Tensorflow : Assign variable by name

Question

I would like to assign/modify the values of my variables and I wish to do so by calling them by their name.

For exemple:

vars = tf.trainable_variables()
print(vars[1].name)

'matrix1:0'

upt = vars['matrix1:0'].assign_add(tf.constant(1))
sess.run(upt)

The reason for this is that indexes for variables aren't reliable as they are dependant as to when they are ran in the code. By adding a new variable, all the indexes would need to be shifted, which is not convenient. Using names would make my life much easier.

How about vars={v.name:v for v in tf.trainable_variables()} — Yaroslav Bulatov
– Yaroslav Bulatov, Commented Jun 8, 2016 at 18:57
Now, how do you assign new values to the variables by calling them by their name? — PhABC
– PhABC, Commented Jun 8, 2016 at 19:30
And then you can do vars['matrix1:0'].assign_add(tf.constant(1)) — Yaroslav Bulatov
– Yaroslav Bulatov, Commented Jun 8, 2016 at 19:40
Great, works well! Thank you. Mind adding a reply so I can mark your comment as the answer? — PhABC
– PhABC, Commented Jun 8, 2016 at 20:12

Yaroslav Bulatov · Accepted Answer · 2016-06-08 20:40:34Z

3

You can use Python generator expression to construct a dictionary like this

vars={v.name:v for v in tf.trainable_variables()}

and then you modify the variable as

vars['matrix1:0'].assign_add(tf.constant(1))

answered Jun 8, 2016 at 20:40

Yaroslav Bulatov

58.1k23 gold badges147 silver badges210 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Tensorflow : Assign variable by name

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related