4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

Navigation

StartHereBlogBooksAboutContact
Search...
Wanthelpwithalgorithms?TaketheFREEMiniCourse.

Home

EmptyMenu

ReturntoContent

5TechniquesToUnderstandMachineLearning
AlgorithmsWithouttheBackgroundinMathematics
ByJasonBrownleeonAugust24,2015inMachineLearningAlgorithms

Wheredoestheoryfitintoatopdownapproachtostudyingmachinelearning?

Inthetraditionalapproachtoteachingmachinelearning,theorycomesfirstrequiringanextensivebackground
inmathematicstobeabletounderstandit.Inmyapproachtoteachingmachinelearning,Istartwithteaching
youhowtoworkproblemsendtoendanddeliverresults.

Sowheredoesthetheoryfit?

Inthispostyouwilldiscoverwhatwereallymeanwhenwetalkabouttheoryinmachinelearning.Hint:itsall
aboutthealgorithms.

Youwilldiscoverthatonceyougetskilledatworkingthroughproblemsanddeliveringresults,youwilldevelopa
compulsiontodivedeeperinordertobetterunderstandingandresults.Nobodywillbeabletoholdyouback.

Finally,youwilldiscover5techniquesthatyoucanusewhenyouarepracticingmachinelearningonstandard
datasetstoincrementallybuildupyourunderstandingofmachinelearningalgorithms.

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 1/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

Thankyouforsigningup!
Pleasecheckyouremailandclickthelink
providedtoconfirmyoursubscription.

HowToLearnMachineLearningwithouttheMath
PhotobyEdBrambley,somerightsreserved

LearnTheoryLast,NotFirst
Thewaymachinelearningistaughttodevelopersiscrap.

Itistaughtbottomup.Thisiscrapifyouareadeveloperwhoisprimarilyinterestedinusingmachinelearning
asatooltosolveproblemsratherthanbeingaresearcherinthefield.

Thetraditionalapproachrequiresthatyoulearnalloftheprerequisitemathematicslikelinearalgebra,probability
andstatisticsbeforelearningthetheoryofalgorithms.Youreluckyifyouevergonearaworkingimplementation
ofanalgorithmordiscusshowtoworkaproblemendtoendanddeliveraworking,reliableandaccurate
predictivemodel.

Iteachatopdownapproachtolearningmachinelearning.Inthisapproachwestartwith1)learninga
systematicprocessforworkingthroughproblemsendtoend,2)maptheprocessontobestofbreedmachine
learningtoolsandplatformsthen3)completetargetedpracticeontestdatasets.

GetYourStartinMachineLearning
YoucanlearnmoreaboutmyapproachtoteachingtopdownmachinelearninginthepostMachineLearning
forProgrammers:Leapfromdevelopertomachinelearningpractitioner.

Sowheredoestheoryfitintothisprocess?

Ifthemodelisflipped,thentheoryistaughtlater.Butwhattheoryarewetalkingaboutandhowexactlydoyou
learnthattheorywhenyouarepracticingontestdatasets?

GetyourFREEAlgorithmsMindMap
http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 2/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

Sampleofthehandymachinelearningalgorithmsmindmap.
Thankyouforsigningup!
I'vecreatedahandymindmapof60+algorithmsorganizedbytype.
Pleasecheckyouremailandclickthelink
providedtoconfirmyoursubscription.
Downloadit,printitanduseit.

DownloadForFree

Alsogetexclusiveaccesstothemachinelearningalgorithmsemailminicourse.

TheTheoryisReallyAllAboutAlgorithms
Thefieldofmachinelearningistheorydense.

Itsdensebecausethereisatraditiontodescribeandexplainconceptsmathematically.

Thisisusefulbecausemathematicaldescriptionscanbeveryconcise,cuttingdownontheambiguity.They
alsolendthemselvestoanalysisbyleveragingthetechniquesfromthecontextinwhichtheyaredescribed
(e.g.aprobabilisticunderstandingofaprocess).
GetYourStartinMachineLearning
Alotofthesetangentialmathematicaltechniquesareoftenbundledinwiththedescriptionofmachinelearning
algorithms.Forsomeonewhojustwantstobuildasuperficialunderstandingofamethodtobeabletoconfigure
andapplyit,thisfeelsoverwhelming.Frustratinglyso.

