ThisrepositorycontainsvariousdemonstrationsoftheFluxmachinelearninglibrary.Anyofthesemayfreelybeusedasastartingpointforyourownmodels.
Themodelsarebroadlycategorisedintothefoldersvision(e.g.largeconvolutionalneuralnetworks(CNNs)),text(e.g.variousrecurrentneuralnetworks(RNNs)andnaturallanguageprocessing(NLP)models),games(ReinforcementLearning/RL).SeetheREADMEsofrespectivemodelsformoreinformation.
UsageThezoocomeswithitsownJuliaproject,whichliststhepackagesyouneedtorunthemodels.YoucanrunthemodelsbyopeningJuliaintheprojectfolderandrunning
usingPkg;Pkg.activate(".");Pkg.instantiate()toinstallallneededpackages.Thenyoucanrunthemodelcodewithinclude("<model-to-run>.jl")orbyrunningthemodelscriptline-by-line.
ModelsmayalsoberunwithNVIDIAGPUsupport,ifyouhaveaCUDAinstalled.Mostmodelswillhavethiscapabilitybydefault,pointedatbycallstogpuinthemodelcode.
GitpodOnlineIDEEachmodelcanbeusedinGitpod,justopentherepositorybygitpod
BasedonGitpod'spolicies,freeaccessislimited.AllofyourworkwillplaceintheGitpod'scloud.Itisn'tanofficiallymaintainedfeature.ContributingWewelcomecontributionsofnewmodels.Theyshouldbeinafolderwithaprojectandmanifestfile,topinallrelevantpackages,aswellasaREADMEtoexplainwhatthemodelisabout,howtorunit,andwhatresultsitachieves(ifapplicable).IfpossiblemodelsshouldnotdependdirectlyonGPUfunctionality,butideallyshouldbeCPU/GPUagnostic.Pleasekeepthecodeshort,cleanandself-explanatory,withaslittleboilerplatecodeaspossible.
ExamplesListingVisionMNISTSimplemulti-layerperceptronSimpleConvNet(LeNet)VariationalAuto-EncoderDeepConvolutionalGenerativeAdversarialNetworksConditionalDeepConvolutionalGenerativeAdversarialNetworksVGG16/19onCIFAR10TextCharRNNCharacter-levellanguagedetectionSeq2SeqphonemedetectiononCMUDictRecursivenetonIMDBsentimenttreebankOther&contributedmodelsDifferentialEquationsBitStringParityChallengeMLPonhousingdata(lowlevelAPI)FizzBuzzMeta-LearningLogisticRegressionIrisSpeechrecognitionTutorialsA60MinuteBlitzDataLoaderexamplewithimagedataTransferLearning
评论