R/starting.values.R

.start.smith <- function(data, coord, covariables, loc.form, scale.form, shape.form,
                         print.start.values = TRUE, method = "Nelder",
                         iso = TRUE, ...){

  n.site <- ncol(data)

  if (iso && (method == "Nelder"))
    method2 <- "BFGS"

  else if (method %in% c("nlm", "nlminb"))
    method2 <- "Nelder"

  else
    method2 <- method

  if (ncol(coord) == 2)
    param <- c("cov11", "cov12", "cov22")

  else
    param <- c("cov11", "cov12", "cov13", "cov22", "cov23", "cov33")

  fixed.param <- list(...)[names(list(...)) %in% param]
  idx.cov <- which(names(fixed.param) %in% param)
  fixed.param.cov <- fixed.param[idx.cov]
  fixed.param.gev <- fixed.param[-idx.cov]

  if (print.start.values)
    cat("Computing appropriate starting values\n")

  if (length(fixed.param.cov) > 0){
    args <- c(list(data = data, coord = coord, marge = "emp", iso = iso, method = method2),
              fixed.param.cov)
    covs <- do.call("fitcovmat", args)$fitted
  }

  else
    covs <- fitcovmat(data, coord, marge = "emp", iso = iso, method = method2)$fitted

  if (iso){
    covs <- covs[1]
    names(covs) <- "cov"
  }

  args <- c(list(data = data, covariables = as.matrix(covariables), loc.form = loc.form,
                 scale.form = scale.form, shape.form = shape.form, method = method, warn = FALSE), fixed.param.gev)

  spatgev <- do.call("fitspatgev", args)

  frech <- data
  gev <- predict(spatgev)
  for (i in 1:n.site)
    frech[,i] <- gev2frech(frech[,i], gev[i,"loc"], gev[i,"scale"], gev[i,"shape"])

  args <- c(list(data = frech, coord = coord, fit.marge = FALSE, start = as.list(covs),
                 iso = iso, warn = FALSE, method = method),
            fixed.param.cov)

  covs <- do.call("smithfull", args)

  start <- c(as.list(covs$fitted), as.list(spatgev$param))

  if (print.start.values){
    cat("Starting values are defined\n")
    cat("Starting values are:\n")
    print(c(covs$fitted, spatgev$fitted))
    cat("\n \n \n")
  }

  return(start)
}


.start.schlather <- function(data, coord, covariables, cov.mod, loc.form,
                             scale.form, shape.form, print.start.values = TRUE,
                             method = "Nelder", ...){

  n.site <- ncol(data)
  param <- c("nugget", "range", "smooth", "smooth2")
  fixed.param <- list(...)[names(list(...)) %in% param]

  idx.cov <- which(names(fixed.param) %in% param)
  fixed.param.cov <- fixed.param[idx.cov]
  fixed.param.gev <- fixed.param[-idx.cov]

  if (print.start.values)
    cat("Computing appropriate starting values\n")

  if (length(fixed.param.cov) > 0){
    args <- c(list(data = data, coord = coord, cov.mod = cov.mod,
                   marge = "emp"), fixed.param.cov)
    cov.param <- do.call("fitcovariance", args)$fitted
  }

  else
    cov.param <- fitcovariance(data, coord, cov.mod, marge = "emp")$fitted

  args <- c(list(data = data, covariables = as.matrix(covariables), loc.form = loc.form,
                 scale.form = scale.form, shape.form = shape.form,
                 method = method, warn = FALSE), fixed.param.gev)

  spatgev <- do.call("fitspatgev", args)

  frech <- data
  gev <- predict(spatgev)
  for (i in 1:n.site)
    frech[,i] <- gev2frech(frech[,i], gev[i,"loc"], gev[i,"scale"], gev[i,"shape"])

  args <- c(list(data = frech, coord = coord, cov.mod = cov.mod, fit.marge = FALSE,
                 start = as.list(cov.param), warn = FALSE, method = method), fixed.param.cov)

  cov.param <- do.call("schlatherfull", args)

  start <- c(as.list(cov.param$param), as.list(spatgev$param))

  if (print.start.values){
    cat("Starting values are defined\n")
    cat("Starting values are:\n")
    print(c(cov.param$fitted, spatgev$fitted))
  }

  return(start)
}


.start.schlatherind <- function(data, coord, covariables, cov.mod, loc.form,
                                scale.form, shape.form, print.start.values = TRUE,
                                method = "Nelder", ...){

  n.site <- ncol(data)
  param <- c("alpha", "nugget", "range", "smooth", "smooth2")
  fixed.param <- list(...)[names(list(...)) %in% param]

  idx.cov <- which(names(fixed.param) %in% param)
  fixed.param.cov <- fixed.param[idx.cov]
  fixed.param.gev <- fixed.param[-idx.cov]

  if (print.start.values)
    cat("Computing appropriate starting values\n")

