【Spark Mllib】决策树，随机森林——预测森林植被类型_sparkmllib decisiontree

重磅推荐专栏： 《大模型AIGC》
本专栏致力于探索和讨论当今最前沿的技术趋势和应用领域，包括但不限于ChatGPT和Stable Diffusion等。我们将深入研究大型模型的开发和应用，以及与之相关的人工智能生成内容（AIGC）技术。通过深入的技术解析和实践经验分享，旨在帮助读者更好地理解和应用这些领域的最新进展
http://blog.csdn.net/u011239443/article/details/51858825
#数据集处理

import org.apache.spark.mllib.linalg._
import org.apache.spark.mllib.regression._
val rawData = sc.textFile("covtype.data")
val data = rawData.map{	
	line =>
	val values = line.split(",").map( _.toDouble)
	//init返回除最后一个值外的所有值
	val featureVector = Vectors.dense(values.init)
	//决策树要求label从0开始
	val label = values.last -1
	LabeledPoint( label,featureVector)
}

val Array(trainData,cvData,testData) = data.randomSplit( Array(0.8,0.1,0.1))
trainData.cache() 
cvData.cache() //交叉检验集
testData.cache()
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

#模型训练

import org.apache.spark.mllib.tree._
import org.apache.spark.mllib.tree.model._
import org.apache.spark.rdd._

def getMetrics(model: DecisionTreeModel,dta: RDD[ LabeledPoint ]):
	MulticlassMetrics = {
		val predictionsAndLabels = data.map( example =>
		( 
			model.predict( example.features), example.label)
		)
		new MulticlassMetrics( predictionsAndLabels)
	}


val model = DecisionTree.trainClassifier(trainData,7,Map[Int,Int](),"gini",4,100)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

决策树有训练分类模型的函数trainClassifier和回归模型的函数trainRegressor，这里我们使用trainClassifier。
我们来看看trainClassifier都需要什么参数：

scala> DecisionTree.trainClassifier
<console>:42: error: ambiguous reference to overloaded definition,
both method trainClassifier in object DecisionTree of type (input: org.apache.spark.api.java.JavaRDD[org.apache.spark.mllib.regression.LabeledPoint], numClasses: Int, categoricalFeaturesInfo: java.util.Map[Integer,Integer], impurity: String, maxDepth: Int, maxBins: Int)org.apache.spark.mllib.tree.model.DecisionTreeModel
and  method trainClassifier in object DecisionTree of type (input: org.apache.spark.rdd.RDD[org.apache.spark.mllib.regression.LabeledPoint], numClasses: Int, categoricalFeaturesInfo: Map[Int,Int], impurity: String, maxDepth: Int, maxBins: Int)org.apache.spark.mllib.tree.model.DecisionTreeModel
match expected type ?
       DecisionTree.trainClassifier
                    ^
1
2
3
4
5
6
7

• input:数据的LabeledPoint
• numClasses：类别数量
• categoricalFeaturesInfo：
  看下Saprk官网的介绍:

categoricalFeaturesInfo: Specifies which features are categorical and how many categorical values each of those features can take. This is given as a map from feature indices to feature arity (number of categories). Any features not in this map are treated as continuous.

E.g., Map(0 -> 2, 4 -> 10) specifies that feature 0 is binary (taking values 0 or 1) and that feature 4 has 10 categories (values {0, 1, …, 9}). Note that feature indices are 0-based: features 0 and 4 are the 1st and 5th elements of an instance’s feature vector.

Note that you do not have to specify categoricalFeaturesInfo. The algorithm will still run and may get reasonable results. However, performance should be better if categorical features are properly designated.

• impurity：不纯度的类型，有基尼不纯度——“gini”，熵——“entropy”
• maxDepth：对层数进行限制，避免过拟合
• maxBins：决策规则集，可以理解成是决策树的孩子节点的数量

#性能评估

import org.apache.spark.mllib.evaluation._
val metrics = getMetrics(model,cvData)
metrics.confusionMatrix
/*
res6: org.apache.spark.mllib.linalg.Matrix =                                    
156710.0  51350.0   203.0    0.0  0.0    0.0  3577.0
68735.0   207253.0  6883.0   0.0  42.0   0.0  388.0
0.0       5872.0    29882.0  0.0  0.0    0.0  0.0
0.0       0.0       2747.0   0.0  0.0    0.0  0.0
105.0     8702.0    557.0    0.0  129.0  0.0  0.0
0.0       4475.0    12892.0  0.0  0.0    0.0  0.0
11290.0   239.0     55.0     0.0  0.0    0.0  8926.0 
*/
1
2
3
4
5
6
7
8
9
10
11
12
13

因为一共有7种类别，所以生成的是7*7的矩阵，aij 表示实际类别是i，而被预测类别是j的次数。

metrics.precision
//res7: Double = 0.6934452300468837 
1
2

#决策树调优

val evaluations =
      for (impurity <- Array("gini", "entropy");
           depth    <- Array(1, 20);
           bins     <- Array(10, 300))
        yield {
          val model = DecisionTree.trainClassifier(
            trainData, 7, Map[Int,Int](), impurity, depth, bins)
          val accuracy = getMetrics(model, cvData).precision
          ((impurity, depth, bins), accuracy)
        }
evaluations.sortBy(_._2).reverse.foreach( println)
/*
((entropy,20,300),0.9380098861985638)
((gini,20,300),0.9319721451536285)
((entropy,20,10),0.9273681094366382)
((gini,20,10),0.9195954644654499)
((gini,1,10),0.633916339077334)
((gini,1,300),0.6335772755123819)
((entropy,1,300),0.48759922342395684)
((entropy,1,10),0.48759922342395684)
*/
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21

• scala语法：

  for (impurity <- Array("gini", "entropy");
           depth    <- Array(1, 20);
           bins     <- Array(10, 300))
        yield {}
1
2
3
4

相当于关于impurity，depth，bins的三层循环。

#关于categoricalFeaturesInfo
关于categoricalFeaturesInfo这个参数，我们前面直接不设定取值个数:

Map[Int,Int]()
1

但是，我们可以参阅下covtype.info关于数据集的描述：

……

Hillshade_9am quantitative 0 to 255 index Hillshade index at 9am, summer solstice

Hillshade_Noon quantitative 0 to 255 index Hillshade index at noon, summer soltice

Hillshade_3pm quantitative 0 to 255 index Hillshade index at 3pm, summer solstice

Wilderness_Area (4 binary columns) qualitative 0 (absence) or 1 (presence) Wilderness area designation

Soil_Type (40 binary columns) qualitative 0 (absence) or 1 (presence) Soil Type designation

……

Wilderness Areas: 1 – Rawah Wilderness Area
2 – Neota Wilderness Area
3 – Comanche Peak Wilderness Area
4 – Cache la Poudre Wilderness Area

Soil Types: 1 to 40 : based on the USFS Ecological
Landtype Units (ELUs) for this study area

可知：

• 三个Hillshade都有256种取值
• Wilderness Areas 有4中类别，Soil Types 有40种。数据集中是以二元特征的形式，有4列，如取值为3，那么第三列为1，其它列都为0

##重新处理数据集

  def unencodeOneHot(rawData: RDD[String]): RDD[LabeledPoint] = {
    rawData.map { line =>
      val values = line.split(',').map(_.toDouble)
      /*我们可以从covtype.info中得知：wilderness是从第10行开始的，
        slice(10, 14) 截取 10 到 13 行
        indexOf(1.0)  返回值为1的位置编号
     */
      val wilderness = values.slice(10, 14).indexOf(1.0).toDouble
      val soil = values.slice(14, 54).indexOf(1.0).toDouble
      val featureVector = Vectors.dense(values.slice(0, 10) :+ wilderness :+ soil)
      val label = values.last - 1
      LabeledPoint(label, featureVector)
    }
  }
  
    val data = unencodeOneHot(rawData)

    val Array(trainData, cvData, testData) = data.randomSplit(Array(0.8, 0.1, 0.1))
    trainData.cache()
    cvData.cache()
    testData.cache()
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21

