# the code to transforme to regression was transfered to spatialMatrixRegression.R

#############################################################################
# Train the models
#

train.all <- function(obj, name2save){
  train.trees(obj, name2save)
  train.svms(obj, name2save)
  train.rfs(obj, name2save)
}

train.rfs <- function(obj, name2save){
#  rf.T <- system.time(rf <- tune.model(data=obj, learner="randomForest", params=list(ntree=c(500,1000,1500))))
  rf.T <- system.time(rf <- tune.model(data=obj, learner="randomForest", params=list(ntree=500)))
  save(rf, rf.T, file=paste("data/",name2save,"_rf",".RData",sep=""))
  print(paste(paste("data/",name2save,"_rf",".RData",sep="")))
}

train.trees <- function(obj, name2save){
#  tree.T <- system.time(tree <- tune.model(data=obj, learner="rpartXse", params=list(se=c(0,.5,1,1.5))))
  tree.T <- system.time(tree <- tune.model(data=obj, learner="rpartXse", params=list(se=0)))
  save(tree, tree.T, file=paste("data/",name2save,"_tree",".RData",sep=""))
  print(paste("data/",name2save,"_tree",".RData",sep=""))
}

train.svms <- function(obj, name2save){
#  svm.T <- system.time(svm <- tune.model(data=obj, learner="svm", params=list(cost=c(1,10,100),gamma=c(.1, .5))))
  svm.T <- system.time(svm <- tune.model(data=obj, learner="svm", params=list(cost=1,gamma=.1)))
  save(svm, svm.T, file=paste("data/",name2save,"_svm",".RData",sep=""))
  print(paste("data/",name2save,"_svm",".RData",sep=""))
}

#############################################################################

# train.base.models(x0,"x0")
train.base.models <- function(obj, obj.name){
  nomes <- c(paste("m",1:3,sep=""), paste("w",1:3,sep=""))
  for(n in nomes){
    aux.t <- system.time(aux.pred <- learn.base.models(obj=obj, type=n))
    # assign time
    assign(paste(obj.name,n,sep="."), aux.pred)
    assign(paste(obj.name,n,"T",sep="."), aux.t)
  }
  save(list=c(paste(obj.name, nomes,sep="."),paste(obj.name,nomes,"T",sep=".")),file=paste("data/",obj.name,"_base.RData",sep=""))
}


#############################################################################
# krige

learn.autokrige.saving <- function(obj, nome, nmax){
  model <- list()
  
  try(kc.T <- system.time(model[["autoKrige"]] <- learn.autokrige(obj, nmax)))
  try(save(model, kc.T, file=paste("data/",nome,"_autoKrige.RData",sep="")))
  try(rm(model, kc.T))
}

##########################################################################

getRealValues <- function(model){
  real <- list()
  for(i in 1:length(model)){
    variations <- list()
    # for 1 ... 30 data variation
    for(j in 1:length(model[[i]]$pred)){
        variations[[length(variations)+1]] <- model[[i]]$real[[j]]
    }
    real[[length(real)+1]] <- variations
  }
  real
}

######################################################################################
# Error


error <- function(preds, type="MAE"){
  er <- NULL
  for(i in 1:length(preds)){
    error <- NULL
    for(j in 1:length(preds[[i]]$pred)){
      error <- c(error, mae(preds[[i]]$pred[[j]], preds[[i]]$real[[j]]))
    }
    er <- c(er, mean(error))
  }
  er    
}

###########################################################################################
# plot the predicted picture

saveNAJpg <- function(obj, file.name="nina", gap.id=1){
  for(i in 1:length(obj$data)){
    mat <- num2mat(obj$data[[i]]$gaps[[gap.id]], obj$num.cols)
    mat[which(is.na(mat))] <- 255
    img.mat <- imagedata(mat)
    writeJpeg(paste("graphics/",file.name,"_",i,"_",gap.id,".jpg",sep=""), img.mat)
    print(paste("graphics/",file.name,"_",i,"_",gap.id,".jpg saved",sep=""))
  }
}

saveModelJpg <- function(obj, model, model.name, file.name, model.tuning.id=1, gap.id=1, type="model"){
  for(i in 1:length(obj$data)){
    gaps <- obj$data[[i]]$gaps[[gap.id]]
    ind.na <- which(is.na(gaps))
    if(type=="base"){
      pred <- model[[i]]$pred[[gap.id]]
    } else if(type=="model"){
      pred <- model[[model.tuning.id]][[i]]$pred[[gap.id]]
    }

    gaps[ind.na] <- pred
    mat <- num2mat(gaps, obj$num.cols)
    img.mat <- imagedata(mat)
    writeJpeg(paste("graphics/",file.name,".jpg",sep=""), img.mat)
  }
}

###########################################################################################
# Mean Absolute Error
mae <- function(pred, real){
  if(length(pred)!=length(real))
    stop("must have the same length")
  mean(abs(pred-real))
}