## Authors 
## Martin Schlather, schlather@math.uni-mannheim.de
##
##
## Copyright (C) 2015 Martin Schlather
##
## This program is free software; you can redistribute it and/or
## modify it under the terms of the GNU General Public License
## as published by the Free Software Foundation; either version 3
## of the License, or (at your option) any later version.
##
## This program is distributed in the hope that it will be useful,
## but WITHOUT ANY WARRANTY; without even the implied warranty of
## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
## GNU General Public License for more details.
##
## You should have received a copy of the GNU General Public License
## along with this program; if not, write to the Free Software
## Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.  

DeleteCoord <- function(coord, nr) {
  if (coord$grid || length(coord$T)>0)
    stop("deleting from a grid not possible")
  if (length(coord$y))
    stop("deleting within a kernel definition not allowed")
  if (coord$dist.given) {
    stop("deleting from distances not programmed yet")
  } else {
    coord$restotal <- coord$restotal - 1
    coord$l <- coord$l - 1
    coord$x <- coord$x[-nr, , drop=FALSE]
  }
  coord
}


summary.RFcrossvalidate <- function(object, ...) {
  ## cf geoR
  if (is.null(object$error)) {
    #Print(names(object))
    res <- list()
    for (i in 1:length(object)) {
      res[[i]] <- summary.RFcrossvalidate(object[[i]])
    }
    names(res) <- names(object)
  } else {
    F <- function(object) {    
      paste(formatC("sd", flag="-", width=7), ":", sep="",
            formatC(object, digits=2, width=6, format = "f"))
    }
    res <- list()
    res$error <- rbind(summary(object$error), F(apply(object$error, 2, sd)))
    res$std.error <- rbind(summary(object$std.error), F(apply(object$std.error, 2, sd)))
  }
  class(res) <- "summary.RFcrossvalidate"  
  return(res)
}


print.RFcrossvalidate <- function(x, ...) {
#  str(x)
  print.summary.RFcrossvalidate(summary.RFcrossvalidate(x, ...))
}

print.summary.RFcrossvalidate <- function(x, ...) {
  ## cf geoR
  if (is.null(x$error)) {
    for (i in 1:length(x)) {
      cat("\n", names(x)[i], "\n")
      print.summary.RFcrossvalidate(x[[i]])
    }
    return(invisible(x))
  } else {
    colnames(x$error) <- paste("err:", colnames(x$error), sep="")
    colnames(x$std.error) <- paste("std:", colnames(x$std.error), sep="")
    res <- cbind(x$error, x$std.error)
    print(res, quote=FALSE) #
    return(invisible(res))
  }
}

RFcrossvalidate <- function(model, x, y=NULL, z=NULL, T=NULL, grid=NULL, data,
                            lower=NULL, upper=NULL,
                            method="ml", # "reml", "rml1"),
                            users.guess=NULL,  
                            distances=NULL,
                            dim, ## space time dim -- later a vector of grouped dimensions
                            optim.control=NULL,
                            transform=NULL,
                            full = FALSE,
                            ...) {

  RFoptOld <- internal.rfoptions(seed=NA, ...)
  on.exit(RFoptions(LIST=RFoptOld[[1]]))
  RFopt <- RFoptOld[[2]]
  refit <- RFopt$fit$cross_refit
  return.variance <- RFopt$krige$return.variance
  general <- RFopt$general
  basic <- RFopt$basic
  printlevel <- basic$printlevel
  details <- general$detailed_output
  if (general$modus_operandi == "neurotic") stop("crossvalidation is not a precise method")
  if (length(method) != 1) stop("exactly 1 method must be given")
   pch <- general$pch
  if (printlevel < PL_IMPORTANT) pch <- ""
 
  if (method %in%LSQMETHODS){
    sub.methods <- method
    methods <- NULL
  } else if (method %in% MLMETHODS) {
    sub.methods <- LSQMETHODS
    methods <- method
  } else stop("unknown method")
  
