negnormL {MQMF}

negnormL an alternative -log-likelihood for normal errors

Description

negnormL is an alternative to negNLL to produce -ve log-likelihoods for normal random errors, allowing for the sigma parameter to vary as a function of the predicted value. In negnormL only one needs both a funk and a funksig, the former to calculate the predicted values using funk, and funksig to calculate the changing sigma values relative to the predicted values. The example code illustrates an example funksig that does nothing to the sigma value. See the chapter on Static Models to see further examples.

Usage

negnormL(pars, funk, funksig, indat, obs = "dl", ...)

Arguments

pars

the vector of parameters, with sigma, the standard deviation of the normal random deviates at the end.

funk

the funk needed to generate the predicted values

funksig

the function used to calculate the sigma value based on the last parameter (=constant sigma) and the predicted values

indat

the data.frame or matrix containing the obs and the independent variable used by funk

obs

identifies the column name or column number that contains the observed data for comparison with the predicted from funk, the default="dl"

...

the standard R for including extra parameters needed by funk but without having to be explicitly defined.

Value

the negative log-likelihood, for use in an optimizer, eg nlm

Examples

 data(blackisland)
 param <- c(Linf=173.65,K=0.2666,sigma=3.6)
 sigfunk <- function(pars,predobs) return(tail(pars,1))
 negnormL(par=param,funk=fabens,funksig=sigfunk,indat=blackisland,
          obs="dl",initL="l1",delT="dt")   # should be 291.1757
 param2 <- c(21.03,130.94,40.65,3.162)  
 negnormL(par=param2,funk=invl,funksig=sigfunk,indat=blackisland,
           obs="dl",initL="l1",delT="dt")  # should be 277.0186