be able to pass init.vals, so that in iterations init.vals need to only be calculated once

R/gevdfit.R

@@ -31,6 +31,7 @@
 #' @param show Logical; if TRUE (the default), print details of the fit.
 #' @param method The optimization method used in \code{\link{optim}}.
 #' @param maxit The maximum number of iterations.
+#' @param init.vals vector of length 5, giving initial values for parameter intercepts
 #' @param ... Other control parameters for the optimization.
 #' @return A list containing the following components. 
 #' A subset of these components are printed after the fit. 
@@ -47,6 +48,9 @@
 #' A zero indicates successful convergence.}
 #' \item{data}{data is standardized to standart Gumbel.} 
 #' \item{cov}{The covariance matrix.} 
+#' \item{vals}{Parameter values for every data point.}
+#' \item{init.vals}{Initial values that where used.}
+#' \item{ds}{Durations for every data point.}
 #' @seealso \code{\link{dgev.d}}, \code{\link{IDF.agg}}, \code{\link{gev.fit}}, \code{\link{optim}}
 #' @export
 #' @importFrom stats optim 
@@ -69,7 +73,7 @@ gev.d.fit<-
   function(xdat, ds, ydat = NULL, mul = NULL, sigl = NULL, shl = NULL, thetal = NULL, etal = NULL, 
            mulink = identity, siglink = identity, shlink = identity, thetalink = identity, etalink = identity,  
            muinit = NULL, siginit = NULL, shinit = NULL, thetainit = NULL, etainit = NULL,
-           show = TRUE, method = "Nelder-Mead", maxit = 10000, ...)
+           show = TRUE, method = "Nelder-Mead", maxit = 10000, init.vals = NULL, ...)
   {
     #
     # obtains mles etc for d-gev distn
@@ -88,7 +92,9 @@ gev.d.fit<-
     z$trans <- FALSE  # indicates if fit is non-stationary
     
     # calculate initial values for mu.d, sigma_0, xi, eta using IDF.init:  (thetainit=0)
-    if(any(is.null(c(muinit,siginit,shinit,etainit)))){
+    if(!is.null(init.vals)){init.vals <- data.frame(mu = init.vals[1], sigma = init.vals[2], xi = init.vals[3]
+                                                    ,theta = init.vals[4], eta = init.vals[5])}
+    if(any(is.null(c(muinit,siginit,shinit,etainit)))& is.null(init.vals)){
       init.vals <- gev.d.init(xdat,ds,ifelse(is.null(thetainit),0,thetainit[1]))
     }
     # TODO: transform initial values with link function
@@ -217,6 +223,7 @@ gev.d.fit<-
         }
     z$se <- sqrt(diag(z$cov)) # sqrt(digonal entries) = standart error of mle's 
     z$vals <- cbind(mut, sc0, xi, theta, eta)
+    z$init.vals <- as.numeric(init.vals)
     colnames(z$vals) <- c('mut','sigma0','xi','theta','eta')
     z$ds <- ds
     if(show) {

man/gev.d.fit.Rd

@@ -9,7 +9,7 @@ gev.d.fit(xdat, ds, ydat = NULL, mul = NULL, sigl = NULL,
   siglink = identity, shlink = identity, thetalink = identity,
   etalink = identity, muinit = NULL, siginit = NULL, shinit = NULL,
   thetainit = NULL, etainit = NULL, show = TRUE,
-  method = "Nelder-Mead", maxit = 10000, ...)
+  method = "Nelder-Mead", maxit = 10000, init.vals = NULL, ...)
 }
 \arguments{
 \item{xdat}{A vector containing maxima for different durations. 
@@ -41,6 +41,8 @@ duration dependency of the location and shape parameter.}
 
 \item{maxit}{The maximum number of iterations.}
 
+\item{init.vals}{vector of length 5, giving initial values for parameter intercepts}
+
 \item{...}{Other control parameters for the optimization.}
 }
 \value{
@@ -58,7 +60,10 @@ duration offset and duration exponent, resp.}
 \item{conv}{The convergence code, taken from the list returned by \code{\link{optim}}. 
 A zero indicates successful convergence.}
 \item{data}{data is standardized to standart Gumbel.} 
-\item{cov}{The covariance matrix.}
+\item{cov}{The covariance matrix.} 
+\item{vals}{Parameter values for every data point.}
+\item{init.vals}{Initial values that where used.}
+\item{ds}{Durations for every data point.}
 }
 \description{
 Modified \code{\link[ismev]{gev.fit}} function for Maximum-likelihood fitting