I have two questions, first:
In the constructor of linear_bayesian_layer.py, where do the default values of the parameters come from?
Second:
How am I to understand the distributions from which we sample our weights and biases? they have simply a mu and rho but surely they cannot all be regular normal distributions, they are probably more akin to mixture models , no?.Should I want to model the distributions, how should I go at it? Thanks
Hello,
Thanks for the library, it is quite useful!
I have two questions, first: In the constructor of linear_bayesian_layer.py, where do the default values of the parameters come from?![image](https://user-images.githubusercontent.com/83427055/129930852-521d63b2-45ca-4fae-9f46-8156443ceb5b.png)
Second: How am I to understand the distributions from which we sample our weights and biases? they have simply a mu and rho but surely they cannot all be regular normal distributions, they are probably more akin to mixture models , no?.Should I want to model the distributions, how should I go at it? Thanks