In BLNNdevs/BLNN: Bayesian Learning for Artificial Neural Networks

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(BLNN)

BLNN allows users to build, train, and make predictions using Bayesian neural networks. The package is built around four core functions all with 'BLNN_' at the begining. Those functions are

1) Build: Used to create the structure or shell of your network. Contains all the relevent information that needs to be passed along through the training process such as the desired network shape and what activation and output functions are to be used.

2) Train: This function takes the desired BLNN object, data, target values and training method to optimize the weights of the network. Returns the samples collected using bayesian methods. If non-Bayesian methods are used then a trained BLNN object is returned

3) Update: Allows the user to update their existing BLNN object with weights gathered from trained samples.

4) Predict: Takes a trained BLNN object along with data and calculates predicted values.

As with other Bayesian methods, propper diagnostics are required to determine if samples are adaquate in understanding your parameter space. To get the most out of diagnostics of the NUTS algorithm and the shinyBLNN function, it is necessary that users install shinystan though it is not necessary to have the library loaded.

install.packages("shinystan")

In an attempt to display how our BLNN methods can be used to investigate a variety of questions, we provide a series of three vignettes demonstrating linear, survival and classification modeling.

BLNNdevs/BLNN documentation built on Dec. 10, 2019, 3:31 a.m.