MachieLearigCollectio
Microsoftcotributiglibraries,tools,recipes,samplecodesadworkshopcotetsformachielearig&deeplearig.
TableofCotetsBoostigAutoMLNeuralNetworkGraph&NetworkVisioTimeSeriesNLPOlieMachieLearigRecommedatioDistributedCausalIfereceResposibleAIOptimizatioReiforcemetLearigWidowsDatasetsDebugPipeliePlatformTaggigDevelopertoolSampleCodeWorkshopBookLearigBlog,News&WebiarBoostigLightGBM-Afast,distributed,highperformacegradietboostigframeworkExplaiableBoostigMachies-iterpretablemodeldevelopediMicrosoftResearchusigbaggig,gradietboostig,adautomaticiteractiodetectiotoestimatedgeeralizedadditivemodels.AutoMLNeuralNetworkItelligece-AopesourceAutoMLtoolkitforautomatemachieleariglifecycle,icludigfeatureegieerig,euralarchitecturesearch,modelcompressioadhyper-parametertuig.Archai-ReproducibleRapidResearchforNeuralArchitectureSearch(NAS).FLAML-AfastadlightweightAutoMLlibrary.AzureAutomatedMachieLearig-AutomatedMachieLearigforTabulardata(regressio,classificatioadforecastig)byAzureMachieLearigNeuralNetworkPyMarli-LightweightDeepLearigModelTraiiglibrarybasedoPyTorch.bayesiaize-ABayesiaeuraletworkwrapperipytorch.O-CNN-Octree-basedcovolutioaleuraletworksfor3Dshapeaalysis.ResNet-deepresidualetwork.CNTK-microsoftcogitivetoolkit(CNTK),opesourcedeep-learigtoolkit.IfiiBatch-Efficiet,check-poiteddataloadigfordeeplearigwithmassivedatasets.ModelsuderHuggigFace-MicrosoftsharestrasformermodelsatHuggigFace.51pretraiedmodels(asofJue28,2021).Muzic-MusicUderstadigadGeeratiowithArtificialItelligece.Graph&Networkgraspologic-utilitiesadalgorithmsdesigedfortheprocessigadaalysisofgraphswithspecializedgraphstatisticalalgorithms.TFGraphNeuralNetworkSamples-tesorFlowimplemetatiosofgrapheuraletworks.ptg-PyTorchGraphNeuralNetworkLibraryStemGNN-spectraltemporalgrapheuraletwork(StemGNN)formultivariatetime-seriesforecastig.SPTAG-adistributedapproximateearesteighborhoodsearch(ANN)library.VisioMicrosoftVisioModelResNet50-alargepretraiedvisioResNet-50modelusigsearchegie'sweb-scaleimagedata.Oscar-Object-SematicsAligedPre-traiigforVisio-LaguageTasks.TorchGeo-aPyTorchdomailibrary,similartotorchvisio,thatprovidesdatasets,trasforms,samplers,adpre-traiedmodelsspecifictogeospatialdata.SwiTrasformer-aofficialimplemetatiofor"SwiTrasformer:HierarchicalVisioTrasformerusigShiftedWidows".TimeSerieslumiol-aomalydetectioadcorrelatiolibrary.SR-CNN-SpectralResidualbasedaomalydetectioalgorithm,SR-CNNimplemetatio.Greykite-flexible,ituitiveadfastforecaststhroughitsflagshipalgorithm,Silverkite.TemporalClusterMatchigforChageDetectioofStructuresfromSatelliteImagery-Aimplemetatioofthetemporalclustermatchigmethodfordetectigchageistructurefootpritsfromtimeseriesofremotelysesedimagery.MicrosoftFiaceTimeSeriesForecastigFramework-aforecastigpackagethatutilizescuttig-edgetimeseriesforecastigadparallelizatioothecloudtoproduceaccurateforecastsforfiacialdata.NLPT-ULRv2-Turigmultiliguallaguagemodel.Turig-NLG-TurigNaturalLaguageGeeratio,17billio-parameterlaguagemodel.DeBERTa-Decodig-ehacedBERTwithDisetagledAttetioUiLM-UifiedLaguageModelPre-traiig/Pre-traiigforNLPadBeyodUicoder-Uicodermodelforuderstadigadgeeratio.NeuroBlocks-buildigyourlpdmodelslikeplayiglegoMultiligualModelTrasfer-ewdeeplearigmodelsforbootstrappiglaguageuderstadigmodelsforlaguageswitholabeleddatausiglabeleddatafromotherlaguages.MT-DNN-multi-taskdeepeuraletworksforaturallaguageuderstadig.imt-iteractiveeuralmachietraislatio-liteOpeKP-automaticallyextractigkeyphrasesthataresaliettothedocumetmeaigsisaessetialstepisematicdocumetuderstadig.DeText-adeepeuraltextuderstadigframeworkforrakigadclassificatiotasks.Gealog-aopesource,cross-platformpythopackageallowiggeeratioofsytheticdocumetimageswithcustomdegradatiosadtextaligmetcapabilities.FastFormers-highlyefficiettrasformermodelsforNLU.VERSEAGILITY-aPytho-basedtoolkittorampupyourcustomaturallaguageprocessig(NLP)task,allowigyoutobrigyourowdataadbrigmodelsitoproductio.ItisacetralcompoetoftheMicrosoftDataScieceToolkit.DPUUtilities-UtilitiesusedbytheDeepProgramUderstadigteam.OlieMachieLearigVowpalWabbit-fast,efficiet,adflexibleoliemachielearigtechiquesforreiforcemetlearig,supervisedlearig,admore.RecommedatioRecommeders-examplesadbestpracticsforbuildigrecommedatiosystems(A2SVD,DKN,xDeepFM,LightGBM,LSTUR,NAML,NPA,NRMS,RLRMC,SAR,VowpalWabbitareiveted/cotributedbyMicrosoft).GDMIX-AdeeprakigpersoalizatioframeworkrakerEval-Afastumpy-basedimplemetatioofrakigmetricsforiformatioretrievaladrecommedatio.DistributedDeepSpeed-DeepSpeedisadeeplearigoptimizatiolibrarythatmakesdistributedtraiigeasy,efficiet,adeffective.MMLSpark-machieleariglibraryospark.photo-ml-ascalablemachieleariglibraryoapachespark.ToY-frameworktoativelyrudeeplearigframeworksoapachehadoop.isolatio-forest-ASpark/Scalaimplemetatiooftheisolatioforestusupervisedoutlierdetectioalgorithm.CausalIfereceEcoML-Pythopackageforestimatigheterogeeoustreatmeteffectsfromobservatioaldataviamachielearig.DoWhy-Pytholibraryforcausaliferecethatsupportsexplicitmodeligadtestigofcausalassumptios.ResposibleAIIterpretML-atoolkittohelpuderstadmodelsadeableresposbilemachielearig.IterpretCommuity-extedsiterpretrepowithadditioaliterpretabilitytechiquesadutilityfuctios.DiCE-diversecouterfactualexplaatios.Iterpret-Text-state-of-the-artexplaiersfortext-basedmlmodelsadvisualizewithdashboard.fairlear-pythopackagetoassessadimprovefairessofmachielearigmodels.LiFT-likedifairesstoolkit.RobustDG-Toolkitforbuildigmachielearigmodelsthatgeeralizetouseedomaisadarerobusttoprivacyadotherattacks.SHAP-agametheoreticapproachtoexplaitheoutputofaymachielearigmodel(scottludbert,MicrosoftResearch).LIME-explaiigthepredictiosofaymachielearigclassifier(Marco,MicrosoftResearch).BackwardCompatibilityML-ProjectforopesourcigresearcheffortsoBackwardCompatibilityiMachieLearigcofidetial-ml-utils-PythoutilitiesfortraiigaddeployigMLmodelsagaistdatayouca'tsee.presidio-cotextaware,pluggableadcustomizabledataprotectioadaoymizatioservicefortextadimages.Presidio-research-Thispackagefeaturesdata-sciecerelatedtasksfordevelopigewrecogizersforPresidio.CofidetialONNXIfereceServer-AOpeEclaveportoftheONNXifereceserverwithdataecryptioadattestatiocapabilitiestoeablecofidetialifereceoAzureCofidetialComputig.Resposible-AI-Widgets-resposibleAIuseriterfacesforFairlear,iterpret-commuity,adErrorAalysis,aswellasfoudatioalbuildigblocksthattheyrelyo.ErrorAalysis-Atoolkittohelpaalyzeadimprovemodelaccuracy.SecureDataSadbox-Atoolkitforcoductigmachielearigtrialsagaistcofidetialdata.shrike-Pythoutilitiestoaid"compliatexperimet"iAzureMachieLearig-traiigMLmodelswithoutseeigthetraiigdata.OptimizatioONNXRutime-cross-platfom,highperformaceMLifereceadtraiigaccelerator.ONNXRutimeforPyTorch-AcceleratePyTorchmodelswithONNXRutime.ONNXRutimeTraiigExamples-examplesforusigoxrutimeformodeltraiig.ONNXCoverter-commoutilitiesforoxcoverters.ONNX.js-ruoxmodelsusigjavascript.ONNX.jsDemo-demosforONNX.js.Olive-asequeceofdockerimagesthatautomatestheprocessofONNXmodelshippig.Hummigbird-compiletraiedmlmodelitotesorcomputatioforfasteriferece.EdgeML-providescodeformachielearigalgorithmsforedgedevicesdevelopedatMicrosoftResearchIdia.DirectML-high-performace,hardware-acceleratedDirectX12libraryformachielearig.MMd-MMdisasetoftoolstohelpusersiter-operateamogdifferetdeeplearigframeworks.E.g.modelcoversioadvisualizatio.iifiibatch-Efficiet,check-poiteddataloadigfordeeplearigwithmassivedatasets.IfereceSchema-Schemadecoratioforiferececodefusio-flexibleadefficietdeepeuraletworkcompiler.ReiforcemetLearigAirSim-opesourcesimulatorforautoomousvehiclesbuildourealegie/uityfrommicrosoftresearch.TextWorld-TextWorldisasadboxlearigevirometforthetraiigadevaluatioofreiforcemetlearig(RL)agetsotext-basedgames.Moab-ProjectMoab,aewope-sourcebalacigrobottohelpegieersaddeveloperslearhowtobuildreal-worldautoomouscotrolsystemswithProjectBosai.MARO-multi-agetresourceoptimizatio(MARO)platfom.TraiigData-DriveorSurrogateSimulators-buildsimulatiofromdataforuseiRLadBosaiplatformformachieteachig.Bosai-lowcodeidustrialmachieteachigplatform.BosaiPythoSDK-ApytholibraryforitegratigdatasourceswithBosaiBRAIN.Securitycouterfit-aCLIthatprovidesageericautomatiolayerforassessigthesecurityofMLmodels.WidowsWidowsMachieLearig-MachieLearigoWidows.DatasetsCOCODataset-COCOisalarge-scaleobjectdetectio,segmetatio,adcaptioigdataset.MSMARCO-collectioofdatasetsfocusedodeeplearigisearch.IerEyeCreateDataset-IerEyedatasetcreatiotoolforIerEye-DeepLeariglibrary.TrasformsDICOMdataitomaskfortraiigDeepLearigmodels.SepsisCohortfromMIMICIII-SepsiscohortfromMIMICdataset.MIND:MicrosoftNewsDataset-alarge-scaledatasetforewsrecommedatioresearch.DatasetforAIforEarth-AIForEarthDataSetsisacollectioofdatasetsforAIresearch.ORBIT-acollectioofvideosofobjectsicleaadclutteredsceesrecordedbypeoplewhoareblid/low-visiooamobilephoe.Debug&Bechmarktesorwatch-debuggig,moitorigadvisualizatioforpythomachielearigaddatasciece.PYRIGHT-statictypecheckerforpytho.BechML-Pytholibrarytobechmarkpopularpre-builtcloudAIAPIs.debugpy-AimplemetatiooftheDebugAdapterProtocolforPythokieto-ACPU+GPUProfiliglibrarythatprovidesaccesstotimelietracesadhardwareperformacecouterscotributedbyAzureAIPlatformteam.SuperBechmark-abechmarkigaddiagosistoolforAIifrastructure(software&hardware).PipelieGitHubActios-Automateallyoursoftwareworkflows,owwithworld-classCI/CD.Build,test,addeployyourcoderightfromGitHub.AzurePipelies-AutomateyourbuildsaddeploymetswithPipeliessoyouspedlesstimewiththeutsadboltsadmoretimebeigcreative.Dagli-frameworkfordefiigmachielearigmodels,icludigfeaturegeeratioadtrasformatiosasDAG.PlatformAIforEarthAPIPlatform-distributedifrastructuredesigedtoprovideasecure,scalable,adcustomizableAPIhostig,desigedtohadletheeedsoflog-ruig/asychroousmachielearigmodeliferece.
OpePlatfomforAI(OpePAI)-resourcescheduligadclustermaagemetforAI.
OpePAIRutime-Rutimefordeeplearigworkload.OpePAIProtocol-OpePAIprotocoleablesjobsharigadportability.Opepaimarketplace-Amarketplacewhichstoresexamplesadjobtemplatesofopepai.OpePAIFrameworkCotroller-builttoorchestrateallkidsofapplicatiosoKuberetesbyasiglecotroller.HivedDScheduler-KuberetesSchedulerforDeepLearig.OpePAIJSSDK-TheJavaScriptSDKisdesigedtofacilitatethedevelopersofOpePAItoofferuserfriedlyexperiece.OpePAIVSCodeCliet-ExtesiotocoectOpePAIclusters,submitAIjobs,simulatejobslocally,maagefiles,adsoo.MLOS-DataSciecepoweredifrastructureadmethodologytodemocratizeadautomatePerformaceEgieerig.
PlatformforSituatedItelligece-aope-sourceframeworkformultimodal,itegrativeAI.
Qlib-aAI-orietedquatitativeivestmetplatform.
TaggigTagAomaly-Aomalydetectioaalysisadlabeligtool,specificallyformultipletimeseries(oetimeseriespercategory)VoTT-VisualobjecttaggigtoolDevelopertoolVisualStudioCode-Codeeditorredefiedadoptimizedforbuildigaddebuggigmoderwebadcloudapplicatios.Gather-addsgatherfuctioalityithePytholaguagetotheJupyterExtesio.Pylace-aextesiothatworksalogsidePythoiVisualStudioCodetoprovideperformatlaguagesupport.AzureMLSippets-VSCodesippetsforAzureMachieLearigSampleCodeMicrosoftAIforEarth
Sharedutilities-AcollectioofutilitiesforworkigwithAzureMachieLearig.acoustic-bird-detectio-Tutorial:AccurateBioacousticSpeciesDetectiofromSmallNumbersofTraiigClipsUsigtheBiophoyModelbeluasoud-Usigmachielearigtodetectbelugawhalecallsihydrophoerecordigs.arcticseals-detect&classifyarcticsealsiaerialimagerytouderstadhowthey’readaptigtoachagigworld.AIDE:AotatioIterfaceforData-driveEcology-Detectigadclassifyigwildlifeiaerialimagery.CameraTrapTool-toolsfortraiigadruigdetectorsadclassifiersforwildlifeimagescollectedfrommotio-triggeredcameras.LadcovermappigtheOrioquíaregio-AtoolforpredictigladcoveritheOrioquiaregioofPeru.PlaetaryComputerHub-adevelopmetevirometthatmakesourdataadAPIsaccessiblethroughfamiliar,ope-sourcetools,adallowsuserstoeasilyscaletheiraalyses.Poultrybarmappig-codefordetectigpoultrybarsfromhigh-resolutioaerialimageryadaaccompayigdatasetofpredictedbarsovertheUitedStates.PlaetaryComputerSDKforPytho-APythoSDKforthePlaetaryComputerHub.SpeciesClassificatio-Atoolforclassifyigspeciesiimages.NewsThreads-TheNewsThreadsprojectaalyzesewsarticlestohelpfidsimilaritiesbetweeewsarticlesadtraceewsproveaceacrosstime.
IerEyeDeepLearig-MedicalImagigDeepLeariglibrarytotraiaddeploymodelsoAzureMachieLearigadAzureStack
DeepSeismic-DeepLearigforSeismicImagigadIterpretatio
Multi-speciesbioacousticclassificatio-Multi-speciesbioacousticclassificatiousigdeeplearigalgorithms.
NestleAceAssessmet-deeplearigmodelstoassesstheaceseveritylevelbasedoselfieimages.
VisualAalogies-explorigthecoectiosbetweeartworkswithdeep"VisualAalogies".
ForecastigBestPractices-timeseriesforecastigbestpractices&examples.
ComputerVisioRecipes-bestpractices,codesamples,addocumetatioforComputerVisio.
AzureMLDesigerSample-samplesofAzureMachieLearigdesiger.
DeepSpeedExamples-ExamplemodelsusigDeepSpeed
ATALEOFTHREECITIES-Aalyzigthesafety(311)datasetpublishedbyAzureOpeDatasetsforChicago,BostoadNewYorkCityusigSparkR,SParkSQL,AzureDatabricks,visualizatiousigggplot2adleaflet.
MicrosoftHealthItelligeceMachieLearigToolbox-MicrosoftHealthItelligeceAzureMachieLearigToolbox.
MicrosoftSolutioAccelerators
MLOpsSolutioAccelerator-thisrepositoryhelpsMLteamstoacceleratetheirmodeldeploymettoproductioleveragigAzure.AomalyDetectioSolutioAccelerator-implemetAomalyDetectiowhichisthetechiqueofidetifyigrareevetsorobservatioswhichcaraisesuspiciosbybeigstatisticallydifferetfromtherestoftheobservatios.ClassificatioSolutioAccelerator-Thisisaclassificatiosolutioacceleratortohelpyoubuildaddeployabiaryclassificatioproject.CommuityAI@EdgeCommuity-fidtheresourcesyoueedtocreatesolutiosusigitelligeceattheedgethroughcombiatiosofhardware,machielearig(ML),artificialitelligece(AI)adMicrosoftAzureservice.GlobalAICommuity-empowersdeveloperswhoarepassioateaboutAItosharekowledgethroughevetsadmeetups.DeepLearigLab(Japa)-providesiformatioodevelopmetcasesadthelatesttechologytredsrelatedtodeeplearig.WorkshopWorkshopWikiCompetitio2020MachieLearigSecurityEvasioCompetitio-codesamplesforthe2020MachieLearigSecurityEvasioCompetitio.BookPRML-PatterRecogitioadMachieLearigbyChristopherBishopFoudatiosofDataSciece-BasicTheoryforDataSciece.MasterigAzureMachieLearig:Performlarge-scaleed-to-edadvacedmachielearigithecloudwithMicrosoftAzureMachieLearigPracticalAutomatedMachieLearigoAzure:UsigAzureMachieLearigtoQuicklyBuildAISolutiosLearigMicrosoftLear-LearigcotetsforMicrosofttechologyDataScietist,AIEgieerDataScieceforMaager-Geeralizatio,Utility,adExperimetatio:MLCoceptsforMakigBetterBusiessDecisiosGithubLearigLab-learigcotetsforGithubtechology.GettigstartedwithPytho-SamplecodeforChael9PythoforBegierscourse.PythoforBegiersMorePythoforBegiersEveMorePythoforBegiersGetstartedwithPyTorch-learthefudametalsofdeeplearigwithPyTorch.DevItrotoDataSciece-Ithis28-videoseries,youwilllearimportatcoceptsadtechologiestobuildyoured-to-edmachielearigapplicatiosoAzure.MachieLearigforBegiers-ACurriculum-12weeks,24lessos,classicMachieLearigforallDataScieceforBegiers-ACurriculum-10Weeks,20Lessos,DataScieceforAll!MicrosoftCloudWorkshop-WideWorldImporters(WWI)deliversiovativesolutiosformaufacturers.Blog,News&Webiarchael9-AIShow-videosfordevelopersfrompeoplebuildigMicrosoftproductsadservices.MicrosoftOpeSourceBlog-blogaboutmicrosoftopesourcetechology.MicrosoftResearchEvet,Coferece&Webiars-Evets,Cofereces&WebiarsbyMicrosoftResearch.MicrosoftIovatioTechHub-AIprojectiMicrosoft.LikedIEgieerigBlog-BlogbyLikedIEgieerigTeamAISystem-systemforAIEducatioResource(Chiese).AIEdu-AIeducatiomaterialsforChiesestudets,teachersadITprofessioals(Chiese).---CotributigThisprojectwelcomescotributiosadsuggestios.MostcotributiosrequireyoutoagreetoaCotributorLiceseAgreemet(CLA)declarigthatyouhavetherightto,adactuallydo,gratustherightstouseyourcotributio.Fordetails,visithttps://cla.opesource.microsoft.com.
Wheyousubmitapullrequest,aCLAbotwillautomaticallydetermiewhetheryoueedtoprovideaCLAaddecoratethePRappropriately(e.g.,statuscheck,commet).Simplyfollowtheistructiosprovidedbythebot.YouwillolyeedtodothisoceacrossallreposusigourCLA.
ThisprojecthasadoptedtheMicrosoftOpeSourceCodeofCoduct.FormoreiformatioseetheCodeofCoductFAQorcotactopecode@microsoft.comwithayadditioalquestiosorcommets.
TrademarksThisprojectmaycotaitrademarksorlogosforprojects,products,orservices.AuthorizeduseofMicrosofttrademarksorlogosissubjecttoadmustfollowMicrosoft'sTrademark&BradGuidelies.UseofMicrosofttrademarksorlogosimodifiedversiosofthisprojectmustotcausecofusioorimplyMicrosoftsposorship.Ayuseofthird-partytrademarksorlogosaresubjecttothosethird-party'spolicies.
评论