search

kumar-shridhar / PyTorch-BayesianCNN

Bayesian Convolutional Neural Network with Variational Inference based on Bayes by Backprop in PyTorch.
MIT License
1.42k stars 323 forks source link

about uncertainties #15

Closed ShellingFord221 closed 5 years ago

ShellingFord221 commented 5 years ago

Hi, are there any possibilities to compute aleatoric uncertainty and epistemic uncertainty of each class of the model (BCNN), other than just one sample?

kumar-shridhar commented 5 years ago

Hi. Yes, it is possible. I will put the updated code on this repo in 2 days.

However, we calculate the uncertainties on the test data on a given batch. So class wise uncertainty can only be derived on the validation set.