probly.train.evidential.torch

Unified Evidential Train Function.

Functions

der_loss	Deep Evidential Regression loss for uncertainty-aware regression.
dirichlet_entropy	Dirichlet entropy for predictive uncertainty estimation.
evidential_ce_loss	Evidential Cross Entropy Loss for classification uncertainty estimation.
evidential_kl_divergence	Evidential KL divergence loss for classification uncertainty estimation.
evidential_log_loss	Evidential Log Loss for classification uncertainty estimation.
evidential_mse_loss	Evidential Mean Squared Error loss for classification uncertainty estimation.
evidential_nignll_loss	Evidence-based Normal-Inverse-Gamma (NIG) regression loss.
evidential_regression_regularization	Regularization term for evidential regression.
ird_loss	Information Robust Dirichlet (IRD) loss for predictive uncertainty estimation.
kl_dirichlet	Compute KL(Dir(alpha_p) || Dir(alpha_q)) for each batch item.
lp_fn	Lp calibration loss for predictive uncertainty estimation.
make_in_domain_target_alpha	Construct target Dirichlet distribution for in-distribution samples.
make_ood_target_alpha	Construct flat Dirichlet target distribution for out-of-distribution samples.
natpn_loss	Natural Posterior Network (NatPN) classification loss.
normal_wishart_log_prob	Compute simplified univariate Normal-Wishart log-likelihood.
pn_loss	Paired ID/OOD training loss for Dirichlet Prior Networks.
postnet_loss	Posterior Networks (PostNet) classification loss.
predictive_probs	Expected categorical probabilities under Dirichlet.
regularization_fn	Regularization term for Information Robust Dirichlet Networks.
rpn_distillation_loss	Compute the distillation loss for Regression Prior Networks (RPN).
rpn_loss	Paired in-distribution and out-of-distribution loss for Regression Prior Networks.
rpn_ng_kl	KL divergence between two Normal-Gamma distributions.
rpn_prior	Normal-Gamma prior with zero evidence for Regression Prior Networks.
unified_evidential_train	Trains a given Neural Network using different learning approaches, depending on the approach of a selected paper.