Whenever using tensorflow variables's name, I have had a question why they ended by blabla:0. So I should use split function like v.name.split(':')[0].
Despite there is same name's variables, the tail's name wouldn't be changed, like my expectation aaa:0, aaa:1.
Does anybody know why tensorflow does this?
I couldn't find this reason on tensorflow.org or via googling.
The offending integer is added in the Tensor class, and is the "value index":
value_index: An
int. Index of the operation's endpoint that produces this tensor.
It's not always zero! For example:
>>> import tensorflow as tf
>>> tf.unique([1,2,3])
Prints:
Unique(y=<tf.Tensor 'Unique:0' shape=(?,) dtype=int32>, idx=<tf.Tensor 'Unique:1' shape=(3,) dtype=int32>)
So the first output has the ":0" suffix, but the second has a ":1" suffix since it's the second output of the op.
