Spark學習筆記——泰坦尼克生還預測

package kaggle

import org.apache.spark.SparkContext
import org.apache.spark.SparkConf
import org.apache.spark.sql.{SQLContext, SparkSession}
import org.apache.spark.mllib.regression.LabeledPoint
import org.apache.spark.mllib.classification.{LogisticRegressionWithLBFGS, LogisticRegressionWithSGD, NaiveBayes, SVMWithSGD}
import org.apache.log4j.{Level, Logger}
import org.apache.spark.mllib.linalg.Vectors
import org.apache.spark.mllib.stat.Statistics


/**
  * Created by mi on 17-5-23.
  */


object Titanic {


  def main(args: Array[String]) {

    //    val sparkSession = SparkSession.builder.
    //      master("local")
    //      .appName("spark session example")
    //      .getOrCreate()
    //    val rawData = sparkSession.read.csv("/home/mi/下載/kaggle/Titanic/nohead-train.csv")
    //    val d = rawData.map{p => p.asInstanceOf[person]}
    //    d.show()

    val conf = new SparkConf().setAppName("WordCount").setMaster("local")
    val sc = new SparkContext(conf)
    val sqlContext = new SQLContext(sc)

    //遮蔽日誌
    Logger.getLogger("org.apache.spark").setLevel(Level.WARN)
    Logger.getLogger("org.eclipse.jetty.server").setLevel(Level.OFF)

    // 讀取資料
    val df = sqlContext.load("com.databricks.spark.csv", Map("path" -> "/home/mi/下載/kaggle/Titanic/train.csv", "header" -> "true"))

    // 分析年齡資料
    val ageAnalysis = df.rdd.filter(d => d(5) != null).map { d =>
      val age = d(5).toString.toDouble
      Vectors.dense(age)
    }
    val ageMean = Statistics.colStats(ageAnalysis).mean(0)
    val ageMax = Statistics.colStats(ageAnalysis).max(0)
    val ageMin = Statistics.colStats(ageAnalysis).min(0)
    val ageDiff = ageMax - ageMin

    // 分析船票價格資料
    val fareAnalysis = df.rdd.filter(d => d(9) != null).map { d =>
      val fare = d(9).toString.toDouble
      Vectors.dense(fare)
    }
    val fareMean = Statistics.colStats(fareAnalysis).mean(0)
    val fareMax = Statistics.colStats(fareAnalysis).max(0)
    val fareMin = Statistics.colStats(fareAnalysis).min(0)
    val fareDiff = fareMax - fareMin


    // 資料預處理
    val trainData = df.rdd.map { d =>
      val label = d(1).toString.toInt
      val sex = d(4) match {
        case "male" => 0.0
        case "female" => 1.0
      }
      val age = d(5) match {
        case null => (ageMean - ageMin) / ageDiff
        case _ => (d(5).toString().toDouble - ageMin) / ageDiff
      }
      val fare = d(9) match {
        case null => (fareMean - fareMin) / fareDiff
        case _ => (d(9).toString().toDouble - fareMin) / fareDiff
      }

      LabeledPoint(label, Vectors.dense(sex, age, fare))
    }

    // 切分資料集和測試集
    val Array(trainingData, testData) = trainData.randomSplit(Array(0.8, 0.2))

    // 訓練資料
    val numIterations = 8
    val lrModel = new LogisticRegressionWithLBFGS().setNumClasses(2).run(trainingData)
    //    val svmModel = SVMWithSGD.train(trainingData, numIterations)

    val nbTotalCorrect = testData.map { point =>
      if (lrModel.predict(point.features) == point.label) 1 else 0
    }.sum
    val nbAccuracy = nbTotalCorrect / testData.count

    println("SVM模型正確率：" + nbAccuracy)

    // 預測
    // 讀取資料
    val testdf = sqlContext.load("com.databricks.spark.csv", Map("path" -> "/home/mi/下載/kaggle/Titanic/test.csv", "header" -> "true"))

    // 分析測試集年齡資料
    val ageTestAnalysis = testdf.rdd.filter(d => d(4) != null).map { d =>
      val age = d(4).toString.toDouble
      Vectors.dense(age)
    }
    val ageTestMean = Statistics.colStats(ageTestAnalysis).mean(0)
    val ageTestMax = Statistics.colStats(ageTestAnalysis).max(0)
    val ageTestMin = Statistics.colStats(ageTestAnalysis).min(0)
    val ageTestDiff = ageTestMax - ageTestMin

    // 分析船票價格資料
    val fareTestAnalysis = testdf.rdd.filter(d => d(8) != null).map { d =>
      val fare = d(8).toString.toDouble
      Vectors.dense(fare)
    }
    val fareTestMean = Statistics.colStats(fareTestAnalysis).mean(0)
    val fareTestMax = Statistics.colStats(fareTestAnalysis).max(0)
    val fareTestMin = Statistics.colStats(fareTestAnalysis).min(0)
    val fareTestDiff = fareTestMax - fareTestMin

    // 資料預處理
    val data = testdf.rdd.map { d =>
      val sex = d(3) match {
        case "male" => 0.0
        case "female" => 1.0
      }
      val age = d(4) match {
        case null => (ageTestMean - ageTestMin) / ageTestDiff
        case _ => (d(4).toString().toDouble - ageTestMin) / ageTestDiff
      }
      val fare = d(8) match {
        case null => (fareTestMean - fareTestMin) / fareTestDiff
        case _ => (d(8).toString().toDouble - fareTestMin) / fareTestDiff
      }

      Vectors.dense(sex, age, fare)
    }

    val predictions = lrModel.predict(data).map(p => p.toInt)
    // 儲存預測結果
    predictions.coalesce(1).saveAsTextFile("file:///home/mi/下載/kaggle/Titanic/test_predict")
  }
}