You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for creating this library. When using fully factorized posterior structure I ran into a bug caused by zero values in weight variance, which can be easily fixed by setting the lower bound of clamp to be a very small number instead of zero.
Hi Yingzhen,
Thanks for creating this library. When using fully factorized posterior structure I ran into a bug caused by zero values in weight variance, which can be easily fixed by setting the lower bound of clamp to be a very small number instead of zero.
bayesianize/bnn/nn/mixins/variational/ffg.py
Line 102 in 6aab95e
I create a pull request to fix this small bug, would be great if you can take a look.
Kind regards,
Rui