-
Notifications
You must be signed in to change notification settings - Fork 74.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Frame of the Variable #4478
Comments
It sounds like you are trying to use external tensors in scan (ie, On Mon, Sep 19, 2016 at 10:59 PM, Linara Adilova notifications@github.com
|
Yes, I am, because I have weights variables that need to be initialized outside. |
Please provide the critical part of the code. On Sep 20, 2016 1:07 PM, "Linara Adilova" notifications@github.com wrote:
|
p.s. is this something for which you can use tf.get_variable? On Sep 20, 2016 2:15 PM, wrote: Please provide the critical part of the code. On Sep 20, 2016 1:07 PM, "Linara Adilova" notifications@github.com wrote: Yes, I am, because I have weights variables that need to be initialized — |
And after this calling initialize_all_variables() to run causes that error
What exactly this call will do? |
I'm having the same error message that is reported in the issue(s). I have a RNN like structure that has some building blocks (component neural networks) that are passed in by the user. Here is a minimal example:
with the error ending in:
With
However, still the same error Of course, moving contents of the external function into the What am I doing wrong? |
This is probably related to #2211 |
Can you do this without the templating mechanism? On Oct 21, 2016 2:55 PM, "Christian" notifications@github.com wrote:
|
No, does not work without templating either. The second code part in the previous post tries this, leading to the same error message. |
Yuan pushed improved error messaging today. Can you try using a fresh build On Oct 25, 2016 2:11 PM, "Christian" notifications@github.com wrote:
|
I have the same problem. Passing an initializer function instead of a tensor/constant as |
@akosiorek Thank you! I didn't try your approach, but a similar one suggested by @mrry here: http://stackoverflow.com/questions/42564698/invalidargumenterror-the-node-has-inputs-from-different-frames. A |
Automatically closing due to lack of recent activity. Since this issue is old at this point, please reopen the issue if it still occurs when tried with the latest version of Tensorflow. Thank you. |
Will now construct while loops, but cannot actually run them due to tensorflow.python.framework.errors_impl.InvalidArgumentError: The node 'while_1/Variable/Assign' has inputs from different frames. The input 'while_1/Size' is in frame 'while_1/while_1/'. The input 'while_1/Variable' is in frame ‘’., seems to be related to tensorflow/tensorflow#4478, but I don’t believe I’m passing in external tensors. Once this is fixed, however, should be able to apply same structure to a left-propagating update of the weights, after which the whole thing should (?) be done.
Hello
I am using tf.scan for implementing memory augmented network and when trying to run tf.initialize_all_variables() getting the error about frame of the variables:
As I understood this frame is some inner identification of the variables, so what should I do to get rid of this error? Scan works if I do not define any variables outside the step, but I need at least weights-biases and loss-accuracy variables.
The text was updated successfully, but these errors were encountered: