Spark机器学习系列之13： 支持向量机SVM

Posted 2020-10-23 大葱拌豆腐

Spark 优缺点分析

以下翻译自Scikit。 
The advantages of support vector machines are: 
（1）Effective in high dimensional spaces.在高维空间表现良好。 
（2）Still effective in cases where number of dimensions is greater than the number of samples.在数据维度大于样本点数时候，依然可以起作用 
（3）Uses a subset of training points in the decision function (called support vectors), so it is also memory efficient.仅仅使用训练数据的一个子集（支持向量），因此是内存友好型的算法。 
（4）Versatile: different Kernel functions can be specified for the decision function. Common kernels are provided, but it is also possible to specify custom kernels.适应广，能解决多种情况下的分类问题，这是由于它支持不同类型的核函数，甚至支持自定义的核函数。 
能灵活使用多种核函数的确使得SVM变得非常强大，而核函数相关的理论很深奥，初步的探讨可参考转载的一篇文章：http://blog.csdn.net/qq_34531825/article/details/52895621 
The disadvantages of support vector machines include: 
（1）If the number of features is much greater than the number of samples, the method is likely to give poor performances.在数据维度（特征个数）多于样本数很多的时候，通常只能训练出一个表现很差的模型。 
（2）SVMs do not directly provide probability estimates, these are calculated using an expensive five-fold cross-validation (see Scores and probabilities, below).SVM不支持直接进行概率估计，Scikit中使用很耗费资源的5折交叉检验来估计概率。

Spark Mllib

 
 

import org.apache.spark.{SparkConf,SparkContext}
import org.apache.log4j.{Level, Logger}
import org.apache.spark.mllib.classification.{SVMModel,SVMWithSGD}
import org.apache.spark.mllib.evaluation.BinaryClassificationMetrics
import org.apache.spark.mllib.util.MLUtils
import org.apache.spark.SparkConf
import org.apache.spark.SparkContext
import org.apache.spark.mllib.optimization.L1Updater
import org.apache.spark.mllib.optimization.SquaredL2Updater

object mySVM {
  def main(args:Array[String]){
    //屏蔽日志
    Logger.getLogger("org.apache.spark").setLevel(Level.ERROR)
    Logger.getLogger("org.eclipse.jetty.server").setLevel(Level.OFF) 
    val conf=new SparkConf().setMaster("local").setAppName("My App")
    val sc=new SparkContext(conf)

    // Load training data in LIBSVM format.
    val data = MLUtils.loadLibSVMFile(sc, "/data/mllib/sample_libsvm_data.txt")
    //println(data.collect()(0))//检查数据   

    // Split data into training (60%) and test (40%).
    val splits = data.randomSplit(Array(0.6, 0.4), seed = 11L)
    val training = splits(0).cache()
    val test = splits(1)

    // Run training algorithm to build the model   
    /*
     * stepSize: 迭代步长，默认为1.0
     * numIterations: 迭代次数，默认为100
     * regParam: 正则化参数，默认值为0.0
     * miniBatchFraction: 每次迭代参与计算的样本比例，默认为1.0
     * gradient：HingeGradient ()，梯度下降；
     * updater：SquaredL2Updater ()，正则化，L2范数；
     * optimizer：GradientDescent (gradient, updater)，梯度下降最优化计算。
     */
    val svmAlg=new SVMWithSGD()
    svmAlg.optimizer
      .setNumIterations(100)
      .setRegParam(0.1)//正则化参数
      .setUpdater(new L1Updater)
    val modelL1=svmAlg.run(training)


    // Clear the default threshold.
    modelL1.clearThreshold()

   // Compute raw scores on the test set.
    val scoreAndLabels = test.map { point =>
     val score = modelL1.predict(point.features)     
     (score, point.label)//return score and label
    }   

   // Get evaluation metrics.
   val metrics = new BinaryClassificationMetrics(scoreAndLabels)
   val auROC = metrics.areaUnderROC() 
   println("Area under ROC = " + auROC)  

  }   

}

参考文献 
（1） 支持向量机通俗导论（理解SVM的三层境界） http://blog.csdn.net/v_july_v/article/details/7624837 
（2）Spark MLlib SVM算法（源代码分析）http://www.itnose.net/detail/6267193.html 
（3）Spark官网与Scikit官网 
（4）LIBSVM: A  Library是 for Support Vector Machines , Chih-Chung Chang and Chih-Jen Lin，Department of Computer Science National Taiwan University, Taipei, Taiwan http://www.csie.ntu.edu.tw/~cjlin/papers/libsvm.pdf 
（5）A Tutorial on Support Vector Regression Alex J. Smola† and Bernhard Scholkopf 
（6）如何解决机器学习中数据不平衡问题 http://www.zhaokv.com/2016/01/learning-from-imbalanced-data.html 
（7）不平衡数据下的机器学习方法简介 http://www.jianshu.com/p/3e8b9f2764c8 
（8）核函数 http://blog.csdn.net/xiaojiegege123456/article/details/7728198，xiaojiegege123456 CSDN 对核函数有一个比较容易的理解的介绍 
（9）http://www.cnblogs.com/zxb1990/p/5098345.html SVM学习笔记（二）—-手写数字识别 贴近实际应用一些 
（10）http://www.cnblogs.com/jerrylead/archive/2011/03/18/1988406.html 
（11）RBF 参数 http://scikit-learn.org/stable/auto_examples/svm/plot_rbf_parameters.html