Microsoftcontributinglibraries,tools,recipes,samplecodesandworkshopcontentsformachinelearning&deeplearning.
TableofContentsBoostingAutoMLNeuralNetworkGraph&NetworkVisionTimeSeriesNLPOnlineMachineLearningRecommendationDistributedCausalInferenceResponsibleAIOptimizationReinforcementLearningWindowsDatasetsDebugPipelinePlatformTaggingDevelopertoolSampleCodeWorkshopBookLearningBlog,News&WebinarBoostingLightGBM-Afast,distributed,highperformancegradientboostingframeworkExplainableBoostingMachines-interpretablemodeldevelopedinMicrosoftResearchusingbagging,gradientboosting,andautomaticinteractiondetectiontoestimatedgeneralizedadditivemodels.AutoMLNeuralNetworkIntelligence-AnopensourceAutoMLtoolkitforautomatemachinelearninglifecycle,includingfeatureengineering,neuralarchitecturesearch,modelcompressionandhyper-parametertuning.Archai-ReproducibleRapidResearchforNeuralArchitectureSearch(NAS).FLAML-AfastandlightweightAutoMLlibrary.AzureAutomatedMachineLearning-AutomatedMachineLearningforTabulardata(regression,classificationandforecasting)byAzureMachineLearningNeuralNetworkPyMarlin-LightweightDeepLearningModelTraininglibrarybasedonPyTorch.bayesianize-ABayesianneuralnetworkwrapperinpytorch.O-CNN-Octree-basedconvolutionalneuralnetworksfor3Dshapeanalysis.ResNet-deepresidualnetwork.CNTK-microsoftcognitivetoolkit(CNTK),opensourcedeep-learningtoolkit.InfiniBatch-Efficient,check-pointeddataloadingfordeeplearningwithmassivedatasets.ModelsunderHuggingFace-MicrosoftsharestransformermodelsatHuggingFace.51pretrainedmodels(asofJune28,2021).Muzic-MusicUnderstandingandGenerationwithArtificialIntelligence.Graph&Networkgraspologic-utilitiesandalgorithmsdesignedfortheprocessingandanalysisofgraphswithspecializedgraphstatisticalalgorithms.TFGraphNeuralNetworkSamples-tensorFlowimplementationsofgraphneuralnetworks.ptgnn-PyTorchGraphNeuralNetworkLibraryStemGNN-spectraltemporalgraphneuralnetwork(StemGNN)formultivariatetime-seriesforecasting.SPTAG-adistributedapproximatenearestneighborhoodsearch(ANN)library.VisionMicrosoftVisionModelResNet50-alargepretrainedvisionResNet-50modelusingsearchengine'sweb-scaleimagedata.Oscar-Object-SemanticsAlignedPre-trainingforVision-LanguageTasks.TorchGeo-aPyTorchdomainlibrary,similartotorchvision,thatprovidesdatasets,transforms,samplers,andpre-trainedmodelsspecifictogeospatialdata.SwinTransformer-anofficialimplementationfor"SwinTransformer:HierarchicalVisionTransformerusingShiftedWindows".TimeSeriesluminol-anomalydetectionandcorrelationlibrary.SR-CNN-SpectralResidualbasedanomalydetectionalgorithm,SR-CNNimplementation.Greykite-flexible,intuitiveandfastforecaststhroughitsflagshipalgorithm,Silverkite.TemporalClusterMatchingforChangeDetectionofStructuresfromSatelliteImagery-Animplementationofthetemporalclustermatchingmethodfordetectingchangeinstructurefootprintsfromtimeseriesofremotelysensedimagery.MicrosoftFinanceTimeSeriesForecastingFramework-aforecastingpackagethatutilizescutting-edgetimeseriesforecastingandparallelizationonthecloudtoproduceaccurateforecastsforfinancialdata.NLPT-ULRv2-Turingmultilinguallanguagemodel.Turing-NLG-TuringNaturalLanguageGeneration,17billion-parameterlanguagemodel.DeBERTa-Decoding-enhancedBERTwithDisentangledAttentionUniLM-UnifiedLanguageModelPre-training/Pre-trainingforNLPandBeyondUnicoder-Unicodermodelforunderstandingandgeneration.NeuronBlocks-buildingyournlpdnnmodelslikeplayinglegoMultilingualModelTransfer-newdeeplearningmodelsforbootstrappinglanguageunderstandingmodelsforlanguageswithnolabeleddatausinglabeleddatafromotherlanguages.MT-DNN-multi-taskdeepneuralnetworksfornaturallanguageunderstanding.inmt-interactiveneuralmachinetrainslation-liteOpenKP-automaticallyextractingkeyphrasesthataresalienttothedocumentmeaningsisanessentialstepinsemanticdocumentunderstanding.DeText-adeepneuraltextunderstandingframeworkforrankingandclassificationtasks.Genalog-anopensource,cross-platformpythonpackageallowinggenerationofsyntheticdocumentimageswithcustomdegradationsandtextalignmentcapabilities.FastFormers-highlyefficienttransformermodelsforNLU.VERSEAGILITY-aPython-basedtoolkittorampupyourcustomnaturallanguageprocessing(NLP)task,allowingyoutobringyourowndataandbringmodelsintoproduction.ItisacentralcomponentoftheMicrosoftDataScienceToolkit.DPUUtilities-UtilitiesusedbytheDeepProgramUnderstandingteam.OnlineMachineLearningVowpalWabbit-fast,efficient,andflexibleonlinemachinelearningtechniquesforreinforcementlearning,supervisedlearning,andmore.RecommendationRecommenders-examplesandbestpracticsforbuildingrecommendationsystems(A2SVD,DKN,xDeepFM,LightGBM,LSTUR,NAML,NPA,NRMS,RLRMC,SAR,VowpalWabbitareinvented/contributedbyMicrosoft).GDMIX-AdeeprankingpersonalizationframeworkrankerEval-Afastnumpy-basedimplementationofrankingmetricsforinformationretrievalandrecommendation.DistributedDeepSpeed-DeepSpeedisadeeplearningoptimizationlibrarythatmakesdistributedtrainingeasy,efficient,andeffective.MMLSpark-machinelearninglibraryonspark.photon-ml-ascalablemachinelearninglibraryonapachespark.TonY-frameworktonativelyrundeeplearningframeworksonapachehadoop.isolation-forest-ASpark/Scalaimplementationoftheisolationforestunsupervisedoutlierdetectionalgorithm.CausalInferenceEconML-Pythonpackageforestimatingheterogeneoustreatmenteffectsfromobservationaldataviamachinelearning.DoWhy-Pythonlibraryforcausalinferencethatsupportsexplicitmodelingandtestingofcausalassumptions.ResponsibleAIInterpretML-atoolkittohelpunderstandmodelsandenableresponsbilemachinelearning.InterpretCommunity-extendsinterpretrepowithadditionalinterpretabilitytechniquesandutilityfunctions.DiCE-diversecounterfactualexplanations.Interpret-Text-state-of-the-artexplainersfortext-basedmlmodelsandvisualizewithdashboard.fairlearn-pythonpackagetoassessandimprovefairnessofmachinelearningmodels.LiFT-linkedinfairnesstoolkit.RobustDG-Toolkitforbuildingmachinelearningmodelsthatgeneralizetounseendomainsandarerobusttoprivacyandotherattacks.SHAP-agametheoreticapproachtoexplaintheoutputofanymachinelearningmodel(scottlundbert,MicrosoftResearch).LIME-explainingthepredictionsofanymachinelearningclassifier(Marco,MicrosoftResearch).BackwardCompatibilityML-ProjectforopensourcingresearcheffortsonBackwardCompatibilityinMachineLearningconfidential-ml-utils-PythonutilitiesfortraininganddeployingMLmodelsagainstdatayoucan'tsee.presidio-contextaware,pluggableandcustomizabledataprotectionandanonymizationservicefortextandimages.Presidio-research-Thispackagefeaturesdata-sciencerelatedtasksfordevelopingnewrecognizersforPresidio.ConfidentialONNXInferenceServer-AnOpenEnclaveportoftheONNXinferenceserverwithdataencryptionandattestationcapabilitiestoenableconfidentialinferenceonAzureConfidentialComputing.Responsible-AI-Widgets-responsibleAIuserinterfacesforFairlearn,interpret-community,andErrorAnalysis,aswellasfoundationalbuildingblocksthattheyrelyon.ErrorAnalysis-Atoolkittohelpanalyzeandimprovemodelaccuracy.SecureDataSandbox-Atoolkitforconductingmachinelearningtrialsagainstconfidentialdata.shrike-Pythonutilitiestoaid"compliantexperiment"inAzureMachineLearning-trainingMLmodelswithoutseeingthetrainingdata.OptimizationONNXRuntime-cross-platfom,highperformanceMLinferenceandtrainingaccelerator.ONNXRuntimeforPyTorch-AcceleratePyTorchmodelswithONNXRuntime.ONNXRuntimeTrainingExamples-examplesforusingonnxruntimeformodeltraining.ONNXConverter-commonutilitiesforonnxconverters.ONNX.js-runonnxmodelsusingjavascript.ONNX.jsDemo-demosforONNX.js.Olive-asequenceofdockerimagesthatautomatestheprocessofONNXmodelshipping.Hummingbird-compiletrainedmlmodelintotensorcomputationforfasterinference.EdgeML-providescodeformachinelearningalgorithmsforedgedevicesdevelopedatMicrosoftResearchIndia.DirectML-high-performance,hardware-acceleratedDirectX12libraryformachinelearning.MMdnn-MMdnnisasetoftoolstohelpusersinter-operateamongdifferentdeeplearningframeworks.E.g.modelconversionandvisualization.inifinibatch-Efficient,check-pointeddataloadingfordeeplearningwithmassivedatasets.InferenceSchema-Schemadecorationforinferencecodennfusion-flexibleandefficientdeepneuralnetworkcompiler.ReinforcementLearningAirSim-opensourcesimulatorforautonomousvehiclesbuildonunrealengine/unityfrommicrosoftresearch.TextWorld-TextWorldisasandboxlearningenvironmentforthetrainingandevaluationofreinforcementlearning(RL)agentsontext-basedgames.Moab-ProjectMoab,anewopen-sourcebalancingrobottohelpengineersanddeveloperslearnhowtobuildreal-worldautonomouscontrolsystemswithProjectBonsai.MARO-multi-agentresourceoptimization(MARO)platfom.TrainingData-DrivenorSurrogateSimulators-buildsimulationfromdataforuseinRLandBonsaiplatformformachineteaching.Bonsai-lowcodeindustrialmachineteachingplatform.BonsaiPythonSDK-ApythonlibraryforintegratingdatasourceswithBonsaiBRAIN.Securitycounterfit-aCLIthatprovidesagenericautomationlayerforassessingthesecurityofMLmodels.WindowsWindowsMachineLearning-MachineLearningonWindows.DatasetsCOCODataset-COCOisalarge-scaleobjectdetection,segmentation,andcaptioningdataset.MSMARCO-collectionofdatasetsfocusedondeeplearninginsearch.InnerEyeCreateDataset-InnerEyedatasetcreationtoolforInnerEye-DeepLearninglibrary.TransformsDICOMdataintomaskfortrainingDeepLearningmodels.SepsisCohortfromMIMICIII-SepsiscohortfromMIMICdataset.MIND:MicrosoftNewsDataset-alarge-scaledatasetfornewsrecommendationresearch.DatasetforAIforEarth-AIForEarthDataSetsisacollectionofdatasetsforAIresearch.ORBIT-acollectionofvideosofobjectsincleanandclutteredscenesrecordedbypeoplewhoareblind/low-visiononamobilephone.Debug&Benchmarktensorwatch-debugging,monitoringandvisualizationforpythonmachinelearninganddatascience.PYRIGHT-statictypecheckerforpython.BenchML-Pythonlibrarytobenchmarkpopularpre-builtcloudAIAPIs.debugpy-AnimplementationoftheDebugAdapterProtocolforPythonkineto-ACPU+GPUProfilinglibrarythatprovidesaccesstotimelinetracesandhardwareperformancecounterscontributedbyAzureAIPlatformteam.SuperBenchmark-abenchmarkinganddiagnosistoolforAIinfrastructure(software&hardware).PipelineGitHubActions-Automateallyoursoftwareworkflows,nowwithworld-classCI/CD.Build,test,anddeployyourcoderightfromGitHub.AzurePipelines-AutomateyourbuildsanddeploymentswithPipelinessoyouspendlesstimewiththenutsandboltsandmoretimebeingcreative.Dagli-frameworkfordefiningmachinelearningmodels,includingfeaturegenerationandtransformationsasDAG.PlatformAIforEarthAPIPlatform-distributedinfrastructuredesignedtoprovideasecure,scalable,andcustomizableAPIhosting,designedtohandletheneedsoflong-running/asynchronousmachinelearningmodelinference.
OpenPlatfomforAI(OpenPAI)-resourceschedulingandclustermanagementforAI.
