Spark.ML分类模型之决策树（数据集为KDD99）

不念不忘少年蓝@ 2022-05-15 15:17 473阅读 0赞

# 环境：IDEA+SBT打包，上传至Spark集群运行 #

不知如何打包运行的，参照这篇博客[博客地址][Link 1]

首先，在项目的build.sbt中添加关于spark.mllib的依赖包。注：需要引用的包见下面完整代码。

"org.apache.spark" % "spark-mllib_2.11" % "2.3.2" % "provided"

然后，大致流程：由于spark.ml是基于DataFrame数据格式的框架，不同于spark.mllib框架中的基于RDD数据格式的。在载入训练数据和测试数据时，可以使用sc.textFile()方法，再由RDD转为DataFrame。也可以直接上传为DataFrame格式，具体做法是先创建一个SparkSession对象，再调用read()方法。

val spark = SparkSession.builder().appName("Kdd99").config("example", "some-value").getOrCreate()
    val data = spark.read.csv("/user/Tian/data/kddcup.data")

DataFrame数据格式需要操作表头，即列的名称，由于载入的数据中没有表头，故需要加上表头。方法如下

val df = data.toDF("duration", "protocol_type", "service", "flag", "src_bytes", "dst_bytes", "land",
        "wrong_fragment", "urgent", "hot", "num_failed_logins", "logged_in", "num_compromised", "root_shell",
        "su_attempted", "num_root", "num_file_creations", "num_shells", "num_access_files", "num_outbound_cmds",
        "is_host_login", "is_guest_login", "count", "srv_count", "serror_rate", "srv_serror_rate", "rerror_rate",
        "srv_rerror_rate", "same_srv_rate", "diff_srv_rate", "srv_diff_host_rate", "dst_host_count",
        "dst_host_srv_count", "dst_host_same_srv_rate", "dst_host_diff_srv_rate", "dst_host_same_src_port_rate",
        "dst_host_srv_diff_host_rate", "dst_host_serror_rate", "dst_host_srv_serror_rate", "dst_host_rerror_rate",
        "dst_host_srv_rerror_rate", "label")

上述代码使用.toDF()方法，传入参数是42列的列名。

由于ML框架下的各种机器学习的算法，传入的数据格式为DataFrame格式，只有两列，一列是features，内容为特征向量；一列是label，内容为标签向量。现在开始把KDD99数据集通过一系列转换变为这种各种。

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MzEzNTg0Ng_size_16_color_FFFFFF_t_70][]

观察数据集格式

0,tcp,http,SF,181,5450,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,8,8,0.00,0.00,0.00,0.00,1.00,0.00,0.00,9,9,1.00,0.00,0.11,0.00,0.00,0.00,0.00,0.00,normal.

可以发现，前41列为特征，最后一列是标签。并且第2、3、4、42列为类别型，不是数值型。由于使用的是决策树算法，故需要把特征中的类别特征转换为数值型，可以用操作DataFrame方法中的StringIndexer()方法。使用示例见[官网][Link 2]

转换完之后，数据集中原来的类别型特征的列（即多出的列）可用DataFrame里的drop()方法删除，传入参数为列名。

之后再把前41列合并为一列，取名为features，合并后的每一行自动成为一个稀疏向量。因合并后的列每一行是一个稀疏向量，故需要待合并的列为double类型，合并前转为double类型，代码如下：

val cols = df_final.columns.map(f => col(f).cast(DoubleType))

接下来就可以合并了，合并列的方法为：

val assembler = new VectorAssembler().setInputCols(Array("duration", "src_bytes", "dst_bytes", "land",
        "wrong_fragment", "urgent", "hot", "num_failed_logins", "logged_in", "num_compromised", "root_shell",
        "su_attempted", "num_root", "num_file_creations", "num_shells", "num_access_files", "num_outbound_cmds",
        "is_host_login", "is_guest_login", "count", "srv_count", "serror_rate", "srv_serror_rate", "rerror_rate",
        "srv_rerror_rate", "same_srv_rate", "diff_srv_rate", "srv_diff_host_rate", "dst_host_count",
        "dst_host_srv_count", "dst_host_same_srv_rate", "dst_host_diff_srv_rate", "dst_host_same_src_port_rate",
        "dst_host_srv_diff_host_rate", "dst_host_serror_rate", "dst_host_srv_serror_rate", "dst_host_rerror_rate",
        "dst_host_srv_rerror_rate", "protocol_typeIndex", "serviceIndex", "flagIndex")).setOutputCol("features")
    
    var data_fea: DataFrame = assembler.transform(df_final.select(cols: _*))

