-
Notifications
You must be signed in to change notification settings - Fork 74k
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
MultivariateNormalDiag probability gradient fails #10766
Comments
Also fails on the released version of tensorflow (v1.2.0) built from source. |
Also fails when |
I had a similar problem as @jakedailey1 which is why I distilled this example down from a Mixture Density Network with a likelihood-based loss function. |
This turns out to be an issue with |
Wow! Just when I thought I was stuck I find there's a solution already on the way. Nice work :D How is it coming along? |
It looks like a fix has been put together here. I can't try this out right now, but perhaps you can just grab the latest files and it will work. |
I can confirm that ea79ba4 (mentioned above) fixes the issue. I cherry-picked it to v1.2.1 to verify. The fix is available on master. |
I upgraded my Tensorflow to v1.2.1. But I am still facing the same error message when I am trying to train a model to learn an appropriate sigma value for the Gaussian filter. Am I doing something wrong? My system is Win10. I installed Tensorflow through Anaonda. Here is my code.
Here are the error messages.
|
v1.2.1 does not have the fix. The fix is on master. You need to cherry-pick the commit I mentioned above (ea79ba4) onto your v1.2.1 branch. |
Oh now I see! Thank you so much @ktarplee ! |
- http://stackoverflow.com/q/45109305 - tensorflow#10766 PiperOrigin-RevId: 162026912
- http://stackoverflow.com/q/45109305 - tensorflow#10766 PiperOrigin-RevId: 162026912
System information
Describe the problem
The gradients for the tensorflow.contrib.distributions.MultiVariateNormalDiag throw an error. This might be related to #10149. Notice that the tensorflow.contrib.distributions.Normal works as expected so I think this is a bug in the gradient implementation of the MultiVarateNormalDiag.
Source code
Output
The text was updated successfully, but these errors were encountered: