8.4 分类与回归
一、逻辑斯蒂回归分类器
逻辑斯蒂回归(logistic regression)是统计学习中的经典分类方法,属于对数线性模型。logistic回归的因变量可以是二分类的,也可以是多分类的。
任务描述:以iris数据集(iris)为例进行分析(iris下载地址:http://dblab.xmu.edu.cn/blog/wp-content/uploads/2017/03/iris.txt)
iris以鸢尾花的特征作为数据来源,数据集包含150个数据集,分为3类,每类50个数据,每个数据包含4个属性,是在数据挖掘、数据分类中非常常用的测试集、训练集。为了便于理解,这里主要用后两个属性(花瓣的长度和宽度)来进行分类。
1.用二项逻辑斯蒂回归来解决二分类问题
首先我们先取其中的后两类数据,用二项逻辑斯蒂回归进行二分类分析
(1)导入需要的包
import org.apache.spark.sql.Row import org.apache.spark.sql.SparkSession import org.apache.spark.ml.linalg.{Vector,Vectors} import org.apache.spark.ml.evaluation.MulticlassClassificationEvaluator import org.apache.spark.ml.{Pipeline,PipelineModel} import org.apache.spark.ml.feature.{IndexToString, StringIndexer, VectorIndexer,HashingTF, Tokenizer} import org.apache.spark.ml.classification.LogisticRegression import org.apache.spark.ml.classification.LogisticRegressionModel import org.apache.spark.ml.classification.{BinaryLogisticRegressionSummary, LogisticRegression} import org.apache.spark.sql.functions;
(2)读取数据,简要分析
scala> import spark.implicits._ import spark.implicits._ scala> case class Iris(features: org.apache.spark.ml.linalg.Vector, label: String) defined class Iris scala> val data = spark.sparkContext.textFile("file:///usr/local/spark/iris.txt").map(_.split(",")).map(p => I ris(Vectors.dense(p(0).toDouble,p(1).toDouble,p(2).toDouble, p(3).toDouble), p(4 ).toString())).toDF() data: org.apache.spark.sql.DataFrame = [features: vector, label: string] scala> data.show() +-----------------+-----------+ | features| label| +-----------------+-----------+ |[5.1,3.5,1.4,0.2]|Iris-setosa| |[4.9,3.0,1.4,0.2]|Iris-setosa| |[4.7,3.2,1.3,0.2]|Iris-setosa| |[4.6,3.1,1.5,0.2]|Iris-setosa| |[5.0,3.6,1.4,0.2]|Iris-setosa| |[5.4,3.9,1.7,0.4]|Iris-setosa| |[4.6,3.4,1.4,0.3]|Iris-setosa| |[5.0,3.4,1.5,0.2]|Iris-setosa| |[4.4,2.9,1.4,0.2]|Iris-setosa| |[4.9,3.1,1.5,0.1]|Iris-setosa| |[5.4,3.7,1.5,0.2]|Iris-setosa| |[4.8,3.4,1.6,0.2]|Iris-setosa| |[4.8,3.0,1.4,0.1]|Iris-setosa| |[4.3,3.0,1.1,0.1]|Iris-setosa| |[5.8,4.0,1.2,0.2]|Iris-setosa| |[5.7,4.4,1.5,0.4]|Iris-setosa| |[5.4,3.9,1.3,0.4]|Iris-setosa| |[5.1,3.5,1.4,0.3]|Iris-setosa| |[5.7,3.8,1.7,0.3]|Iris-setosa| |[5.1,3.8,1.5,0.3]|Iris-setosa| +-----------------+-----------+ only showing top 20 rows
因为我们现在处理的是2分类问题,所以我们不需要全部的3类数据,我们要从中选出两类的数据。
首先把刚刚得到的数据注册成一个表iris,注册成这个表之后,我们就可以通过sql语句进行数据查询。
scala> data.createOrReplaceTempView("iris") scala> val df = spark.sql("select * from iris where label != 'Iris-setosa'") df: org.apache.spark.sql.DataFrame = [features: vector, label: string] scala> df.map(t => t(1)+":"+t(0)).collect().foreach(println) Iris-versicolor:[7.0,3.2,4.7,1.4] Iris-versicolor:[6.4,3.2,4.5,1.5] Iris-versicolor:[6.9,3.1,4.9,1.5] …… ……
(3)构建ML的pipeline
a.分别获取标签列和特征列,进行索引,并进行了重命名。