Itisfrustratingifyoudonothavethegroundingtobeabletoparseandunderstandthedescriptionofan
algorithm.Itsfrustratingbecausecomingfromafieldlikecomputerscience,algorithmsaredescribedallthe
time,butthedifferenceisthedescriptionsareintendedforfastcomprehension(e.g.fordeskchecking)and
implementation.

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 3/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

Weknowthatforexamplewhenlearningwhatahashtableisandhowtouseit,thatwealmostneverneedto
knowthespecificsofthehashingfunctioninourdaytoday.Butwealsoknowwhatahashingfunctionisand
wheretogotolearnmoreabouthashingfunctionspecificsandhowtowriteyourown.Whycantmachine
learningworklikethat?

Thebulkofthetheoryoneencountersinmachinelearningisrelatedtomachinelearningalgorithms.Ifyouask
anybeginneraboutwhytheyarefrustratedwiththetheory,youwilllearnthatitisinrelationtolearninghowto
understandoruseaspecificmachinelearningalgorithm.

Here,algorithmsismorebroadthanaprocessforcreatingapredictivemodel.Italsoreferstoalgorithmsfor
selectingfeatures,engineeringnewfeatures,transformingdataandestimatingtheaccuracyofamodelon
unseendata(e.g.crossvalidation).

So,learningtheorylast,reallymeanslearningaboutmachinelearningalgorithms.
Thankyouforsigningup!
ACompulsionToDiveIntoTheoryPleasecheckyouremailandclickthelink
providedtoconfirmyoursubscription.
Igenerallyadvisetargetedpracticeonwellknownmachinelearningdatasets.

Thisisbecausewellknownmachinelearningdataset,likethoseontheUCIMachineLearningRepositoryare
easytoworkwith.Theyaresmallsotheyfitintomemoryandcanbeprocessedonyourworkstation.Theyare
alsowellstudiedandunderstoodsoyouhaveabaselineforcomparison.

YoucanlearnmoreabouttargetedpracticeofmachinelearningdatasetsinthepostPracticeMachineLearning
withSmallInMemoryDatasetsfromtheUCIMachineLearningRepository.

Understandingmachinelearningalgorithmsfitsintothisprocess.Thereasonisinthepursuitofgettingresults
onstandardmachinelearningalgorithmsyouaregoingtorunintolimitations.Youaregoingtowanttoknow
howtogetmoreoutofagivenalgorithmortoknowmoreabouthowtobestconfigureit,orhowitactually
works.

Thisneedtoknowmoreandcuriositywilldriveyouintostudyingthetheoryofmachinelearningalgorithms.You
willbecompelledtopiecetogetheranunderstandofthealgorithmsinordertoachievebetterresults.

Weseethissameeffectinyoungdevelopersfromvariedbackgroundsthatendupeventuallystudyingthecode
ofopensourceprojects,textbooksandevenresearchpapersinordertohonetheircraft.Theneedtobeinga
bettermorecapableprogrammerdrivesthemtoit.

GetYourStartinMachineLearning
Ifyouarecuriousandmotivatedtosucceed,youcannotresiststudyingthetheory.

5TechniquesToUnderstandMachineLearningAlgorithms
Thetimewillcometodiveintomachinelearningalgorithmsaspartofyourtargetedpractice

Whenthattimecomes,thereareanumberoftechniquesandtemplatethatyoucanusetoshortcutthe
process.

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 4/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

Inthissectionyouwilldiscover5techniquesthatyoucanusetounderstandthetheoryofmachinelearning
algorithms,fast.

1)CreateListsofMachineLearningAlgorithms

Whenyouarejuststartingoutyoumayfeeloverwhelmedbythelargernumberofalgorithmsavailable.

Evenwhenspottestingalgorithms,youmaybeunsureofwhichalgorithmstoincludeinyourmix(hint,be
diverse).

Anexcellenttrickyoucanusewhenstartingoutistokeeptrackofthealgorithmsyoureadabout.Theselists
canbeassimpleasthenameofthealgorithm,andcanincreaseincomplexityasyouinterestandcuriosity
build.
Thankyouforsigningup!
Capturedetailsliketheproblemtypetowhichtheyaresuited(classificationorregression),relatedalgorithms,
andtaxonomicclass(decisiontree,kernel,etc.).Whenyouseethenameofanalgorithmthatisnewtoyou,
Pleasecheckyouremailandclickthelink
addittoyourlist.Whenyoustartanewproblem,trysomealgorithmsyouhaveneverusedbefore.Marka
providedtoconfirmyoursubscription.
checknexttoalgorithmsyouhaveusedbefore.Andsoon.

Controllingthenamesofalgorithmsinlistsgivesyoupower.Thisridiculouslysimpletacticcanhelpyougeton
topoftheoverwhelm.Examplesofwhereyoursimplealgorithmlistscansaveyoualotoftimeandfrustration
are:

Ideasofalgorithmstotryonnewanddifferentproblemtypes(timeseries,ratingsystems,etc.)
Algorithmsthatyoucaninvestigatetolearnmoreabouthowtoapply.
Getahandleonalgorithmtypesbycategory(trees,kernels,etc.).
Avoidtheproblemoffixatingonafavoritealgorithm.

Startbycreatinglistsofalgorithms,openaspreadsheetandgetstarted.

SeethepostTakeControlByCreatingTargetedListsofMachineLearningAlgorithmsformoreinformationon
thistactic.

2)ResearchMachineLearningAlgorithms

Whenyouwanttoknowmoreaboutamachinelearningalgorithmyouneedtoresearchit.

Themainreasonsyouwillbeinterestedtoresearchanalgorithmistolearnhowtoconfigureitandtolearnhow
GetYourStartinMachineLearning
itworks.

Researchisnotjustforacademics.Afewsimpletipscantakeyoualongwayingatheringinformationona
givenmachinelearningalgorithm.

Thekeyisdiversityofinformationsources.Thefollowingisashortlistofthetypesofsourcesyoucanconsult
forinformationonanalgorithmyouareresearching.

1.Authoritativesourcesliketextbooks,lecturenotes,slideandoverviewpapers.
2.Seminalsourceslikethepapersandarticlesinwhichthealgorithmwasfirstdescribed.
http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 5/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

3.Leadingedgesourcesthatdescribestateoftheartextensionsandexperimentsonthealgorithm.
4.Heuristicsourceslikethosethatcomeoutofmachinelearningcompetitions,postsonQ&Awebsitesand
conferencepapers.
5.Implementationsourcessuchasopensourcecodefortoolsandlibraries,blogpostsandtechnical
reports.

YoudonotneedtobeaPhDresearchernoramachinelearningalgorithmexpert.

Takeyourtimeandpickovermanysourcescollectingfactsonamachinelearningalgorithmyouaretryingto
figureout.Focusonthepracticaldetailsyoucanapplyorunderstandandleavetherest.

FormoreinformationonresearchingmachinelearningalgorithmsseethepostHowtoResearchaMachine
LearningAlgorithm.

Thankyouforsigningup!
3)CreateYourOwnAlgorithmDescriptions
Pleasecheckyouremailandclickthelink
Machinelearningalgorithmdescriptionsyouwilldiscoverinyourresearchwillbeincompleteandinconsistent.
providedtoconfirmyoursubscription.

Anapproachthatyoucanuseistoputtogetheryourownminialgorithmdescriptions.Thisisanothervery
simpleandverypowerfultactic.

Youcandesignastandardalgorithmdescriptiontemplatewithonlythosedetailsthatareusefultoyouingetting
themostfromalgorithms,likealgorithmusageheuristics,pseudocodelistings,parameterrangesandresource
lists.

Youcanthenusethesamealgorithmdescriptiontemplateacrossanumberofkeyalgorithmsandstarttobuild
upyourownlittlealgorithmencyclopediathatyoucanrefertoonfutureprojects.

Somequestionsyoumightliketouseinyourownalgorithmdescriptiontemplateinclude:

Whatarethestandardabbreviationsusedforthealgorithm?
Whatistheobjectiveorgoalforthealgorithm?
Whatisthepseudocodeorflowchartdescriptionofthealgorithm?
Whataretheheuristicsorrulesofthumbforusingthealgorithm?
Whatareusefulresourcesforlearningmoreaboutthealgorithm?

Youwillbesurprisedathowusefulandpracticalthesedescriptionscanbe.Forexample,Iusedthisapproach
towriteabookofnatureinspiredalgorithmdescriptionsthatIstillreferbacktoyearslater.
GetYourStartinMachineLearning
Formoreonhowtocreateeffectivealgorithmdescriptiontemplates,seethepostHowtoLearnaMachine
LearningAlgorithm.

Formoreinformationonmybookofalgorithmsdescribedusingastandardalgorithmdescriptiontemplate,see
CleverAlgorithms:NatureInspiredProgrammingRecipes.

4)InvestigateAlgorithmBehavior

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 6/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

Machinelearningalgorithmsarecomplexsystemsthataresometimesbestunderstoodbytheirbehaviorson
actualdatasets.

Bydesigningsmallexperimentsonmachinelearningalgorithmsusingsmalldatasetsyoucanlearnalotabout
howanalgorithmworks,itslimitationsandhowtoconfigureitinwaysthatmaytransfertoexceptionalresults
onotherproblems.

Asimpleprocedurethatyoucanusetoinvestigateamachinelearningalgorithmisasfollows:

1.Selectanalgorithmthatyouwouldliketoknowmoreabout(e.g.randomforests).
2.Identifyaquestionaboutthatalgorithmyouwouldlikeanswered(e.g.theeffectofthenumberoftrees).
3.Designanexperimenttofindananswertothatquestion(e.g.trydifferentnumbersoftreesonafew
binaryclassificationproblemsandcharttherelationshipwithclassificationaccuracy).
4.Executetheexperimentandwriteupyourresultssothatyoucanmakeuseoftheminthefuture.
Thankyouforsigningup!
5.Repeattheprocess.
Pleasecheckyouremailandclickthelink
Thisisoneofthetrulyexcitingaspectsofappliedmachinelearning,thatthroughyourownsimpleinvestigations
providedtoconfirmyoursubscription.
youcanachievesurprisingandstateoftheartresults.

Formoreinformationonhowtostudyalgorithmsfromtheirbehavior,seethepostHowToInvestigateMachine
LearningAlgorithmBehavior.

5)ImplementMachineLearningAlgorithms

Youcannotgetmoreintimatewithamachinelearningalgorithmthanbyimplementingit.

Inimplementingamachinelearningalgorithmfromscratchyouwillbeconfrontedwiththemyriadofmicro
decisionsthatgointoagivenimplementation.Youmaydecidetocoversomeupwithrulesofthumbofexpose
themallasparameterstotheuser.

Belowisarepeatableprocessthatyoucanusetoimplementmachinelearningalgorithmsfromscratch.

1.Selectaprogramminglanguage,onethatyouaremostfamiliarwithisprobablybest.
2.Selectanalgorithmtoimplement,startwithsomethingeasy(seebelowforalist).
3.Selectaproblemtotestyourimplementationonasyoudevelop,2Ddataisgoodforvisualizing(evenin
Excel).
4.Researchthealgorithmandleveragemanyanddiversesourcesofinformation(e.g.readtutorials,
papers,otherimplementations,andsoon).
GetYourStartinMachineLearning
5.Unittestthealgorithmtoconfirmyourunderstandingandvalidatetheimplementation.

Startsmallandbuildconfidence.

Forexample3algorithmsthatyouselectasyourfirstmachinelearningalgorithmimplementationfromscratch
are:

LinearRegressionusingGradientDescent
kNearestNeighbor(seemytutorialinPython)
NaiveBayes(seemytutorialinPython)
http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 7/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

Formoreinformationonhowtoimplementmachinelearningalgorithms,seethepostHowtoImplementa
MachineLearningAlgorithm.

Alsoseetheposts:

BenefitsofImplementingMachineLearningAlgorithmsFromScratch
DontStartwithOpenSourceCodeWhenImplementingMachineLearningAlgorithms

TheoryisNotJustFortheMathematicians
Machinelearningisnotjustforthemathematicalelite.Youcanlearnhowmachinelearningalgorithmsworkand
howtogetthemostfromthemwithoutdivingdeepintomultivariatestatistics.

