grnnet: Generalized Regression Neural Network

In dsnavega/grnnet: Generalized Regression Neural Network

Description

Implementation of the Generalized Regression Neural Network architecture for Single and Multi-Output Regression with Uncertainty Estimation Using Conformal Prediction.

Usage

grnnet(x, y, alpha = 0.05, crossvalidation = T, kfold = 5,
  seed = 1984)

Arguments

x	a matrix of numeric inputs
y	a matrix of numeric outputs
alpha	a numeric value. confidence level (i.e, 0.05)
crossvalidation	a logical indicating if k-fold cross-validation should be performed. Default is TRUE (T)
kfold	number of fold for k-fold cross-validation
seed	seed for the random number generator. Controls reproducibility of k-fold cross-validation

Value

a trained generalized regression neural network

Note

The network training is performed with an internal leave-one-out sampling loop and derivate free optimization (Brent's algorithm). The bounds of the sigma parameter, the spread of the squared exponential kernel, are initialized following Tomandl & Schober (2001) based on the proprieties of the distance matrix between cases. The internal loop also computes the conformity factors needed to compute the prediction intervals based on conformal prediction.

Author(s)

David Navega

dsnavega/grnnet documentation built on May 9, 2019, 5 a.m.