Scala具有名稱（標簽）的隨機森林特征重要性提取

Question

有什么方法可以從模型中提取特征重要性並附加featureCols名稱，以便於分析？

我有類似的東西：

val featureCols = Array("a","b","c".......... like 67 more)

val assembler = new VectorAssembler().setInputCols(featureCols).setOutputCol("features")
val df2 = assembler.transform(modeling_db)
val labelIndexer = new StringIndexer().setInputCol("def").setOutputCol("label")
val df3 = labelIndexer.fit(df2).transform(df2)
val splitSeed = 5043
val Array(trainingData, testDataCE) = df3.randomSplit(Array(0.7, 0.3), splitSeed)
val classifier = new RandomForestClassifier().setImpurity("gini").setMaxDepth(19).setNumTrees(57).setFeatureSubsetStrategy("auto").setSeed(5043)
val model = classifier.fit(trainingData)

之后，我們嘗試通過以下方式提取重要性：

model.featureImportances

答案確實很難分析：

res14: org.apache.spark.mllib.linalg.Vector = (71,[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,20,23,25,27,33,34,35,38,39,41,42,45,47,48,49,50,51,52,53,54,55,56,57,58,60,61,62,63,64,65,66,67,68,69,70],[0.22362951804309808,0.1830148359365108,0.10246542303449771,0.1699399958851977,0.06486419413350401,0.05187244974385025,0.02627047699833213,0.014498050071723645,0.026182513062665076,0.007126662761055224,0.012375060477018274,0.004354513006816487,0.004361008357237427,0.008435852744278544,0.003195472326415685,0.0023071401643885753,0.004602370417578224,0.0030394399903992345,6.92408316823549E-4,0.011207695216651398,7.609910745572573E-4,8.316382113306638E-4,0.0021506289318167916,0.0013468620354363688,0.006968754359778437,0.018796331618729723,0.0024516591941419444,0.005980997035580654,0.0027983...

有沒有辦法“答應”這個答案並將其附加到原始標簽名稱上？

Answer 1

您在featureCols具有原始列名，並且似乎沒有涉及任何向量，因此您只需將兩個數組zip在一起即可。 對於這樣的輸入數據：

val featureCols = Array("a", "b", "c", "d", "e")
val featureImportance = Vectors.dense(Array(0.15, 0.25, 0.1, 0.35, 0.15)).toSparse

簡單地做

val res = featureCols.zip(featureImportance.toArray).sortBy(-_._2)

通過打印將導致

(d,0.35)
(b,0.25)
(a,0.15)
(e,0.15)
(c,0.1)