OpenPAIRuntime-Runtimefordeeplearningworkload.OpenPAIProtocol-OpenPAIprotocolenablesjobsharingandportability.Openpaimarketplace-Amarketplacewhichstoresexamplesandjobtemplatesofopenpai.OpenPAIFrameworkController-builttoorchestrateallkindsofapplicationsonKubernetesbyasinglecontroller.HivedDScheduler-KubernetesSchedulerforDeepLearning.OpenPAIJSSDK-TheJavaScriptSDKisdesignedtofacilitatethedevelopersofOpenPAItoofferuserfriendlyexperience.OpenPAIVSCodeClient-ExtensiontoconnectOpenPAIclusters,submitAIjobs,simulatejobslocally,managefiles,andsoon.MLOS-DataSciencepoweredinfrastructureandmethodologytodemocratizeandautomatePerformanceEngineering.
PlatformforSituatedIntelligence-anopen-sourceframeworkformultimodal,integrativeAI.
Qlib-anAI-orientedquantitativeinvestmentplatform.
TaggingTagAnomaly-Anomalydetectionanalysisandlabelingtool,specificallyformultipletimeseries(onetimeseriespercategory)VoTT-VisualobjecttaggingtoolDevelopertoolVisualStudioCode-Codeeditorredefinedandoptimizedforbuildinganddebuggingmodernwebandcloudapplications.Gather-addsgatherfunctionalityinthePythonlanguagetotheJupyterExtension.Pylance-anextensionthatworksalongsidePythoninVisualStudioCodetoprovideperformantlanguagesupport.AzureMLSnippets-VSCodesnippetsforAzureMachineLearningSampleCodeMicrosoftAIforEarth
Sharedutilities-AcollectionofutilitiesforworkingwithAzureMachineLearning.acoustic-bird-detection-Tutorial:AccurateBioacousticSpeciesDetectionfromSmallNumbersofTrainingClipsUsingtheBiophonyModelbeluasound-Usingmachinelearningtodetectbelugawhalecallsinhydrophonerecordings.arcticseals-detect&classifyarcticsealsinaerialimagerytounderstandhowthey’readaptingtoachangingworld.AIDE:AnnotationInterfaceforData-drivenEcology-Detectingandclassifyingwildlifeinaerialimagery.CameraTrapTool-toolsfortrainingandrunningdetectorsandclassifiersforwildlifeimagescollectedfrommotion-triggeredcameras.LandcovermappingtheOrinoquíaregion-AtoolforpredictinglandcoverintheOrinoquiaregionofPeru.PlanetaryComputerHub-adevelopmentenvironmentthatmakesourdataandAPIsaccessiblethroughfamiliar,open-sourcetools,andallowsuserstoeasilyscaletheiranalyses.Poultrybarnmapping-codefordetectingpoultrybarnsfromhigh-resolutionaerialimageryandanaccompanyingdatasetofpredictedbarnsovertheUnitedStates.PlanetaryComputerSDKforPython-APythonSDKforthePlanetaryComputerHub.SpeciesClassification-Atoolforclassifyingspeciesinimages.NewsThreads-TheNewsThreadsprojectanalyzesnewsarticlestohelpfindsimilaritiesbetweennewsarticlesandtracenewsprovenanceacrosstime.
InnerEyeDeepLearning-MedicalImagingDeepLearninglibrarytotrainanddeploymodelsonAzureMachineLearningandAzureStack
DeepSeismic-DeepLearningforSeismicImagingandInterpretation
Multi-speciesbioacousticclassification-Multi-speciesbioacousticclassificationusingdeeplearningalgorithms.
NestleAcneAssessment-deeplearningmodelstoassesstheacneseveritylevelbasedonselfieimages.
VisualAnalogies-exploringtheconnectionsbetweenartworkswithdeep"VisualAnalogies".
ForecastingBestPractices-timeseriesforecastingbestpractices&examples.
ComputerVisionRecipes-bestpractices,codesamples,anddocumentationforComputerVision.
AzureMLDesignerSample-samplesofAzureMachineLearningdesigner.
DeepSpeedExamples-ExamplemodelsusingDeepSpeed
ATALEOFTHREECITIES-Analyzingthesafety(311)datasetpublishedbyAzureOpenDatasetsforChicago,BostonandNewYorkCityusingSparkR,SParkSQL,AzureDatabricks,visualizationusingggplot2andleaflet.
MicrosoftHealthIntelligenceMachineLearningToolbox-MicrosoftHealthIntelligenceAzureMachineLearningToolbox.
