README.md

In mgoulet847/tagi: Tractable Approximate Gaussian Inference in Neural Networks

tagi

Tractable Approximate Gaussian Inference (TAGI) is a method used in Bayesian neural networks developped by Goulet et al. (2020). This package is its R implementation. From TAGI, the package enables inference of the weights and baises posterior distributions, treats uncertainty in all layers and uses a single observation at a time for online inference of model parameters. It supports many activation functions and can solve regression problems. Moreover, first and second derivatives calculations are also available using TAGI.

Installation

install.packages("devtools")
devtools::install_github("mgoulet847/tagi")

Examples

• Toy 1D Example: read vignette("ToyExample")
• Boston Housing Regression Problem: read vignette("BostonHousing")
• Insurance Regression Problem: read vignette("MedicalCost")
• Derivatives applied to the Toy 1D Example: read vignette("Derivatives")

Note: If the vignettes were not installed, you can run the following code. It will take approximately 25 minutes to run.

devtools::install(build_vignettes = TRUE)

mgoulet847/tagi documentation built on Dec. 21, 2021, 5:10 p.m.