API

Primary Interface

Lightning-UQ-Box.

A toolbox for Uncertainty Quantification in Deep Learning.

The lightning_uq_box package consists of various uncertainty quantification methods for deep learning models implemented in PyTorch and Lightning.

Package Reference

• lightning_uq_box.models
  • MLP
• lightning_uq_box.datamodules
  • Toy Datamodules
  • Image Regression Datamodules
• lightning_uq_box.uq_methods
• Single Forward Pass Methods
  • Mean Variance Estimation
  • Quantile Regression
  • Deep Evidential Regression
  • Zig Zag
  • Mixture Density Networks
• Approximate Bayesian Methods
  • Monte Carlo Dropout
  • Laplace Approximation
  • Bayesian Neural Networks ELBO
  • Bayesian Neural Networks with Alpha Divergence
  • Bayesian Neural Networks with Latent Variables (BNN-LV)
  • Stochastic Weight Averaging Gaussian (SWAG)
  • Stochastic Gradient Langevin Dynamics (SGLD)
  • Variational Bayes Last Layer
  • Spectral Normalized Gaussian Process (SNGP)
  • Deep Kernel Learning (DKL)
  • Deterministic Uncertainty Estimation (DUE)
  • Deep Ensembles
  • Masked Ensemble
  • Density Uncertainty Model
• Generative Models
  • Classification and Regression Diffusion (CARD)
  • Probabilistic UNet
• UQ Calibration Methods
  • Test Time Augmentation (TTA)
  • Conformal Quantile Regression
  • Temperature TempScaling
  • Regularized Adaptive Prediction Sets (RAPS)
  • Image to Image Conformal