MicrosoftSolutionAccelerators
MLOpsSolutionAccelerator-thisrepositoryhelpsMLteamstoacceleratetheirmodeldeploymenttoproductionleveragingAzure.AnomalyDetectionSolutionAccelerator-implementAnomalyDetectionwhichisthetechniqueofidentifyingrareeventsorobservationswhichcanraisesuspicionsbybeingstatisticallydifferentfromtherestoftheobservations.ClassificationSolutionAccelerator-Thisisaclassificationsolutionacceleratortohelpyoubuildanddeployabinaryclassificationproject.CommunityAI@EdgeCommunity-findtheresourcesyouneedtocreatesolutionsusingintelligenceattheedgethroughcombinationsofhardware,machinelearning(ML),artificialintelligence(AI)andMicrosoftAzureservice.GlobalAICommunity-empowersdeveloperswhoarepassionateaboutAItoshareknowledgethrougheventsandmeetups.DeepLearningLab(Japan)-providesinformationondevelopmentcasesandthelatesttechnologytrendsrelatedtodeeplearning.WorkshopWorkshopWikiCompetition2020MachineLearningSecurityEvasionCompetition-codesamplesforthe2020MachineLearningSecurityEvasionCompetition.BookPRML-PatternRecognitionandMachineLearningbyChristopherBishopFoundationsofDataScience-BasicTheoryforDataScience.MasteringAzureMachineLearning:Performlarge-scaleend-to-endadvancedmachinelearninginthecloudwithMicrosoftAzureMachineLearningPracticalAutomatedMachineLearningonAzure:UsingAzureMachineLearningtoQuicklyBuildAISolutionsLearningMicrosoftLearn-LearningcontentsforMicrosofttechnologyDataScientist,AIEngineerDataScienceforManager-Generalization,Utility,andExperimentation:MLConceptsforMakingBetterBusinessDecisionsGithubLearningLab-learningcontentsforGithubtechnology.GettingstartedwithPython-SamplecodeforChannel9PythonforBeginnerscourse.PythonforBeginnersMorePythonforBeginnersEvenMorePythonforBeginnersGetstartedwithPyTorch-learnthefundamentalsofdeeplearningwithPyTorch.DevIntrotoDataScience-Inthis28-videoseries,youwilllearnimportantconceptsandtechnologiestobuildyourend-to-endmachinelearningapplicationsonAzure.MachineLearningforBeginners-ACurriculum-12weeks,24lessons,classicMachineLearningforallDataScienceforBeginners-ACurriculum-10Weeks,20Lessons,DataScienceforAll!MicrosoftCloudWorkshop-WideWorldImporters(WWI)deliversinnovativesolutionsformanufacturers.Blog,News&Webinarchannel9-AIShow-videosfordevelopersfrompeoplebuildingMicrosoftproductsandservices.MicrosoftOpenSourceBlog-blogaboutmicrosoftopensourcetechnology.MicrosoftResearchEvent,Conference&Webinars-Events,Conferences&WebinarsbyMicrosoftResearch.MicrosoftInnovationTechHub-AIprojectinMicrosoft.LinkedInEngineeringBlog-BlogbyLinkedInEngineeringTeamAISystem-systemforAIEducationResource(Chinese).AIEdu-AIeducationmaterialsforChinesestudents,teachersandITprofessionals(Chinese).---ContributingThisprojectwelcomescontributionsandsuggestions.MostcontributionsrequireyoutoagreetoaContributorLicenseAgreement(CLA)declaringthatyouhavetherightto,andactuallydo,grantustherightstouseyourcontribution.Fordetails,visithttps://cla.opensource.microsoft.com.
Whenyousubmitapullrequest,aCLAbotwillautomaticallydeterminewhetheryouneedtoprovideaCLAanddecoratethePRappropriately(e.g.,statuscheck,comment).Simplyfollowtheinstructionsprovidedbythebot.YouwillonlyneedtodothisonceacrossallreposusingourCLA.
ThisprojecthasadoptedtheMicrosoftOpenSourceCodeofConduct.FormoreinformationseetheCodeofConductFAQorcontactopencode@microsoft.comwithanyadditionalquestionsorcomments.
TrademarksThisprojectmaycontaintrademarksorlogosforprojects,products,orservices.AuthorizeduseofMicrosofttrademarksorlogosissubjecttoandmustfollowMicrosoft'sTrademark&BrandGuidelines.UseofMicrosofttrademarksorlogosinmodifiedversionsofthisprojectmustnotcauseconfusionorimplyMicrosoftsponsorship.Anyuseofthird-partytrademarksorlogosaresubjecttothosethird-party'spolicies.
评论