taitanic004.r

yamamoto — Feb 10, 2014, 1:40 PM

# Titanic

rm(list=ls(all=TRUE))

# load file
    filepath <- "C:/R/data/train_mv.csv"
    X0 <- read.csv(filepath)
    attach(X0)

# data
    Sex <- as.numeric(Sex)
    X1 <- cbind(Sex,Age,Pclass,SibSp,Parch,Fare)
  Y1 <- X0$Survived
    X1 <- data.frame(X1)
    X1_ALL <- data.frame(X1,Y1)

# SVM   
    library(kernlab)
  # gaussian kernel
  x0 <- ksvm(as.factor(Y1)~Sex+Age+Pclass+SibSp+Parch+Fare,data=X1_ALL, type ="C-svc", kernel="rbfdot", cross=10, prob.model=TRUE)

Using automatic sigma estimation (sigest) for RBF or laplace kernel 

    # parameter, sigma
  sigma0 <- as.numeric(x0@kernelf@kpar)

# Cross-validation, ROC
  library(Epi)

Attaching package: 'Epi'

 以下のオブジェクトはマスクされています (from 'package:base') : 

     merge.data.frame 

  cross <- 10 # 10-fold CV
  pp <- NaN;
  for(i in 1:cross){
    x <- ksvm(as.factor(Y1)~Sex+Age+Pclass+SibSp+Parch+Fare,data=X1_ALL[-seq(i,nrow(X1_ALL),cross),], type ="C-svc", kernel="rbfdot", cross=10, prob.model=TRUE, kpar=list(sigma=sigma0), C=x0@param)
    pp[seq(i,nrow(X1_ALL),cross)] <- predict(x,X1_ALL[seq(i,nrow(X1_ALL),cross),], type="probabilities")[,2]
  }
  ROC(test=pp,stat=Y1)

# clear
    detach(X0)
  rm(Sex)

# test data
    filepath_test <- "C:/R/data/test_mv.csv"
    X1_test <- read.csv(filepath_test)
    attach(X1_test)

    # data
    Sex <- as.numeric(Sex)
    X_test <- cbind(Sex,Age,Pclass,SibSp,Parch,Fare)
    X2_ALL <- data.frame(X_test)

    # prediction
    qqq1 <- predict(x0,X2_ALL,type="response")