scala> val labelIndexer = new StringIndexer().setInputCol("label").setOutputCol("indexedLabel").fit(df) labelIndexer: org.apache.spark.ml.feature.StringIndexerModel = strIdx_e53e67411169 scala> val featureIndexer = new VectorIndexer().setInputCol("features").setOutputCol("indexedFeatures").fit(df) featureIndexer: org.apache.spark.ml.feature.VectorIndexerModel = vecIdx_53b988077b38
b.接下来,我们把数据集随机分成训练集和测试集,其中训练集占70%
scala> val Array(trainingData, testData) = df.randomSplit(Array(0.7, 0.3)) trainingData: org.apache.spark.sql.Dataset[org.apache.spark.sql.Row] = [features: vector, label: string] testData: org.apache.spark.sql.Dataset[org.apache.spark.sql.Row] = [features: vector, label: string]
c.然后,我们设置logistic的参数,这里我们统一用setter的方法来设置,也可以用ParamMap来设置(具体的可以查看spark mllib的官网)。这里我们设置了循环次数为10次,正则化项为0.3等
scala> val lr = new LogisticRegression().setLabelCol("indexedLabel").setFeaturesCol("indexedFeatures").setMaxIter(10).setRegParam(0.3).setElasticNetParam(0.8) lr: org.apache.spark.ml.classification.LogisticRegression = logreg_692899496c23
d.这里我们设置一个labelConverter,目的是把预测的类别重新转化成字符型的。
scala> val labelConverter = new IndexToString().setInputCol("prediction").setOut putCol("predictedLabel").setLabels(labelIndexer.labels) labelConverter: org.apache.spark.ml.feature.IndexToString = idxToStr_c204eafabf57
e.构建pipeline,设置stage,然后调用fit()来训练模型
scala> val lrPipeline = new Pipeline().setStages(Array(labelIndexer, featureIndexer, lr, labelConverter)) lrPipeline: org.apache.spark.ml.Pipeline = pipeline_eb1b201af1e0 scala> val lrPipelineModel = lrPipeline.fit(trainingData) lrPipelineModel: org.apache.spark.ml.PipelineModel = pipeline_eb1b201af1e0
f.pipeline本质上是一个Estimator,当pipeline调用fit()的时候就产生了一个PipelineModel,本质上是一个Transformer。然后这个PipelineModel就可以调用transform()来进行预测,生成一个新的DataFrame,即利用训练得到的模型对测试集进行验证。
scala> val lrPredictions = lrPipelineModel.transform(testData) lrPredictions: org.apache.spark.sql.DataFrame = [features: vector, label: string ... 6 more fields]
g.最后我们可以输出预测的结果,其中select选择要输出的列,collect获取所有行的数据,用foreach把每行打印出来。其中打印出来的值依次分别代表该行数据的真实分类和特征值、预测属于不同分类的概率、预测的分类
scala> lrPredictions.select("predictedLabel", "label", "features", "probability").collect().foreach { case Row(predictedLabel: String, label: String, features: Vector, prob: Vector) => println(s"($label, $features) --> prob=$prob, predicted Label=$predictedLabel")} (Iris-virginica, [4.9,2.5,4.5,1.7]) --> prob=[0.4796551461409372,0.5203448538590628], predictedLabel=Iris-virginica (Iris-versicolor, [5.1,2.5,3.0,1.1]) --> prob=[0.5892626391059901,0.41073736089401], predictedLabel=Iris-versicolor (Iris-versicolor, [5.5,2.3,4.0,1.3]) --> prob=[0.5577310241453046,0.4422689758546954], predictedLabel=Iris-versicolor
(4)模型评估
创建一个MulticlassClassificationEvaluator实例,用setter方法把预测分类的列名和真实分类的列名进行设置;然后计算预测准确率和错误率