  res <- list()
  if (full) {
    ats <- approx_test_single(model, method=method)
    result <- ats$result
    models <- ats$fitted.model
  } else {
    if (class(model) == "RF_fit") {
      models <- list(model[[method]])
    } else if (class(model) == "RFfit") {
      models <- list(PrepareModel2(model[method], ...))
    } else {
      models <- list(model)
    }
  }

  
  for (i in 1:length(models)) {

    if (class(model) == "RFfit" || class(model) == "RF_fit") {
      fit <- models[[i]]
      if (details) fitted <- list(fit)
      refit <- FALSE
    } else if (!refit) {
      fit <- RFfit(model=model, x=x, y=y, z=z, T=T, grid=grid, data=data,
                   lower=lower, upper=upper, 
                   methods=methods, sub.methods=sub.methods,
                   optim.control=optim.control, users.guess = users.guess,
                     distances = distances, dim=dim,
                   transform = transform, spConform=FALSE)
      if (details) fitted <- list(fit)
    }
    
    if (class(model) == "RFfit" && missing(data)) Z <- model@Z
    else Z <- StandardizeData(x=x, y=y, z=z, T=T, grid=grid, data=data,
                              distances=distances, dim=dim, RFopt=RFopt)
 
    if (length(Z$data) > 1)
      stop("cross-valdation currently only works for single sets")
    dta <- Z$data[[1]]
    

    predicted <- var <- error <- std.error <-
      matrix(NA, ncol=ncol(dta), nrow=nrow(dta)) ## just the size
    fitted <- list()
    len <- Z$dimdata[1, 1]
    
    Coords <- Z$coord[[1]]
    if (Coords$grid) Coords <- ExpandGrid(Coords)
     for (nr in 1:len) {
      if (length(base::dim(dta)) == 2)
        Daten <- dta[-nr,   , drop=FALSE] ## nicht data -- inferiert!!
      else # dim == 3 assumed
        Daten <- dta[-nr, ,  , drop=FALSE]
      coords <- DeleteCoord(Coords, nr)
      if (printlevel >= PL_IMPORTANT && pch!="") cat(pch)
      if (refit) {
        if (Z$dist.given) {
          stop("not programmed yet. Please contact author") # to do
        } else {
          fit <- RFfit(model=model,
                       x=coords, grid=grid, data=Daten,
                       lower=lower, upper=upper, 
                       methods=methods, sub.methods=sub.methods,
                       optim.control=optim.control, users.guessP = users.guess,
                       dim=dim,
                       transform = transform, spConform=FALSE)
          if (details) fitted[[nr]] <- fit
        }
      }
      
      interpol <- RFinterpolate(fit, x = coords,
                                grid=FALSE,
                                given = coords,
                                data =  Daten,
                                spConform = FALSE,
                                return_variance=TRUE, pch="")


      var[nr, ] <- as.vector(interpol$var)
      predicted[nr, ] <- as.vector(interpol$estim)
    }
    
    if (printlevel <= PL_IMPORTANT && pch!="") cat("\n")

    error <- dta - predicted
    std.error <- error / sqrt(var)
    
    res[[i]] <- list(data=Z$data, predicted=predicted, krige.var=var,
                error=error, std.error=std.error,
                p = 2 * pnorm(-abs(std.error)))
    if (details) res[[i]]$fitted <- fitted
    class(res[[i]]) <- "RFcrossvalidate"
  }
  

  if (length(res) == 1) res <- res[[1]]
  else {
     names(res) <- c(as.character(result$model1.name),
                     as.character(result$model2.name)[1])
     class(res) <- "RFcrossvalidate"
   }

  if (RFopt$general$returncall)
    attr(res, "call") <- as.character(deparse(match.call())) 
  attr(res, "coord_system") <- c(orig=RFopt$coords$coord_system,
                                 model=RFopt$coords$new_coord_system)
  res
}