上述代码.setInputCols()传入的参数是一个字符串数组，即将要合并前41个特征列的列名，组成的数组，.setOutputCol()传入的参数是合并成的新列的列名。

合并完成后，需要把原来41列删除掉，由于删除的列比较多，写一个for循环吧，代码如下：

val colNames = data_fea.columns
      var dataset = data_fea.drop(colNames(0))
    
      for (colId <- 0 to 40) {
        dataset = dataset.drop(colNames(colId))
      }

现在数据格式已经修改完毕，需要把数据随机分成测试集和训练集。代码如下：

val Array(trainingData, testData) = dataset.randomSplit(Array(0.7, 0.3))

接下来的步骤就是固定的流水线步骤了。建立特征索引、创建决策树模型、配置流水线、配置网格参数、实例化交叉验证、通过交叉验证模型，获取最优参数集，并测试模型。最后预测正确率。

完成代码如下：

package ml
    
    import org.apache.spark.ml.classification.{DecisionTreeClassificationModel, DecisionTreeClassifier}
    import org.apache.spark.ml.feature.{StringIndexer, VectorAssembler}
    import org.apache.spark.sql.{DataFrame, SparkSession}
    import org.apache.spark.{SparkConf, SparkContext}
    import org.apache.spark.sql.types._
    import org.apache.spark.sql.functions._
    import org.apache.spark.ml.{Pipeline, PipelineModel}
    import org.apache.spark.ml.evaluation.BinaryClassificationEvaluator
    import org.apache.spark.ml.feature.VectorIndexer
    import org.apache.spark.ml.tuning.{CrossValidator, ParamGridBuilder}
    
    object Kdd99 extends App {
    
      val conf = new SparkConf().setAppName("DecisionTree").setMaster("local[16]")
      val sc = new SparkContext(conf)
      sc.setLogLevel("ERROR")
    
      val spark = SparkSession.builder().appName("Kdd99").config("example", "some-value").getOrCreate()
    
    
      //读取数据并给数据添加表头
      val data = spark.read.csv("/user/Tian/data/kddcup.data")
    
      val df = data.toDF("duration", "protocol_type", "service", "flag", "src_bytes", "dst_bytes", "land",
        "wrong_fragment", "urgent", "hot", "num_failed_logins", "logged_in", "num_compromised", "root_shell",
        "su_attempted", "num_root", "num_file_creations", "num_shells", "num_access_files", "num_outbound_cmds",
        "is_host_login", "is_guest_login", "count", "srv_count", "serror_rate", "srv_serror_rate", "rerror_rate",
        "srv_rerror_rate", "same_srv_rate", "diff_srv_rate", "srv_diff_host_rate", "dst_host_count",
        "dst_host_srv_count", "dst_host_same_srv_rate", "dst_host_diff_srv_rate", "dst_host_same_src_port_rate",
        "dst_host_srv_diff_host_rate", "dst_host_serror_rate", "dst_host_srv_serror_rate", "dst_host_rerror_rate",
        "dst_host_srv_rerror_rate", "label")
    
      //修改"protocol_type"这一列为数值型
      val indexer_2 = new StringIndexer().setInputCol("protocol_type").setOutputCol("protocol_typeIndex")
      val indexed_2 = indexer_2.fit(df).transform(df)
    
      //修改"service"这一列为数值型
      val indexer_3 = new StringIndexer().setInputCol("service").setOutputCol("serviceIndex")
      val indexed_3 = indexer_3.fit(indexed_2).transform(indexed_2)
    
      //修改"flag"这一列为数值型
      val indexer_4 = new StringIndexer().setInputCol("flag").setOutputCol("flagIndex")
      val indexed_4 = indexer_4.fit(indexed_3).transform(indexed_3)
    
      //修改"label"这一列为数值型
      val indexer_final = new StringIndexer().setInputCol("label").setOutputCol("labelIndex")
      val indexed_df = indexer_final.fit(indexed_4).transform(indexed_4)
    
      //删除原有的类别列
      val df_final = indexed_df.drop("protocol_type").drop("service")
        .drop("flag").drop("label")
    
