关于SparkMllib特征工程的案例详解(自己看的)

1. 读取SparkSQL的数据进行统计实战

• 1-读取单个列的数据

import org.apache.spark.mllib.linalg.{
     Vector, Vectors}
import org.apache.spark.mllib.stat.{
     MultivariateStatisticalSummary, Statistics}
import org.apache.spark.rdd.RDD
import org.apache.spark.{
     SparkConf, SparkContext}
import org.apache.spark.sql.SparkSession
/**
 * @author liu a fu
 * @date 2021/2/1 0001
 * @version 1.0
 * @DESC 基于SparkMllib的RDD的结构完成统计---因为在2.2.0中dataframe还没有实现基本数据统计
 *       1-准备环境
 *       2-读取数据
 *       3-转化为Vector
 *       4-数据统计操作
 *       5-展示
 */
object _01SpetalLengthStaticesDemo {
     
  def main(args: Array[String]): Unit = {
     
    //    * 1-准备环境
    val conf: SparkConf = new SparkConf().setAppName("IrisSparkCoreLoader").setMaster("local[*]")
    val spark: SparkSession = SparkSession.builder().config(conf).getOrCreate()
    val sc: SparkContext = spark.sparkContext
    sc.setLogLevel("WARN")

    //    * 2-读取数据
    val datapath="C:\\software\\studysoft\\BigdataCode\\Spark_Code\\Spark_Mllib\\data\\Iris\\length.csv"
    //    * 3-转化为Vector   密集向量
    val data: RDD[Vector] = sc.textFile(datapath).map(_.toDouble).map(x=>Vectors.dense(x))

    //    * 4-数据统计操作  统计特征Statistics.colStats  详见SparkMllib的统计特征实践一文
    val stats: MultivariateStatisticalSummary = Statistics.colStats(data)

    //    * 5-展示
    println("states nonzeros:",stats.numNonzeros)
    println("states min:",stats.min)
    println("states max:",stats.max)
    println("states mean:",stats.mean)
    println("states varience:",stats.variance)
  }
}

• 2-读取全部数据指定列的特征

import org.apache.spark.mllib.linalg.{
     Vector, Vectors}
import org.apache.spark.mllib.stat.{
     MultivariateStatisticalSummary, Statistics}
import org.apache.spark.rdd.RDD
import org.apache.spark.sql.SparkSession

/**
 * @author liu a fu
 * @date 2021/2/1 0001
 * @version 1.0
 * @DESC
 */
object _02irisDataStaticesDemo {
     
  def main(args: Array[String]): Unit = {
     
    //1-准备环境
    val spark: SparkSession = SparkSession
      .builder()
      .appName(this.getClass.getSimpleName.stripSuffix("$"))
      .master("local[*]")
      .getOrCreate()
    spark.sparkContext.setLogLevel("WARN")

    //2-读取数据 转化为Vectors
    val path = "C:\\software\\studysoft\\BigdataCode\\Spark_Code\\Spark_Mllib\\data\\Iris\\iris.data"
    val data: RDD[Vector] = spark.sparkContext.textFile(path).map(x=>(x.split(",")(0)))
      .map(_.toDouble)
      .map(x=>Vectors.dense(x))

    //3-数据统计操作
    /**
     * Statistics:  API for statistical functions in MLlib.
     * colStats: Computes column-wise summary statistics for the input RDD[Vector].
     */
    val stats: MultivariateStatisticalSummary = Statistics.colStats(data)
    //    * 5-展示
    println("states nonzeros:",stats.numNonzeros)
    println("states min:",stats.min)
    println("states max:",stats.max)
    println("states mean:",stats.mean)
    println("states varience:",stats.variance)

    //获取相关系数的double的列的值  第一列
    val data1: RDD[Double] =spark.sparkContext.textFile(path).map(x=>x.split(",")(0)).map(_.toDouble)
    //第二例
    val data2: RDD[Double] = spark.sparkContext.textFile(path).map(x=>x.split(",")(2)).map(_.toDouble)

    /**
     * corr: Compute the Pearson correlation for the input RDDs.
     *       Returns NaN if either vector has 0 variance.
     */
    val corr1: Double = Statistics.corr(data1,data2)
    println("data1 and data2 corr value is:",corr1) //(data1 and data2 corr value is:,0.8717541573048727)
  }
}

• 3-通过SQL的方式进行统计

import org.apache.spark.ml.feature.VectorAssembler
import org.apache.spark.ml.stat.Correlation
import org.apache.spark.sql.{
     DataFrame, SparkSession}

