tensorflow cifar10_eval.py error:RuntimeError: Attempted to use a closed Session.RuntimeError: Attempted to use a closed Session

Question

I\'m running the cifar10 network on my PC and after finishing the training and running eval script the following error appears:

2016-06-01 14:37:14.238317: p         


        

Answer 1

Looking at the code you posted, the problem is between lines 50 and 51 in eval_once():

with tf.Session() as sess:
    ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
    if ckpt and ckpt.model_checkpoint_path:
        saver.restore(sess, ckpt.model_checkpoint_path)
        global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
    else:
        print('No checkpoint file found')
        return
# <<< The Session is closed here >>>
coord = tf.train.Coordinator()
try:
    # ...

When the code exits a with tf.Session() as sess: block, sess is automatically closed, and you cannot use it any more. There are (at least) two ways to fix this problem:

1. Indent lines 51 through 76 by 4 spaces, so that they are also inside the with block.

2. Create the session without using a with block and close it manually:

   def eval_once():
       sess = tf.Session()
       ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
       if ckpt and ckpt.model_checkpoint_path:
           saver.restore(sess, ckpt.model_checkpoint_path)
           global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
       else:
           print('No checkpoint file found')
           sess.close()
           return
   
       coord = tf.train.Coordinator()
       try:
           # ...
       finally:
           sess.close()