def main(text, checkpoint_path):
q_vocabutil = VocabUtil("./data/xhj_q.vocab")
a_vocabutil = VocabUtil("./data/xhj_a.vocab")
tf.reset_default_graph()
# 定义训练用的循环神经网络模型。
with tf.variable_scope("nmt_model", reuse=None):
model = NMTModel()
print(datetime.now(), text)
# 根据词汇表，将测试句子转为ids。
text_ids = q_vocabutil.get_ids_word(text)
print(datetime.now(), text_ids)
# 建立解码所需的计算图。
output_op = model.inference(text_ids)
sess = tf.Session()
saver = tf.train.Saver()
saver.restore(sess, checkpoint_path)
# 读取翻译结果。
output_ids = sess.run(output_op)
print(datetime.now(), output_ids)
output_text = a_vocabutil.get_text(output_ids)
# 输出翻译结果。
print(datetime.now(), output_text)
sess.close()
def main(test_text, checkpoint_path):
tf.reset_default_graph()
# 定义训练用的循环神经网络模型。
with tf.variable_scope("nmt_model", reuse=None):
model = NMTModel()
# 定义个测试句子。
# test_en_text = "This is a test . <eos>"
# print(test_en_text)
vocabutil = VocabUtil("./data/xhj.vocab")
# 根据英文词汇表，将测试句子转为单词ID。
test_ids = vocabutil.get_ids_word(test_text)
print(test_ids)
# 建立解码所需的计算图。
output_op = model.inference(test_ids)
sess = tf.Session()
saver = tf.train.Saver()
saver.restore(sess, checkpoint_path)
# 读取翻译结果。
output_ids = sess.run(output_op)
print(output_ids)
# 根据中文词汇表，将翻译结果转换为中文文字。
answer = vocabutil.get_text(output_ids)
# 输出翻译结果。
print(datetime.now(), answer.encode('utf8').decode(sys.stdout.encoding))
sess.close()
class TestModel():
def __init__(self, vocabfile, checkpoint_path):
self.vocabutil = VocabUtil(vocabfile)
tf.reset_default_graph()
# 定义训练用的循环神经网络模型。
with tf.variable_scope("nmt_model", reuse=tf.AUTO_REUSE):
self.model = NMTModel()
# 建立解码所需的计算图。
text_ids = self.vocabutil.get_ids_word("你是谁")
output_op = self.model.inference(text_ids)
# self.saver = tf.train.import_meta_graph(checkpoint_path+".meta")
self.sess = tf.Session()
self.saver = tf.train.Saver()
self.saver.restore(self.sess, checkpoint_path)
# self.sess.run(tf.global_variables_initializer())
answer = self.vocabutil.get_text(self.sess.run(output_op))
print(answer)
# print(self.predict("你是谁"))
def close(self):
self.sess.close()
def predict(self, q):
# tf.reset_default_graph()
text_ids = self.vocabutil.get_ids_word(q)
output_op = self.model.inference(text_ids)
self.sess.run(output_op)
answer = self.vocabutil.get_text(output_op)
return answer
def talk_word(checkpoint_path):
vocabutil = VocabUtil("./data/xhj.vocab")
try:
while True:
q = _prompt_input()
if q.lower() == 'exit':
break
ids = vocabutil.get_ids_word(q)
answer = get_answer(ids, checkpoint_path)
answer = vocabutil.get_text(answer)
print(answer)
except Exception:
traceback.print_exc()
except KeyboardInterrupt:
print("Ctrl+c exit.")
def talk_word(checkpoint_path):
vocabutil = VocabUtil("./data/xhj.vocab")
#time consumption: 00.002783
try:
while True:
q = _prompt_input()
if q.lower() == 'exit':
break
ids = vocabutil.get_ids_word(q)
#time_consumption:00.000027
answer = get_answer(ids, checkpoint_path)
#time_consumption:03.158722
answer = vocabutil.get_text(answer)
#modi_answser 0:00:00.000066
print(answer)
except Exception:
traceback.print_exc()
except KeyboardInterrupt:
print("Ctrl+c exit.")