/**
 * @author liu a fu
 * @date 2021/2/2 0002
 * @version 1.0
 * @DESC 使用SQL的方式读取数据
 *       1-准备环境
 *       2-准备读取数据---option方法读取
 *       3-解析数据
 *       4-打印schema
 */
object _03IrisSparkSQLStaticesDemo {
     
  def main(args: Array[String]): Unit = {
     
    //1-读取数据
    val spark: SparkSession = SparkSession
      .builder()
      .appName(this.getClass.getSimpleName.stripSuffix("$"))
      .master("local[*]")
      .getOrCreate()
    spark.sparkContext.setLogLevel("WARN")

    //2-准备读取数据---option方法读取
    val path = "C:\\software\\studysoft\\BigdataCode\\Spark_Code\\Spark_Mllib\\data\\Iris\\iris.csv"
    val valueDF: DataFrame = spark.read.format("csv")   //读取CSV文件的数据
      .option("header", "true")
      .option("inferschema", true)
      .load(path)

    valueDF.printSchema()
    valueDF.show()
    /**
     * root
     * |-- sepal_length: double (nullable = true)
     * |-- sepal_width: double (nullable = true)
     * |-- petal_length: double (nullable = true)
     * |-- petal_width: double (nullable = true)
     * |-- class: string (nullable = true)
     */

    val vec: VectorAssembler = new VectorAssembler()
      .setInputCols(Array("sepal_length", "sepal_width", "petal_length", "petal_width"))
      .setOutputCol("feaures")
    val vecResult: DataFrame = vec.transform(valueDF)

    //Compute the Pearson correlation matrix for the input Dataset of Vectors.
    val corr: DataFrame = Correlation.corr(vecResult, "feaures", "pearson")
    println("corr matrix is:")
    corr.show(false)

  }
}

2.特征工程实践

• 1-对数据有敏感性(搞大数据要对数据有想法)
• 2-特征工程分类
  • 特征抽取
  • 特征选择
  • 特征转换-----重要
  • 特征降维
    • 高纬度降低到低纬度，低纬度的物理意义不明确

特征工程案例:
Iris部分数据集展示:

/**
  * DESC: 使用SQL方式读取数据
  * Complete data processing and modeling process steps:
  * 1-准备环境
  * 2-准备读取数据---option方法读取
  * 3-解析数据
  * 4-打印schema
  */
object IrisSparkSQLFeaturesEngineer {
     
  def main(args: Array[String]): Unit = {
     
    //    * 1-准备环境
    val conf: SparkConf = new SparkConf().setAppName("IrisSparkCoreLoader").setMaster("local[*]")
    val spark: SparkSession = SparkSession.builder().config(conf).getOrCreate()
    val sc: SparkContext = spark.sparkContext
    sc.setLogLevel("WARN")
    import spark.implicits._
      
    //    * 2-准备读取数据---option方法读取
    val datapath = "C:\\software\\studysoft\\BigdataCode\\Spark_Code\\Spark_Mllib\\data\\Iris\\iris.csv"
    val data: DataFrame = spark.read.format("csv").option("header", "true").option("inferschema", true).load(datapath)
    //    * 3-解析数据
    data.printSchema()
    data.show(false)
    //    * 4-打印schema
    //    root
    //    |-- sepal_length: double (nullable = true)
    //    |-- sepal_width: double (nullable = true)
    //    |-- petal_length: double (nullable = true)
    //    |-- petal_width: double (nullable = true)
    //    |-- class: string (nullable = true)
      
    //1-首先将数据的标签列进行labelencoder的编码的操作0-1-2
    val strIndex: StringIndexer = new StringIndexer().setInputCol("class").setOutputCol("labelclass")
    val strModel: StringIndexerModel = strIndex.fit(data)
    val strResult: DataFrame = strModel.transform(data)
    strResult.show(false)
      
    //2-可以将4个特征列转化为3个特征列
    //2-1特征选择------df.secelt------ChiSquareSeletor
    data.select("sepal_length").show(false)
    data.select($"sepal_length").show(false)
    data.select(col("sepal_length"))
    data.select($"sepal_length",col("sepal_width")).show(false)
    
    val vec: VectorAssembler = new VectorAssembler()
      .setInputCols(Array("sepal_length","sepal_width","petal_length","petal_width"))
      .setOutputCol("features")
    val vecResult: DataFrame = vec.transform(data)
      
    
    //卡方验证选特征
    val chi: ChiSqSelector = new ChiSqSelector().setFeaturesCol("features").setLabelCol("class").setNumTopFeatures(3)
    val chiModel: ChiSqSelectorModel = chi.fit(vecResult)
    val chiResult: DataFrame = chiModel.transform(vecResult)
    chiResult.show(false)
      
    //2-2特征降维------PCA  setK(2) 降维到2维度
    println("pca transfomation:")
    val pca: PCA = new PCA().setInputCol("features").setOutputCol("pca_features").setK(2)
    val pcaModel: PCAModel = pca.fit(vecResult)
    pcaModel.transform(vecResult).show(false)
  }
}