(
np.array(preds).argmax(-1))
reply = answer.loc[int(labels[0]), 'answer']
prob = np.array(preds).max()
return labels[0], prob, reply
# 先只训练一轮
model.fit(train_x, train_y, valid_x, valid_y, batch_size=32, epochs=10)
model.evaluate(test_x, test_y, batch_size=512)
model.save('bert_model')
import kashgari
kashgari.utils.convert_to_saved_model(model, 'tf_bert_model', version=1)
test = pd.read_csv("test_bert_remove.csv")
print(test.head())
import numpy as np
predict_list = list(test['comment'].apply(lambda x: tokenizer.tokenize(x)))
#predict_list = list(test['comment'])
np_pre = np.asarray(predict_list)
print(np_pre[1])
from kashgari import utils
processor = utils.load_processor(model_path='tf_bert_model/1')
tensor = processor.process_x_dataset(np_pre)
# tensor = [{
# "Input-Token:0": i.tolist(),
# "Input-Segment:0": np.zeros(i.shape).tolist()
# } for i in tensor]
print(tensor[1])
result = model.predict(tensor)
print(f"{result}")