Youdonotneedtobegoodatmath.
Thankyouforsigningup!
Aswesawinthetechniquessection,youcanstartwithalgorithmlistsandtransitiondeeperintoalgorithm
Pleasecheckyouremailandclickthelink
research,descriptionsandalgorithmbehavior. providedtoconfirmyoursubscription.

Youcangoveryfarwiththesemethodswithoutdivingmuchatallintothemath.

Youdonotneedtobeanacademicresearcher.

Researchisnotjustforacademics.Anyonecanreadbooksandpapersandcompiletheirownunderstandingof
atopiclikeaspecificmachinelearningalgorithm.

Yourbiggestbreakthroughswillcomewhenyoutakeonthepersonaofthescientistandstartexperimenting
onmachinelearningalgorithmsasthoughtheywerecomplexsystemsinneedofstudy.Youwilldiscoverall
kindsofinterestingquirksinbehaviorthatmaynotevenbedocumented.

TakeAction
Pickoneofthetechniqueslistedaboveandgetstarted.

Imeantoday,now.

Unsurewheretostart?

Heres5greatideasofwhereyoucouldstart:
GetYourStartinMachineLearning
1.Makealistof10machinealgorithmsforclassification(takealookatmytourofalgorithmstogetsome
ideas).
2.FindfivebooksthatgivedetaileddescriptionsofRandomForests.
3.CreateafiveslidepresentationonNaiveBayesusingyourownalgorithmdescriptiontemplate.
4.OpenWekaandseehowthekparameteraffectsaccuracyofknearestneighborontheirisflowers
dataset.
5.Implementlinearregressionusingstochasticgradientdescent.

Didyoutakeaction?Enjoythispost?Leaveacommentbelow.

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 8/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

FrustratedWithMachineLearningMath?
SeeHowAlgorithmsWorkinMinutes

...withjustarithmeticandsimpleexamples

DiscoverhowinmynewEbook:MasterMachineLearningAlgorithms

Itcoversexplanationsandexamplesof10topalgorithms,including:
LinearRegression,kNearestNeighbors,SupportVectorMachinesandmuchmore...

Finally,PullBacktheCurtainon
MachineLearningAlgorithms
Thankyouforsigningup!
SkiptheAcademics.JustResults.
Pleasecheckyouremailandclickthelink
Clicktolearnmore.
providedtoconfirmyoursubscription.

AboutJasonBrownlee

Dr.JasonBrownleeisahusband,proudfather,academicresearcher,author,professionaldeveloperanda
machinelearningpractitioner.Heisdedicatedtohelpingdevelopersgetstartedandgetgoodatappliedmachine
learning.Learnmore.
ViewallpostsbyJasonBrownlee
PracticeMachineLearningwithDatasetsfromtheUCIMachineLearningRepository
ChoosingMachineLearningAlgorithms:LessonsfromMicrosoftAzure

22Responsesto5TechniquesToUnderstandMachineLearning
AlgorithmsWithouttheBackgroundinMathematics

1.
TamimSyedSeptember3,2015at2:28am# GetYourStartinMachineLearning

FantasticPost.Enjoyedit.

Reply

JasonBrownleeSeptember21,2015at3:21pm#

ThanksTamim.

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 9/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

Reply

2.
AlexisSeptember5,2015at7:32am#

Greatpost.
Iwould,however,objectthattheoryisimportantifyouwanttobeabletointerpretresultsandhaveaclear
understandingoftheassumptionsmadeandtheirimplications.Henceasolidbackgroundinstatisticsand
probabilitiesisalwaysnecessary.
Justrelyingonalgorithmsasblackboxestosolvecertainkindsofproblemswithoutanyunderstandingof
theirprinciples,theirstrenghtsandweaknessesissomewhatdangerous.

Agreatbookforstarters:DataSciencefromScratch(OReilly).

Greetings, Thankyouforsigningup!
Alexis Pleasecheckyouremailandclickthelink
providedtoconfirmyoursubscription.
Reply

JasonBrownleeDecember23,2015at10:03am#

GoodpointandgoodbookrecommendationAlexis.

Reply

3.
BobNovember6,2015at6:52am#

Jason:

SpotonforwhatIneedtoknow,thanks!

