Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.
And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.
Reparametrization in VGAE model can cause NAN's
See original GitHub issue❓ Questions & Help
While, the implementation of the reparametrization is clean and conventional it can potentially lead to NaN's as the self.__logvar__ term is not controlled. This behavior is dependent on the use case, and for the most part will not cause problems for many applications but I believe a better fix could be something like to add another line after : https://github.com/rusty1s/pytorch_geometric/blob/master/torch_geometric/nn/models/autoencoder.py#L231
self.__logvar__ = torch.clamp(self.__logvar__,min=LOG_VAR_MIN,max=LOG_VAR_MAX)
Where, LOG_VAR_MIN and LOG_VAR_MAX are suitable constants. This prevents numerical issues (underflow/overflow) that can happen during optimization.
Issue Analytics
- State:
- Created 4 years ago
- Comments:14 (7 by maintainers)
Top Results From Across the Web
Top Related Medium Post
Top Related StackOverflow Question
Troubleshoot Live Code
Top Related Reddit Thread
Top Related Hackernoon Post
Top Related Tweet
Top Related Dev.to Post
Top Related Hashnode Post
Added to master. Thank you!
Small correction
LOG_VAR_MAX=20blows up still so a lower value like10is more stable. But other than that everything is as you predicted.