Makingdevelopersawesomeatmachinelearning
MakingDevelopersAwesomeatMachineLearning
Mybestadviceforgettingstartedinmachinelearningisbrokendownintoa5-stepprocess:
Formoreonthistop-downapproach,see:
ManyofmystudentshaveusedthisapproachtogoonanddowellinKagglecompetitionsandgetjobsasMachineLearningEngineersandDataScientists.
Thebenefitofmachinelearningarethepredictionsandthemodelsthatmakepredictions.
Tohaveskillatappliedmachinelearningmeansknowinghowtoconsistentlyandreliablydeliverhigh-qualitypredictionsonproblemafterproblem.Youneedtofollowasystematicprocess.
Belowisa5-stepprocessthatyoucanfollowtoconsistentlyachieveaboveaverageresultsonpredictivemodelingproblems:
Foragoodsummaryofthisprocess,seetheposts:
Probabilityisthemathematicsofquantifyingandharnessinguncertainty.Itisthebedrockofmanyfieldsofmathematics(likestatistics)andiscriticalforappliedmachinelearning.
Belowisthe3stepprocessthatyoucanusetogetup-to-speedwithprobabilityformachinelearning,fast.
StatisticalMethodsanimportantfoundationareaofmathematicsrequiredforachievingadeeperunderstandingofthebehaviorofmachinelearningalgorithms.
Belowisthe3stepprocessthatyoucanusetogetup-to-speedwithstatisticalmethodsformachinelearning,fast.
Linearalgebraisanimportantfoundationareaofmathematicsrequiredforachievingadeeperunderstandingofmachinelearningalgorithms.
Belowisthe3stepprocessthatyoucanusetogetup-to-speedwithlinearalgebraformachinelearning,fast.
Optimizationisthecoreofallmachinelearningalgorithms.Whenwetrainamachinelearningmodel,itisdoingoptimizationwiththegivendataset.
Youcangetfamiliarwithoptimizationformachinelearningin3steps,fast.
Calculusisthehiddendriverforthesuccessofmanymachinelearningalgorithms.Whenwetalkaboutthegradientdescentoptimizationpartofamachinelearningalgorithm,thegradientisfoundusingcalculus.
Youcangetfamiliarwithcalculusformachinelearningin3steps.
Pythonisthelinguafrancaofmachinelearningprojects.NotonlyalotofmachinelearninglibrariesareinPython,butalsoitiseffectivetohelpusfinishourmachinelearningprojectsquickandneatly.HavinggoodPythonprogrammingskillscanletyougetmoredoneinshortertime!
YoucangetfamiliarwithPythonformachinelearningin3steps.
Machinelearningisaboutmachinelearningalgorithms.
Youneedtoknowwhatalgorithmsareavailableforagivenproblem,howtheywork,andhowtogetthemostoutofthem.
Wekaisaplatformthatyoucanusetogetstartedinappliedmachinelearning.
Ithasagraphicaluserinterfacemeaningthatnoprogrammingisrequiredanditoffersasuiteofstateoftheartalgorithms.
Pythonisoneofthefastestgrowingplatformsforappliedmachinelearning.
Youcanusethesametoolslikepandasandscikit-learninthedevelopmentandoperationaldeploymentofyourmodel.
BelowarethestepsthatyoucanusetogetstartedwithPythonmachinelearning:
Risaplatformforstatisticalcomputingandisthemostpopularplatformamongprofessionaldatascientists.
Youcanlearnalotaboutmachinelearningalgorithmsbycodingthemfromscratch.
Learningviacodingisthepreferredlearningstyleformanydevelopersandengineers.
Timeseriesforecastingisanimportanttopicinbusinessapplications.
Manydatasetscontainatimecomponent,butthetopicoftimeseriesisrarelycoveredinmuchdepthfromamachinelearningperspective.
Theperformanceofyourpredictivemodelisonlyasgoodasthedatathatyouusetotrainit.
Assuchdatapreparationmaythemostimportantpartsofyourappliedmachinelearningproject.