scala> val evaluator = new MulticlassClassificationEvaluator().setLabelCol("indexedLabel").setPredictionCol("prediction") evaluator: org.apache.spark.ml.evaluation.MulticlassClassificationEvaluator = mcEval_a80353e4211d scala> val lrAccuracy = evaluator.evaluate(lrPredictions) lrAccuracy: Double = 1.0 scala> println("Test Error = " + (1.0 - lrAccuracy)) Test Error = 0.0
接下来我们可以通过model来获取我们训练得到的逻辑斯蒂模型。前面已经说过model是一个PipelineModel,因此我们可以通过调用它的stages来获取模型,具体如下:
scala> val lrModel = lrPipelineModel.stages(2).asInstanceOf[LogisticRegressionModel] lrModel: org.apache.spark.ml.classification.LogisticRegressionModel = logreg_692899496c23 scala> println("Coefficients: " + lrModel.coefficients+"Intercept: "+lrModel.intercept+"numClasses: "+lrModel.numClasses+"numFeatures: "+lrModel.numFeatures) Coefficients: [-0.0396171957643483,0.0,0.0,0.07240315639651046]Intercept: -0.23127346342015379numClasses: 2numFeatures: 4
2.用多项逻辑斯蒂回归来解决二分类问题
3.用多项逻辑斯蒂回归来解决多分类问题
二、决策树分类器
决策树(decision tree)是一种基本的分类与回归方法,这里主要介绍用于分类的决策树。决策树模式呈树形结构,其中每个内部节点表示一个属性上的测试,每个分支代表一个测试输出,每个叶节点代表一种类别。学习时利用训练数据,根据损失函数最小化的原则建立决策树模型;预测时,对新的数据,利用决策树模型进行分类
决策树学习通常包括3个步骤:特征选择、决策树的生成和决策树的剪枝。
1.特征选择
特征选择在于选取对训练数据具有分类能力的特征,这样可以提高决策树学习的效率。通常特征选择的准则是信息增益(或信息增益比、基尼指数等),每次计算每个特征的信息增益,并比较它们的大小,选择信息增益最大(信息增益比最大、基尼指数最小)的特征
2.决策树的生成
- 从根结点开始,对结点计算所有可能的特征的信息增益,选择信息增益最大的特征作为结点的特征,由该特征的不同取值建立子结点,再对子结点递归地调用以上方法,构建决策树;直到所有特征的信息增均很小或没有特征可以选择为止,最后得到一个决策树。
- 决策树需要有停止条件来终止其生长的过程。一般来说最低的条件是:当该节点下面的所有记录都属于同一类,或者当所有的记录属性都具有相同的值时。这两种条件是停止决策树的必要条件,也是最低的条件。在实际运用中一般希望决策树提前停止生长,限定叶节点包含的最低数据量,以防止由于过度生长造成的过拟合问题。
3.决策树的剪枝
决策树生成算法递归地产生决策树,直到不能继续下去为止。这样产生的树往往对训练数据的分类很准确,但对未知的测试数据的分类却没有那么准确,即出现过拟合现象。解决这个问题的办法是考虑决策树的复杂度,对已生成的决策树进行简化,这个过程称为剪枝。
我们以iris数据集(iris)为例进行分析(iris下载地址:http://dblab.xmu.edu.cn/blog/wp-content/uploads/2017/03/iris.txt)iris以鸢尾花的特征作为数据来源,数据集包含150个数据集,分为3类,每类50个数据,每个数据包含4个属性,是在数据挖掘、数据分类中非常常用的测试集、训练集。
(1)导入需要的包
import org.apache.spark.sql.SparkSession import org.apache.spark.ml.linalg.{Vector,Vectors} import org.apache.spark.ml.Pipeline import org.apache.spark.ml.feature.{IndexToString, StringIndexer, VectorIndexer}
(2)读取数据,简要分析
scala> import spark.implicits._ import spark.implicits._ scala> case class Iris(features: org.apache.spark.ml.linalg.Vector, label: String) defined class Iris scala> val data = spark.sparkContext.textFile("file:///usr/local/spark/iris.txt").map(_.split(",")).map(p => Iris(Vectors.dense(p(0).toDouble,p(1).toDouble,p(2).toDouble, p(3).toDouble),p(4).toString())).toDF() data: org.apache.spark.sql.DataFrame = [features: vector, label: string] scala> data.createOrReplaceTempView("iris") scala> val df = spark.sql("select * from iris") df: org.apache.spark.sql.DataFrame = [features: vector, label: string] scala> df.map(t => t(1)+":"+t(0)).collect().foreach(println) Iris-setosa:[5.1,3.5,1.4,0.2] Iris-setosa:[4.9,3.0,1.4,0.2] Iris-setosa:[4.7,3.2,1.3,0.2] Iris-setosa:[4.6,3.1,1.5,0.2] Iris-setosa:[5.0,3.6,1.4,0.2] Iris-setosa:[5.4,3.9,1.7,0.4] Iris-setosa:[4.6,3.4,1.4,0.3] ... ...