      //合并前41列为features
      val assembler = new VectorAssembler().setInputCols(Array("duration", "src_bytes", "dst_bytes", "land",
        "wrong_fragment", "urgent", "hot", "num_failed_logins", "logged_in", "num_compromised", "root_shell",
        "su_attempted", "num_root", "num_file_creations", "num_shells", "num_access_files", "num_outbound_cmds",
        "is_host_login", "is_guest_login", "count", "srv_count", "serror_rate", "srv_serror_rate", "rerror_rate",
        "srv_rerror_rate", "same_srv_rate", "diff_srv_rate", "srv_diff_host_rate", "dst_host_count",
        "dst_host_srv_count", "dst_host_same_srv_rate", "dst_host_diff_srv_rate", "dst_host_same_src_port_rate",
        "dst_host_srv_diff_host_rate", "dst_host_serror_rate", "dst_host_srv_serror_rate", "dst_host_rerror_rate",
        "dst_host_srv_rerror_rate", "protocol_typeIndex", "serviceIndex", "flagIndex")).setOutputCol("features")
    
      //转换为double类型
      val cols = df_final.columns.map(f => col(f).cast(DoubleType))
      var data_fea: DataFrame = assembler.transform(df_final.select(cols: _*))
    
      //删除前41列，只留labelIndex和features这两列（data_fea.columns.length = 43）
      val colNames = data_fea.columns
      var dataset = data_fea.drop(colNames(0))
    
      for (colId <- 0 to 40) {
        dataset = dataset.drop(colNames(colId))
      }
    
      dataset = dataset.withColumnRenamed("labelIndex", "label")
    
      /*
      组装
       */
      //把数据随机分成训练集合测试集
      val Array(trainingData, testData) = dataset.randomSplit(Array(0.7, 0.3))
    
      //建立特征索引
      val featureIndexer = new VectorIndexer().setInputCol("features").setOutputCol("indexedFeatures").fit(dataset)
    
      //创建决策树模型
      val decisionTree = new DecisionTreeClassifier().setLabelCol("label")
        .setFeaturesCol("indexedFeatures").setImpurity("entropy").setMaxBins(100).setMaxDepth(5).setMinInfoGain(0.01)
      println("创建决策树模型...")
    
      //配置流水线
      val dtPipline = new Pipeline().setStages(Array(featureIndexer, decisionTree))
      println("配置流水线...")
    
      /*
      模型优化
       */
      //配置网格参数
      val dtParamGrid = new ParamGridBuilder()
        .addGrid(decisionTree.maxDepth, Array(3, 5, 7))
        .build()
    
      //实例化交叉验证模型
      val evaluator = new BinaryClassificationEvaluator()
      val dtCV = new CrossValidator()
        .setEstimator(dtPipline)
        .setEvaluator(evaluator)
        .setEstimatorParamMaps(dtParamGrid)
        .setNumFolds(2)
    
      //通过交叉验证模型，获取最优参数集，并测试模型
      val dtCVModel = dtCV.fit(trainingData)
      val dtPrediction = dtCVModel.transform(testData)
    
      //查看决策树匹配模型的参数
      val dtBestModel = dtCVModel.bestModel.asInstanceOf[PipelineModel]
      val dtModel = dtBestModel.stages(1).asInstanceOf[DecisionTreeClassificationModel]
      print("决策树模型深度：")
      println(dtModel.getMaxDepth)
    
      //统计预测正确率
      //t:决策树预测值的数组
      //label:测试集的标签值数组
      //count:测试集的数量
      val (t, label, count) = (dtPrediction.select("prediction").collect,
        testData.select("label").collect(),
        testData.count().toInt)
      var dt = 0
      for (i <- 0 to count - 1) {
        if (t(i) == label(i)) {
          dt += 1
        }
      }
      //打印正确率
      println("正确率：" + 1.0 * dt / count)
    
    }

最后的运行结果如下图所示：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MzEzNTg0Ng_size_16_color_FFFFFF_t_70 1][]

原创文章，转载请注明出处： [https://blog.csdn.net/weixin\_43135846/article/details/83541415][https_blog.csdn.net_weixin_43135846_article_details_83541415]

[Link 1]: https://blog.csdn.net/weixin_43135846/article/details/83182208
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MzEzNTg0Ng_size_16_color_FFFFFF_t_70]: /images/20220503/af0ce8d64ce0484c920b5f7afe70d29c.png
[Link 2]: http://spark.apache.org/docs/latest/ml-features.html#stringindexer
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MzEzNTg0Ng_size_16_color_FFFFFF_t_70 1]: /images/20220503/270f70c1e0e144c98ef05050389b903a.png
[https_blog.csdn.net_weixin_43135846_article_details_83541415]: https://blog.csdn.net/weixin_43135846/article/details/83541415