mllib数据类型基础统计分类与回归特征提取转换ector.pptxVIP

MLlib

汪文涛

2016.3.8

DataTypes

目录

BasicStatistics

ClassificationandRegression

FeatureExtractionandTransformation

DataTypes

Vector

Labeledpoint

DataTypes

Matrix

RowMatrix

IndexedRowMatrix

CoordinateMatrix

BlockMatrix

BasicStatistics

Summarystatistics

RDD[Vector]

Statistics.colStats()

MultivariateStatisticalSummary

max,min,mean,variance,numberofnonzeros,count

BasicStatistics

Correlations

RDD[Double]

Statistics.corr()

Double

X

Y

RDD[Vector]

Statistics.corr()

Matrix

pearsonorspearman

BasicStatistics

Stratifiedsampling

RDD[(K,V)]

sampleByKey

RDD[(K,V)]

withReplacement(true/false)

fractions(Map[K,Double])

ClassificationandRegression

SVMs

1.SVMWithSGD(Object)

Input:

input:RDD[LabeledPoint],

numIterations:Int,

stepSize:Double,

regParam:Double,

miniBatchFraction:Double,

initialWeights:Vector

(SVMWithSGD.train)

Output:

SVMModel

Predict:

SVMModel.predict(features)

SVMs

2.SVMWithSGD(Class)

valsvmAlg=newSVMWithSGD()

svmAlg.optimizer.

setNumIterations(200).

setRegParam(0.1).

setUpdater(newL1Updater)

valmodelL1=svmAlg.run(training)

ClassificationandRegression

DecisionTrees

Input

numClasses:Numberofclasses(forClassificationonly)

categoricalFeaturesInfo:Specifieswhichfeaturesarecategoricalandhowmanycategoricalvalueseachofthosefeaturescantake

maxDepth:Maximumdepthofatree

minInfoGain:Foranodetobesplitfurther,thesplitmustimproveatleastthismuch

maxBins:Numberofbinsusedwhendiscretizingcontinuousfeatures

impurity:Impuritymeasure(discussedabove)usedtochoosebetweencandidatesplits

trainingData

trainClassifier,trainRegressorortrain(algo:ClassificationorRegression)

Output

DecisionTreeModelthatcanbeusedforprediction

RandomForests,Gradient-BoostedTrees

ClassificationandRegression

NaiveBayes

Input

input:RDDof`(label,arrayoffeatures)`pairs.Everyvectorshouldbeafrequencyvectororacountvector

lambda:Thesmoothingparameter

modelType:ThetypeofNBmod