Reputation: 770
I'm training a convnet with Tensorflow but I'm having troubles with it, so I'm trying to figure out what's happening. I've defined some variable in the following way:
with tf.variable_scope('myscope') as scope:
var = some_value
But when I try to retrieve it:
var = tf.get_variable('myscope/var')
I get the following:
ValueError: Shape of a new variable (myscope/var) must be fully defined, but instead was <unknown>.
I think there might be something wrong with my tensors, because when I print the output shape I get (?, 10)
while I should have the batch size instead of the ?
. Here's my code for completeness.
Upvotes: 0
Views: 740
Reputation: 11
I had experienced the same problem. The shape is undefined if it has a ?, which means it can take any value. It works well if you manually define your shape (like [10, 10] instead of [?,10]) while declaring the tensor variable.
Upvotes: 0