CaDENCE-package {CaDENCE}

Conditional Density Estimation Network Construction and Evaluation (CaDENCE)

Description

A conditional density estimation network (CDEN) is a probabilistic extension of the standard multi-layer perceptron neural network (MLP) (Neuneier et al., 1994). A CDEN model allows users to estimate parameters of a specified probability distribution conditioned upon values of a set of predictors using the MLP architecture. The result is a flexible nonlinear model that can be used to calculate the conditional mean, variance, prediction intervals, etc. based on the specified distribution. Because the CDEN is based on the MLP, nonlinear relationships, including those involving complicated interactions between predictors, can be described by the modelling framework. The CaDENCE (Conditional Density Estimation Network Creation & Evaluation) package provides routines for creating and evaluating CDEN models in the R programming language.

Details

Procedures for fitting CaDENCE models are provided by cadence.fit, which relies on the standard optim function, the CaDENCE rprop function, or, optionally, the psoptim function from the pso package. Once a model has been developed, cadence.predict is used to evaluate the distribution parameters as a function of predictors.

The package also provides a variety of zero-inflated distributions, including the Bernoulli-gamma (bgamma), Bernoulli-Weibull (bweibull), Bernoulli-Pareto 2 (bpareto2), and Bernoulli-lognormal (blnorm), for use in the CaDENCE models.

gam.style, dummy.code, xval.buffer, and rbf are helper functions that may be useful for data preprocessing, model evaluation, and interpretation of fitted relationships.

Most other functions are used internally and should not normally need to be called directly by the user.

References

Cannon, A.J., 2012. Neural networks for probabilistic environmental prediction: Conditional Density Estimation Network Creation & Evaluation (CaDENCE) in R. Computers & Geosciences 41: 126-135. doi:10.1016/j.cageo.2011.08.023

Neuneier, R., F. Hergert, W. Finnoff, and D. Ormoneit, 1994., Estimation of conditional densities: a comparison of neural network approaches. In: M. Marinaro and P. Morasso (eds.), Proceedings of ICANN 94, Berlin, Springer, p. 689-692.