分佈式模型保存時,提示:TypeError: ‘sess’ must be a Session
具體如下:
File "/users/spraak/spch/prog/spch/tensorflow-1.0.0/lib/python2.7/site-packages/tensorflow/python/training/saver.py", line 1369, in save
raise TypeError("'sess' must be a Session; %s" % sess)
TypeError: 'sess' must be a Session; <tensorflow.python.training.monitored_session.MonitoredSession object at 0x2be710d0>
出現這種情況時的代碼大概爲:
with tf.train.MonitoredTrainingSession(...) as sess:
...
saver.save(sess, 'model.ckpt')
查找問題時,發現了一個解決辦法:
地址爲:https://github.com/tensorflow/tensorflow/issues/8425
解決辦法:
實現一個函數獲得真實的session
def get_session(sess):
session = sess
while type(session).__name__ != 'Session':
#pylint: disable=W0212
session = session._sess
return session
在保存模型時,使用如下方式調用
saver.save(get_session(sess), 'model.ckpt')