DMUBA TP2 AA - Random Forest & J48

j48_random_forest.R

setwd("D:/Datamining")

# para eliminar error de java
if (Sys.getenv("JAVA_HOME")!="")
  Sys.setenv(JAVA_HOME="")

# paquetes requeridos
if(!require (RWeka)){
  install.packages('RWeka')  
  require (RWeka)
}

if(!require (xlsx)){
  install.packages('xlsx')  
  require (xlsx)
}

if(!require (FSelector)){
  install.packages('FSelector')  
  require (FSelector)
}



if(!require (randomForest)){
 install.packages('randomForest')
 library(randomForest)
}
library(caret)

loadFile()


informe.J48 = as.data.frame(matrix(ncol = 5))
colnames(informe.J48) = c("confianza", "nodos", "hojas", "aciertos train", "aciertos test")

# saco el 80 % de ids de la muestra
ratio<- 0.7
# hago un muestreo sobre muestra del ratio eslecificado por default 80
trainId <- sample(1:nrow(cargaAux),size=(nrow(cargaAux)*ratio))
cargaAux.train <- na.omit(cargaAux[trainId,-1])
cargaAux.test<- na.omit(cargaAux[-trainId,-1])

confianza <- 0
while (confianza <= 0.5) {
  confianza <- confianza + 0.025
  print (c(  format(Sys.time(), "%d-%m-%Y_%H-%M-%OS") , "J48", confianza ))
  #
  # Algoritmo J48
  #
  currentTree.train <- J48(Clase ~ ., data = cargaAux.train,control = Weka_control(C = confianza) )
  predictWeka <- evaluate_Weka_classifier(currentTree.train,newdata = cargaAux.test)
  
  # confianza, nodos , cantidad hojas
  nodos <- currentTree.train$classifier$measureTreeSize()
  hojas <- currentTree.train$classifier$measureNumLeaves()
  aciertosTrain <- summary(currentTree.train)$details["pctCorrect"]
  aciertosTest <- predictWeka$details["pctCorrect"]
  
  informe.J48 = rbind(informe.J48, c(confianza, nodos, hojas, aciertosTrain, aciertosTest))
    
}

set.seed(415)
arboles<-1000
informe.RF = as.data.frame(matrix(ncol = 2))
colnames(informe.RF) = c("Arboles", "Acierto")

while (arboles <= 3000) {  
  
print (c(  format(Sys.time(), "%d-%m-%Y_%H-%M-%OS") , "Ramdom Forest", arboles ))
#
# Algoritmo Ramdom Forest
#  
fit <- randomForest(as.factor(Clase) ~ lugar+ tipoprop + piso + sup_tot_m2 + sup_cub_m2+
                    cant_amb,
                    data=cargaAux.train, 
                    importance=TRUE, 
                    ntree=arboles)

arboles <- arboles + 250
Prediction <- predict(fit, cargaAux.test)
result <- confusionMatrix(cargaAux.test$Clase,Prediction)
precision <- as.data.frame(result$overall)[1,]

informe.RF = rbind(informe.RF, arboles, precision)
}

importance(fit)
varImpPlot(fit)
write.csv(informe.RF, file = "firstforest.csv", row.names = FALSE)

loadFile<- function(){
  # read dataset crea por default un data frame
  carga<-read.xlsx("tp2-work.xlsx",1)

  cargaAux= carga[,c(2,5,6,9,10,11,12,15)]
  cargaAux$Clase= as.factor(cargaAux$Clase)
 
  cargaAux
}