(3)进一步处理特征和标签,以及数据分组
//分别获取标签列和特征列,进行索引,并进行了重命名。 scala> val labelIndexer = new StringIndexer().setInputCol("label").setOutputCol( "indexedLabel").fit(df) labelIndexer: org.apache.spark.ml.feature.StringIndexerModel = strIdx_107f7e530fa7 scala> val featureIndexer = new VectorIndexer().setInputCol("features").setOutpu tCol("indexedFeatures").setMaxCategories(4).fit(df) featureIndexer: org.apache.spark.ml.feature.VectorIndexerModel = vecIdx_0649803dfa70 //这里我们设置一个labelConverter,目的是把预测的类别重新转化成字符型的。 scala> val labelConverter = new IndexToString().setInputCol("prediction").setOut putCol("predictedLabel").setLabels(labelIndexer.labels) labelConverter: org.apache.spark.ml.feature.IndexToString = idxToStr_046182b2e571 //接下来,我们把数据集随机分成训练集和测试集,其中训练集占70%。 scala> val Array(trainingData, testData) = data.randomSplit(Array(0.7, 0.3)) trainingData: org.apache.spark.sql.Dataset[org.apache.spark.sql.Row] = [features: vector, label: string] testData: org.apache.spark.sql.Dataset[org.apache.spark.sql.Row] = [features: vector, label: string]
(4)构建决策树分类模型
//导入所需要的包 import org.apache.spark.ml.classification.DecisionTreeClassificationModel import org.apache.spark.ml.classification.DecisionTreeClassifier import org.apache.spark.ml.evaluation.MulticlassClassificationEvaluator //训练决策树模型,这里我们可以通过setter的方法来设置决策树的参数,也可以用ParamMap来设置(具体的可以查看spark mllib的官网)。具体的可以设置的参数可以通过explainParams()来获取。 scala> val dtClassifier = new DecisionTreeClassifier().setLabelCol("indexedLabel ").setFeaturesCol("indexedFeatures") dtClassifier: org.apache.spark.ml.classification.DecisionTreeClassifier = dtc_029ea28aceb1 //在pipeline中进行设置 scala> val pipelinedClassifier = new Pipeline().setStages(Array(labelIndexer, featureIndexer, dtClassifier, labelConverter)) pipelinedClassifier: org.apache.spark.ml.Pipeline = pipeline_a254dfd6dfb9 //训练决策树模型 scala> val modelClassifier = pipelinedClassifier.fit(trainingData) modelClassifier: org.apache.spark.ml.PipelineModel = pipeline_a254dfd6dfb9 //进行预测 scala> val predictionsClassifier = modelClassifier.transform(testData) predictionsClassifier: org.apache.spark.sql.DataFrame = [features: vector, label: string ... 6 more fields] //查看部分预测的结果 scala> predictionsClassifier.select("predictedLabel", "label", "features").show(20) +---------------+---------------+-----------------+ | predictedLabel| label| features| +---------------+---------------+-----------------+ | Iris-setosa| Iris-setosa|[4.4,2.9,1.4,0.2]| | Iris-setosa| Iris-setosa|[4.6,3.6,1.0,0.2]| | Iris-virginica|Iris-versicolor|[4.9,2.4,3.3,1.0]| | Iris-setosa| Iris-setosa|[4.9,3.1,1.5,0.1]| | Iris-setosa| Iris-setosa|[4.9,3.1,1.5,0.1]|
(5)评估决策树分类模型
scala> val evaluatorClassifier = new MulticlassClassificationEvaluator().s etLabelCol("indexedLabel").setPredictionCol("prediction").setMetricName("accuracy") evaluatorClassifier: org.apache.spark.ml.evaluation.MulticlassClassificationEvaluator = mcEval_4abc19f3a54d scala> val accuracy = evaluatorClassifier.evaluate(predictionsClassifier) accuracy: Double = 0.8648648648648649 scala> println("Test Error = " + (1.0 - accuracy)) Test Error = 0.1351351351351351 scala> val treeModelClassifier = modelClassifier.stages(2).asInstanceOf[De cisionTreeClassificationModel] treeModelClassifier: org.apache.spark.ml.classification.DecisionTreeClassificati onModel = DecisionTreeClassificationModel (uid=dtc_029ea28aceb1) of depth 5 