##重新评估性能
这里进行参数设置时发现这样的错误：

java.lang.IllegalArgumentException: requirement failed: DecisionTree requires maxBins (= 40) to be at least as large as the number of values in each categorical feature, but categorical feature 6 has 256 values. Considering remove this and other categorical features with a large number of values, or add more training examples.
1

所以：bins数量必须大于等于Max（各个feature的values数量）

val evaluations =
      for (impurity <- Array("gini", "entropy");
      depth    <- Array(10, 20,30);
      bins     <- Array(256, 300))
      yield{
      val model = DecisionTree.trainClassifier(
      trainData,7,Map(6 -> 256,7 -> 256,8 -> 256,10 -> 4,11 -> 40),
      impurity, depth, bins)
      val accurary = getMetrics(model, cvData).precision
      (( impurity,depth,bins), accurary)
      }

evaluations.sortBy(_._2).reverse.foreach( println)
/*
((gini,30,300),0.6327390828416625)
((gini,20,300),0.6319645721602997)
((gini,10,256),0.6190078690285227)
((gini,30,256),0.6165724632193483)
((gini,20,256),0.6149373851142489)
((gini,10,300),0.596522963381135)
((entropy,30,256),0.5868863293701335)
((entropy,20,256),0.5792754710746078)
((entropy,30,300),0.570642258679683)
((entropy,10,256),0.5678006650465051)
((entropy,20,300),0.5645890274211204)
((entropy,10,300),0.5548353562404907)

*/
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

可以看到，结果反而比之前差了很多。这说明这些特征的类别取值有倾斜。

#随机森林
随机森林可以理解将数据集合分成n个子集，然后在每个子集上建立决策树，最后结果是n棵决策树的平均值。
我们看一下所需要的参数：

scala> RandomForest.trainClassifier
<console>:42: error: ambiguous reference to overloaded definition,
both method trainClassifier in object RandomForest of type (input: org.apache.spark.api.java.JavaRDD[org.apache.spark.mllib.regression.LabeledPoint], numClasses: Int, categoricalFeaturesInfo: java.util.Map[Integer,Integer], numTrees: Int, featureSubsetStrategy: String, impurity: String, maxDepth: Int, maxBins: Int, seed: Int)org.apache.spark.mllib.tree.model.RandomForestModel
and  method trainClassifier in object RandomForest of type (input: org.apache.spark.rdd.RDD[org.apache.spark.mllib.regression.LabeledPoint], numClasses: Int, categoricalFeaturesInfo: Map[Int,Int], numTrees: Int, featureSubsetStrategy: String, impurity: String, maxDepth: Int, maxBins: Int, seed: Int)org.apache.spark.mllib.tree.model.RandomForestModel
match expected type ?
       RandomForest.trainClassifier
                    ^

1
2
3
4
5
6
7
8

这里新增的参数有：

• numTrees：树的数量
• featureSubsetStrategy：我们看下spark文档:

featureSubsetStrategy: Number of features to use as candidates for splitting at each tree node. The number is specified as a fraction or function of the total number of features. Decreasing this number will speed up training, but can sometimes impact performance if too low.

我们可以将featureSubsetStrategy设置为auto，让算法自己来决定。

val forest = RandomForest.trainClassifier(
trainData, 7, Map[Int,Int](), 20, "auto", "entropy", 30, 300)
val predictionsAndLabels = data.map(example =>
 (forest.predict(example.features), example.label)
val mul = new  MulticlassMetrics(predictionsAndLabels)
mul.precision
//res59: Double = 0.8690027056239802    
1
2
3
4
5
6
7