As one of the ways to evaluate different perspectives for model selection, I’m trying to build Bayesian Neural Network and sample parameters for various evaluations.
This is the implementation notebook of Bayesian NN that attempts to evaluate it from a different perspective than the previous one.
Based on the experiment results of this Notebook, I have found the following
- The ensemble evaluation by Bayesian NN is higher than the average of the distribution of the evaluations, so the ensemble is working effectively in NN.
- Uncertainty that can be computed from Bayesian NN may be able to detect the Burn Era.
And I hope it helps for anyone!