with 13 nodes scala> println("Learned classification tree model:\n" + treeModelClassifier.toDebugString) Learned classification tree model: DecisionTreeClassificationModel (uid=dtc_029ea28aceb1) of depth 5 with 13 nodes If (feature 2 <= 1.9) Predict: 2.0 Else (feature 2 > 1.9) If (feature 2 <= 4.7) If (feature 0 <= 4.9) Predict: 1.0 Else (feature 0 > 4.9) Predict: 0.0 Else (feature 2 > 4.7) If (feature 3 <= 1.6) If (feature 2 <= 4.8) Predict: 0.0 Else (feature 2 > 4.8) If (feature 0 <= 6.0) Predict: 0.0 Else (feature 0 > 6.0) Predict: 1.0 Else (feature 3 > 1.6) Predict: 1.0
(6)构建决策树回归模型
//导入所需要的包 import org.apache.spark.ml.evaluation.RegressionEvaluator import org.apache.spark.ml.regression.DecisionTreeRegressionModel import org.apache.spark.ml.regression.DecisionTreeRegressor //训练决策树模型 scala> val dtRegressor = new DecisionTreeRegressor().setLabelCol("indexedLabel") .setFeaturesCol("indexedFeatures") dtRegressor: org.apache.spark.ml.regression.DecisionTreeRegressor = dtr_358e08c37f0c //在pipeline中进行设置 scala> val pipelineRegressor = new Pipeline().setStages(Array(labelIndexer, featureIndexer, dtRegressor, labelConverter)) pipelineRegressor: org.apache.spark.ml.Pipeline = pipeline_ae699675d015 //训练决策树模型 scala> val modelRegressor = pipelineRegressor.fit(trainingData) modelRegressor: org.apache.spark.ml.PipelineModel = pipeline_ae699675d015 //进行预测 scala> val predictionsRegressor = modelRegressor.transform(testData) predictionsRegressor: org.apache.spark.sql.DataFrame = [features: vector, label: string ... 4 more fields] //查看部分预测结果 scala> predictionsRegressor.select("predictedLabel", "label", "features").show(20) +---------------+---------------+-----------------+ | predictedLabel| label| features| +---------------+---------------+-----------------+ | Iris-setosa| Iris-setosa|[4.4,2.9,1.4,0.2]| | Iris-setosa| Iris-setosa|[4.6,3.6,1.0,0.2]| | Iris-virginica|Iris-versicolor|[4.9,2.4,3.3,1.0]| | Iris-setosa| Iris-setosa|[4.9,3.1,1.5,0.1]| | Iris-setosa| Iris-setosa|[4.9,3.1,1.5,0.1]|
(7)评估决策树回归模型
scala> val evaluatorRegressor = new RegressionEvaluator().setLabelCol("ind exedLabel").setPredictionCol("prediction").setMetricName("rmse") evaluatorRegressor: org.apache.spark.ml.evaluation.RegressionEvaluator = regEval_425d2aeea2dd scala> val rmse = evaluatorRegressor.evaluate(predictionsRegressor) rmse: Double = 0.3676073110469039 scala> println("Root Mean Squared Error (RMSE) on test data = " + rmse) Root Mean Squared Error (RMSE) on test data = 0.3676073110469039 scala> val treeModelRegressor = modelRegressor.stages(2).asInstanceOf[Deci sionTreeRegressionModel] treeModelRegressor: org.apache.spark.ml.regression.DecisionTreeRegressionModel = DecisionTreeRegressionModel (uid=dtr_358e08c37f0c) of depth 5 with 13 nodes scala> println("Learned regression tree model:\n" + treeModelRegressor.toDebugString) Learned regression tree model: DecisionTreeRegressionModel (uid=dtr_358e08c37f0c) of depth 5 with 13 nodes If (feature 2 <= 1.9) Predict: 2.0 Else (feature 2 > 1.9) If (feature 2 <= 4.7) If (feature 0 <= 4.9) Predict: 1.0 Else (feature 0 > 4.9) Predict: 0.0 Else (feature 2 > 4.7) If (feature 3 <= 1.6) If (feature 2 <= 4.8) Predict: 0.0 Else (feature 2 > 4.8) If (feature 0 <= 6.0) Predict: 0.5 Else (feature 0 > 6.0) Predict: 1.0 Else (feature 3 > 1.6) Predict: 1.0
从上述结果可以看到模型的标准误差为 0.3676073110469039以及训练的决策树模型结构。