README.md

In dsnavega/grnnet: Generalized Regression Neural Network

grnnet - An R Implementation of Generalized Regression Neural Network

About

grnnet is currently under active development.

grnnet implements the Generalized Regression Neural Network (GRNN) architecture by Specht (1991).

A GRNN is a variation of radial basis neural and can be related to nonparametric regression. In its essence a GRNN can be seen as a formulation of the k-nearest neighbor algorithm in the terminology of neural networks, where each instance is regarded as artificial neuron and its activation is given by a radial basis function or any other suitable kernel-like function.

The current package allows to create single- and multi-output regression neural network and produces heteroscedastic predictive interval using the conformal prediction framework.

A grnnet object is trained with an internal leave-one-out cross-validation loop and statistically validated with K-fold cross-validation loop.

TODO

• Implement conditional density estimation using this network architecture.

References

Specht, D. F. (1991). A general regression neural network. IEEE Transactions on Neural Networks, 2(6), 568-576.

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