Tensorflow: Is it always more convenient to use InteractiveSession() compared to Session()?

Question

It seems to be more convenient to simply use something like sub.eval() instead of sess.run(eval), so would it be more convenient to always use InteractiveSession()? Are there any tradeoffs if we were to use InteractiveSession() all the time?

So far the only 'disadvantage' I see is that I can't use something like:

with tf.InteractiveSession() as sess:
   result = product.eval() #Where product is a simple matmul
   print result
   sess.close()

Instead I've to just define sess = tf.InteractiveSession right away.

score 1 · Accepted Answer · edited May 23 '17 at 12:30

From their implementation, the InteractiveSession sets itself as the default session and your subsequent eval() calls can use this session. You should be able to use the InteractiveSession in almost all the cases where you use Session.

One small difference is that you don't need to use InteractiveSession in a with block:

sess = tf.InteractiveSession()
# do your work
sess.close()

So don't forget to close the session after doing your work.

Here is an comparison between session.run() and eval(): In TensorFlow, what is the difference between Session.run() and Tensor.eval()?

Tensorflow: Is it always more convenient to use InteractiveSession() compared to Session()?

1 Answers1