  if (length(fixed.param.cov) > 0){
    args <- c(list(data = data, coord = coord, cov.mod = paste("i", cov.mod, sep=""),
                   marge = "emp"), fixed.param.cov)
    cov.param <- do.call("fitcovariance", args)$fitted
  }

  else
    cov.param <- fitcovariance(data, coord, paste("i", cov.mod, sep=""),
                               marge = "emp")$fitted

  args <- c(list(data = data, covariables = as.matrix(covariables), loc.form = loc.form,
                 scale.form = scale.form, shape.form = shape.form,
                 method = method, warn = FALSE), fixed.param.gev)

  spatgev <- do.call("fitspatgev", args)

  frech <- data
  gev <- predict(spatgev)
  for (i in 1:n.site)
    frech[,i] <- gev2frech(frech[,i], gev[i,"loc"], gev[i,"scale"], gev[i,"shape"])

  args <- c(list(data = frech, coord = coord, cov.mod = cov.mod, fit.marge = FALSE,
                 start = as.list(cov.param), warn = FALSE, method = method), fixed.param.cov)

  cov.param <- do.call("schlatherindfull", args)

  start <- c(as.list(cov.param$param), as.list(spatgev$param))

  if (print.start.values){
    cat("Starting values are defined\n")
    cat("Starting values are:\n")
    print(c(cov.param$fitted, spatgev$fitted))
  }

  return(start)
}

.start.geomgauss <- function(data, coord, covariables, cov.mod, loc.form,
                             scale.form, shape.form, print.start.values = TRUE,
                             method = "Nelder", ...){

  n.site <- ncol(data)
  param <- c("sigma2", "nugget", "range", "smooth", "smooth2")
  fixed.param <- list(...)[names(list(...)) %in% param]

  idx.cov <- which(names(fixed.param) %in% param)
  fixed.param.cov <- fixed.param[idx.cov]
  fixed.param.gev <- fixed.param[-idx.cov]

  if (print.start.values)
    cat("Computing appropriate starting values\n")

  if (length(fixed.param.cov) > 0){
    args <- c(list(data = data, coord = coord, cov.mod = paste("g", cov.mod, sep=""),
                   marge = "emp"), fixed.param.cov)
    cov.param <- do.call("fitcovariance", args)$fitted
  }

  else
    cov.param <- fitcovariance(data, coord, paste("g", cov.mod, sep=""),
                               marge = "emp")$fitted

  args <- c(list(data = data, covariables = as.matrix(covariables), loc.form = loc.form,
                 scale.form = scale.form, shape.form = shape.form,
                 method = method, warn = FALSE), fixed.param.gev)

  spatgev <- do.call("fitspatgev", args)

  frech <- data
  gev <- predict(spatgev)
  for (i in 1:n.site)
    frech[,i] <- gev2frech(frech[,i], gev[i,"loc"], gev[i,"scale"], gev[i,"shape"])

  args <- c(list(data = frech, coord = coord, cov.mod = cov.mod, fit.marge = FALSE,
                 start = as.list(cov.param), warn = FALSE, method = method), fixed.param.cov)

  cov.param <- do.call("geomgaussfull", args)

  start <- c(as.list(cov.param$param), as.list(spatgev$param))

  if (print.start.values){
    cat("Starting values are defined\n")
    cat("Starting values are:\n")
    print(c(cov.param$fitted, spatgev$fitted))
  }

  return(start)
}

.start.extremalt <- function(data, coord, covariables, cov.mod, loc.form,
                             scale.form, shape.form, print.start.values = TRUE,
                             method = "Nelder", ...){

  n.site <- ncol(data)
  param <- c("nugget", "range", "smooth", "smooth2", "DoF")
  fixed.param <- list(...)[names(list(...)) %in% param]

  idx.cov <- which(names(fixed.param) %in% param)
  fixed.param.cov <- fixed.param[idx.cov]
  fixed.param.gev <- fixed.param[-idx.cov]

  if (print.start.values)
    cat("Computing appropriate starting values\n")

  if (length(fixed.param.cov) > 0){
    args <- c(list(data = data, coord = coord, cov.mod = paste("t", cov.mod, sep=""),
                   marge = "emp"), fixed.param.cov)
    cov.param <- do.call("fitcovariance", args)$fitted
  }

  else
    cov.param <- fitcovariance(data, coord, paste("t", cov.mod, sep=""),
                               marge = "emp")$fitted

  args <- c(list(data = data, covariables = as.matrix(covariables), loc.form = loc.form,
                 scale.form = scale.form, shape.form = shape.form,
                 method = method, warn = FALSE), fixed.param.gev)

  spatgev <- do.call("fitspatgev", args)

  frech <- data
  gev <- predict(spatgev)
  for (i in 1:n.site)
    frech[,i] <- gev2frech(frech[,i], gev[i,"loc"], gev[i,"scale"], gev[i,"shape"])

  args <- c(list(data = frech, coord = coord, cov.mod = cov.mod, fit.marge = FALSE,
                 start = as.list(cov.param), warn = FALSE, method = method), fixed.param.cov)