Datascienceistousemathematics,statistics,andmachinelearningtechniquestoextractinsightsfromdata.
XGBoostisahighlyoptimizedimplementationofgradientboosteddecisiontrees.
Itispopularbecauseitisbeingusedbysomeofthebestdatascientistsintheworldtowinmachinelearningcompetitions.
Imbalancedclassificationreferstoclassificationtaskswheretherearemanymoreexamplesforoneclassthananotherclass.
Thesetypesofproblemsoftenrequiretheuseofspecializedperformancemetricsandlearningalgorithmsasthestandardmetricsandmethodsareunreliableorfailcompletely.
Deeplearningisafascinatingandpowerfulfield.
State-of-the-artresultsarecomingfromthefieldofdeeplearninganditisasub-fieldofmachinelearningthatcannotbeignored.
BesidesKeras,PyTorchisanotherlibraryfordeeplearningwithahugemarket-share.ItisimportanttoknowaboutPyTorchandbecomefamiliarwithitssyntax.
OpenCVisthemostpopularlibraryforimageprocessingbutitsmachinelearningmoduleislesswell-known.
IfyouarealreadyusingOpenCV,addingmachinelearningtoyourprojectshouldbeatnoadditionalcost.Youcanmakeuseoftheexperiencesyoulearnedinscikit-learnorKerastobringyourimageprocessingprojecttothenextlevel.
BelowarethestepsthatyoucanusetogetstartedwithmachinelearninginOpenCV:
Althoughitiseasytodefineandfitadeeplearningneuralnetworkmodel,itcanbechallengingtogetgoodperformanceonaspecificpredictivemodelingproblem.
Therearestandardtechniquesthatyoucanusetoimprovethelearning,reduceoverfitting,andmakebetterpredictionswithyourdeeplearningmodel.
Predictiveperformanceisthemostimportantconcernonmanyclassificationandregressionproblems.Ensemblelearningalgorithmscombinethepredictionsfrommultiplemodelsandaredesignedtoperformbetterthananycontributingensemblemember.
LongShort-TermMemory(LSTM)RecurrentNeuralNetworksaredesignedforsequencepredictionproblemsandareastate-of-the-artdeeplearningtechniqueforchallengingpredictionproblems.
Workingwithtextdataishardbecauseofthemessynatureofnaturallanguage.
Workingwithimagedataishardbecauseofthegulfbetweenrawpixelsandthemeaningintheimages.
Computervisionisnotsolved,buttogetstate-of-the-artresultsonchallengingcomputervisiontaskslikeobjectdetectionandfacerecognition,youneeddeeplearningmethods.
Deeplearningneuralnetworksareabletoautomaticallylearnarbitrarycomplexmappingsfrominputstooutputsandsupportmultipleinputsandoutputs.
MethodssuchasMLPs,CNNs,andLSTMsofferalotofpromisefortimeseriesforecasting.
GenerativeAdversarialNetworks,orGANsforshort,areanapproachtogenerativemodelingusingdeeplearningmethods,suchasconvolutionalneuralnetworks.
GANsareanexcitingandrapidlychangingfield,deliveringonthepromiseofgenerativemodelsintheirabilitytogeneraterealisticexamplesacrossarangeofproblemdomains,mostnotablyinimage-to-imagetranslationtasks.
Attentionmechanismsarethetechniquesinventedtomitigatetheissuewhererecurrentneuralnetworksfailedtoworkwellwithlongsequencesofinput.Welearnedthattheattentionmechanismitselfcanbeusedasabuildingblockofneuralnetworksandthereforewenowhavethetransformerarchitecture.
Attentionmechanismsandtransformermodelsareshowntodeliveramazingresults,especiallyinnaturallanguageprocessing.Thereareexamplesofusingtransformermodelsinonewayoranotherthatmakecomputersunderstandhumanlanguageandperformtaskssuchastranslationorsummarizingaparagraph,inhuman-likequality.
Ifyoustillhavequestionsandneedhelp,youhavesomeoptions: