Supportvectormachines(SVMs)andrelatedkernel-basedlearningalgorithmsareawell-knownclassofmachinelearningalgorithms,fornon-parametricclassificationandregression.liquidSVMisanimplementationofSVMswhosekeyfeaturesare:
fullyintegratedhyper-parameterselection,extremespeedonbothsmallandlargedatasets,BindingsforR,Python,MATLAB/Octave,Java,andSpark,fullflexibilityforexperts,andinclusionofavarietyofdifferentlearningscenarios:multi-classclassification,ROC,andNeyman-Pearsonlearning,least-squares,quantile,andexpectileregression.Forquestionsandcommentsjustcontactusviamail.Thereyoualsocanasktoberegisterdtoourmailinglist.
liquidSVMislicensedunderAGPL3.0.Incaseyouneedanotherlicense,pleasecontactme.
CommandLineinterfaceInstallationinstructionsforthecommandlineversions.
TerminalversionforLinux/OSXliquidSVM.tar.gzTerminalversionforWindows(64bit)avx2:liquidSVM.zipavx:liquidSVM.zipsse2:liquidSVM.zipPreviousversionsv1.1(June2016),v1.0(January2016)OnLinuxandMacontheterminalliquidSVMcanbeusedinthefollowingway:
wgetwww.isa.uni-stuttgart.de/software/liquidSVM.tar.gztarxzfliquidSVM.tar.gzcdliquidSVMmakeallscripts/mc-svm.shbanana-mc12RReadthedemovignetteforatutorialoninstallingliquidSVM-packageandhowtouseitandthedocumentationvignetteformoreadvancedinstallationoptionsandusage.
Aneasyusageis:
install.packages("liquidSVM")library(liquidSVM)banana<-liquidData('banana-mc')model<-mcSVM(Y~.,banana$train,display=1,threads=2)result<-test(model,banana$test)errors(result)PythonReadthedemonotebookforatutorialoninstallingliquidSVM-packageandhowtouseitandthehomepageformoreadvancedinstallationoptionsandusage.
Toinstalluse:
pipinstall--userliquidSVMandtheninPythonyoucanuseite.g.like:
fromliquidSVMimport*banana=LiquidData('banana-mc')model=mcSVM(banana.train,display=1,threads=2)result,err=model.test(banana.test)MATLAB/OctaveTheMATLABbindingsarecurrentlygettingabetterinterface,andthisisapreviewversion.
ItdoescurrentlynotworkonWindows.
ForinstallationdownloadtheToolboxliquidSVM.mltbxandinstallitinMATLABbydoubleclickingit.Tocompileandaddpathsissue:
makeliquidSVMnativeThenyoucanuseitlike:
banana=liquidData('banana-mc');model=svm_mc(banana.train,'DISPLAY',1,'THREADS',2);[result,err]=model.test(banana.test);MostofthecodealsoworksinOctaveifyouuseliquidSVM-octave.zip.
JavaThemainhomepageishere.ForinstallationdownloadliquidSVM-java.zipandunzipit.Theclassesareallinpackagede.uni_stuttgart.isa.liquidsvmandaneasyexampleis:
LiquidDatabanana=newLiquidData("banana-mc");SVMmodel=newMC(banana.train,newConfig().display(1).threads(2));ResultAndErrorsresult=model.test(banana.test);IfthisisimplementedinthefileExample.javathiscanbecompiledandrunusing
#ifyouwanttocompiletheJNI-nativelibrary:makelib#compileyourJava-Codejavac-classpathliquidSVM.jarExample.java#andrunitjava-Djava.library.path=.-cp.:liquidSVM.jarExampleSparkThisisapreviewversion,seeSparkformoredetails.DownloadliquidSVM-spark.zipandunzipit.AssumeyouhaveSparkinstalledin$SPARK_HOMEyoucanissue:
makelibexportLD_LIBRARY_PATH=.:$LD_LIBRARY_PATH$SPARK_HOME/bin/spark-submit\--classde.uni_stuttgart.isa.liquidsvm.spark.App\liquidSVM-spark.jarbanana-mcIfyouhaveconfiguredSparktobeusedonaclusterwithHadoopuse:
hdfsdfs-putdata/covtype-full.train.csvdata/covtype-full.test.csv.makelib$SPARK_HOME/bin/spark-submit--files../libliquidsvm.so\--confspark.executor.extraLibraryPath=.\--confspark.driver.extraLibraryPath=.\--classde.uni_stuttgart.isa.liquidsvm.spark.App\--num-executors14liquidSVM-spark.jarcovtype-fullExtraDatasetsfortheDemocovertypedatasetwith35.090trainingand34.910testsamples
covertypedatasetwith522.909trainingand58.103testsamples
BothdatasetswerecompiledfromLIBSVM'sversionofthecovertypedataset,whichinturnwastakenfromtheUCIrepositoryandpreprocessedasin[RC02a].CopyrightforthisdatasetisbyJockA.BlackardandColoradoStateUniversity.
CitationIfyouuseliquidSVM,pleaseciteitas:
I.SteinwartandP.Thomann.liquidSVM:AfastandversatileSVMpackage.ArXive-prints1702.06899,February2017.
评论