  cov.param <- do.call("extremaltfull", args)

  start <- c(as.list(cov.param$param), as.list(spatgev$param))

  if (print.start.values){
    cat("Starting values are defined\n")
    cat("Starting values are:\n")
    print(c(cov.param$fitted, spatgev$fitted))
  }

  return(start)
}

.start.nsgeomgauss <- function(data, coord, covariables, cov.mod, loc.form,
                               scale.form, shape.form, sigma2.form,
                               print.start.values = TRUE, method = "Nelder",
                               ...){

  n.site <- ncol(data)
  sigma2.terms <- terms(sigma2.form)

  if (length(attributes(sigma2.terms)$factor) == 0)
    n.sigma2coeff <- attributes(sigma2.terms)$intercept

  else
    n.sigma2coeff <- attributes(sigma2.terms)$intercept +
      ncol(attributes(sigma2.terms)$factors)

  if (n.sigma2coeff == 1)
    sigma2.names <- "sigma2"

  else
    sigma2.names <- paste("sigma2Coeff", 1:n.sigma2coeff, sep="")

  param <- c(sigma2.names, "nugget", "range", "smooth", "smooth2")
  fixed.param <- list(...)[names(list(...)) %in% param]

  idx.cov <- which(names(fixed.param) %in% param)
  fixed.param.cov <- fixed.param[idx.cov]
  fixed.param.gev <- fixed.param[-idx.cov]

  if (print.start.values)
    cat("Computing appropriate starting values\n")

  if (length(fixed.param.cov) > 0){
    args <- c(list(data = data, coord = coord, cov.mod = cov.mod,
                   marge = "emp"), fixed.param.cov)
    cov.param <- do.call("fitcovariance", args)$fitted
  }

  else
    cov.param <- fitcovariance(data, coord, cov.mod, marge = "emp")$fitted

  args <- c(list(data = data, covariables = as.matrix(covariables), loc.form = loc.form,
                 scale.form = scale.form, shape.form = shape.form,
                 method = method, warn = FALSE), fixed.param.gev)

  spatgev <- do.call("fitspatgev", args)

  frech <- data
  gev <- predict(spatgev)
  for (i in 1:n.site)
    frech[,i] <- gev2frech(frech[,i], gev[i,"loc"], gev[i,"scale"], gev[i,"shape"])

  sigma2param <- rep(0, length(sigma2.names))
  names(sigma2param) <- sigma2.names

  args <- c(list(data = frech, coord = coord, cov.mod = cov.mod, fit.marge = FALSE,
                 sigma2.form = sigma2.form, warn = FALSE, method = method, start =
                 c(as.list(sigma2param), as.list(cov.param))),
            fixed.param.cov)

  cov.param <- do.call("nsgeomgaussfull", args)

  start <- c(as.list(cov.param$param), as.list(spatgev$param))

  if (print.start.values){
    cat("Starting values are defined\n")
    cat("Starting values are:\n")
    print(c(cov.param$fitted, spatgev$fitted))
  }

  return(start)
}

.start.brownresnick <- function(data, coord, covariables, loc.form, scale.form,
                                shape.form, print.start.values = TRUE,
                                method = "Nelder", ...){

  n.site <- ncol(data)
  param <- c("range", "smooth")
  fixed.param <- list(...)[names(list(...)) %in% param]

  idx.cov <- which(names(fixed.param) %in% param)
  fixed.param.cov <- fixed.param[idx.cov]
  fixed.param.gev <- fixed.param[-idx.cov]

  if (print.start.values)
    cat("Computing appropriate starting values\n")

  if (length(fixed.param.cov) > 0){
    args <- c(list(data = data, coord = coord, cov.mod = "brown",
                   marge = "emp"), fixed.param.cov)
    cov.param <- do.call("fitcovariance", args)$fitted
  }

  else
    cov.param <- fitcovariance(data, coord, "brown", marge = "emp")$fitted

  args <- c(list(data = data, covariables = as.matrix(covariables), loc.form = loc.form,
                 scale.form = scale.form, shape.form = shape.form,
                 method = method, warn = FALSE), fixed.param.gev)

  spatgev <- do.call("fitspatgev", args)

  frech <- data
  gev <- predict(spatgev)
  for (i in 1:n.site)
    frech[,i] <- gev2frech(frech[,i], gev[i,"loc"], gev[i,"scale"], gev[i,"shape"])

  args <- c(list(data = frech, coord = coord, fit.marge = FALSE,
                 start = as.list(cov.param), warn = FALSE, method = method), fixed.param.cov)

  cov.param <- do.call("brownresnickfull", args)

  start <- c(as.list(cov.param$param), as.list(spatgev$param))

  if (print.start.values){
    cat("Starting values are defined\n")
    cat("Starting values are:\n")
    print(c(cov.param$fitted, spatgev$fitted))
  }

  return(start)
}

Try the SpatialExtremes package in your browser

Any scripts or data that you put into this service are public.

SpatialExtremes documentation built on May 2, 2019, 5:45 p.m.