require(uba)

# ============================================================
# EVALUTION STATISTICS
# This function will be used by the different learning routines
# to get the stats we are estimating by CV
# ============================================================
eval.stats <- function(form,train,test,preds) {
  trues <- resp(form,test)
  trainY <- resp(form,train)

  phiF.args <- phi.control(trainY, method="extremes")
  lossF.args <- loss.control(trainY)

  prec <- util(preds,trues,phiF.args,lossF.args,util.control(umetric="P"))
  rec  <- util(preds,trues,phiF.args,lossF.args,util.control(umetric="R"))
  F05  <- util(preds,trues,phiF.args,lossF.args,util.control(umetric="Fm",beta=0.5))
  F1   <- util(preds,trues,phiF.args,lossF.args,util.control(umetric="Fm",beta=1))
  F2   <- util(preds,trues,phiF.args,lossF.args,util.control(umetric="Fm",beta=2))

  c(mad=mean(abs(trues-preds)),
    mse=mean((trues-preds)^2),
    prec=prec,rec=rec,F05=F05,F1=F1,F2=F2
    )
}



# ============================================================
# Obtraining a model using the standard data set,
# i.e. all given data
# ============================================================
allData <- function(form,train,test,learner,...) {
  preds <- do.call(paste('cv',learner,sep='.'),
                   list(form,train,test,...)
                   )
  eval.stats(form,train,test,preds)
}


# ============================================================
# Obtraining a model using  random undersampling
# i.e. all given data
# ============================================================
underSampl <- function(form,train,test,learner,thr.rel=0.5,perc.under=200,...) {
  newtrain <- random.underSample(form,train,thr.rel,perc.under)
  preds <- do.call(paste('cv',learner,sep='.'),
                   list(form,newtrain,test,...)
                   )
  eval.stats(form,train,test,preds)
}


# ============================================================
# Obtraining a model using smote
# i.e. all given data
# ============================================================
smote <- function(form,train,test,learner,thr.rel,perc.over,k,perc.under,...) {
  newtrain <- SMOTE.R(form,train,thr.rel,perc.over,k,perc.under)
  preds <- do.call(paste('cv',learner,sep='.'),
                   list(form,newtrain,test,...)
                   )
  eval.stats(form,train,test,preds)
}


# ==============================
# the learn/test functions for the different systems 
cv.svm <- function(form,train,test,...) {
  m <- svm(form,train,...)
  predict(m,test)
}
cv.rf <- function(form,train,test,...) {
  m <- randomForest(form,train,...)
  predict(m,test)
}
cv.earth <- function(form,train,test,...) {
  m <- earth(form,train,...)
  predict(m,test)
}