removing variables from Tensorflow

Question

I'm trying some deep learning experiments with various hyper parameters. I build model for each hyper parameter settings separately. After first hyperparameter setting is trained and evaluated, when I try to build a new model with second setting, it gives me an error related to variable reused and stuff.

So I want to reset my session after each experiment. How can I do that?

I've tried tf.reset_default_graph(), but when I call sess.run(tf.global_variables_initializer()), it gives me following error:

ValueError: Fetch argument cannot be interpreted as a Tensor. (Operation name: "init" op: "NoOp" input: "^v/Assign" is not an element of this graph.)

How can I remove all variables and reset my session peacefully?

Answer 1

After you have reset the default graph, you will also need to create a new session. A small example:

import tensorflow as tf

hello = tf.constant('Hello, TensorFlow!')

sess = tf.Session()
print(sess.run(hello))

tf.reset_default_graph()
sess = tf.Session()  # Create new session
sess.run(tf.global_variables_initializer())