Alexis:

Fromtheperspectiveofasolutionsorientedpractitioner,Iwouldneedmanylifetimestounderstandthefull
theoryofeachblackboxIuse.

Iinsteadhavetobesatisfiedwithamorefunctionalapproach,andlearnjustenoughaboutthetheoryand
behaviorofeachblackboxtouseitcorrectly.Detailedtheorycomesafterthat,timeandinterestpermitting.
GetYourStartinMachineLearning
Usually,itstime.

Cheers,

Bob

Reply

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 10/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

JasonBrownleeDecember23,2015at10:06am#

Icouldnotagreemorebob,thisisexactlythetopdownapproachthatIteach.

Reply

4.
JayJanuary13,2016at9:37am#

HiJason,
WhenIhearaboutpapersthatdescribealgorithmsthatbrokerecords(say,thisonethatwonfirstplacein
theILSVRCcompetitionin2015:http://arxiv.org/pdf/1512.03385v1.pdf),Igetexcitedandtrytoreadit.

ThenIrememberthatIcantreallyreadthemathnotation.Samestoryfor,say,thisStanfordMLtutorialon
softmaxregression:http://ufldl.stanford.edu/tutorial/supervised/SoftmaxRegression/
Thankyouforsigningup!
Nocluewhatsomeofthesesymbolsmean.Looksmaybelikesettheorynotationsometimes?
Pleasecheckyouremailandclickthelink
providedtoconfirmyoursubscription.
Anyways,myrealquestionis:howdoIgoaboutfiguringthenotationout?IsthereareferenceIcoulduse
tolearnmoreaboutthesymbolandwhen/whyitsused?

Reply

JasonBrownleeJanuary13,2016at8:16pm#

HiJay.

Hmmm.Iagreemate,itshard.

So,withsoftmax,youcouldgoandfindoneofthemanyimplementationsinpythonorwhatever,ripit
outandstudyitinisolation.Youcouldalsodumpitintoaspreadsheetandplotthefunctionsmapping
ofinputstooutputs.

Thegeneralpointofthepostistolookforalternativedescriptionsofalgorithmsthatareeasierto
grokandtinkerwiththemtounderstandthembetter.

Backtothepapers,especiallythelatterone.Ittakestime.Youcanstepthrougheachsymbol,they
arealldefinedandyesyouarerighttherearesomegeneralsetnotationfornumbersinthereal
numberset,etc.Perhapsstartwithasmallerexampleandpractice.Ifyoumustgroknotationthen
youmustpracticeworkingthroughalotofnotation.Thesoftmaxregressiondoeslookprettystraight
GetYourStartinMachineLearning
forwardbutdoesrequirealittlelinalgtogetthematrixnotation.Notaundergradcourseworth,but
perhapsafewvideosonkhanacademy.

Reply

5.
PozyJanuary17,2016at2:55pm#

Greatpost.Thisisreallybuiltupmyconfidentassoftwareengineeringbackground.

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 11/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

Reply

6.
JamesBurkillMarch23,2016at10:23pm#

Iwastaughtthemathandtheorypiecemealandimmediatelybeforeputtingitintopractice.

ThatshowAndrewNgteachesanditsalsohowitwascoveredinmypostgraduatecomputerscience
study.

AnytimeIhaveattemptedtorushaheadandusesomethinglikeWEKAbeforeproperlyunderstandingthe
algortihms,Ihavefoundmyselfwastingalotoftimetryingtocomprehendtheresultsandtunethe
parameters.

Advancedknowledgeofthemathmightbenecessaryinauniversitycourseoryouriskgettinglostinthe
Thankyouforsigningup!
lecturesbutthatconcerndoesntexistforonlinelecturesthatyoucanpauseorrewatch.
Pleasecheckyouremailandclickthelink
Inmyview,thebestwayfordeveloperstolearnmachinelearningistostartwithonlinecourses.The
providedtoconfirmyoursubscription.
momentaconceptappearsthatyoudontunderstand,simplypauseitandfindaKahnAcademyvideotofill
intheblanks.

Reply

7.
RajeshAugust11,2016at11:32pm#

Greatpost!Thanksalot.

Reply

JasonBrownleeAugust15,2016at12:28pm#

YourewelcomeRajesh.

Reply

8.
GaneshNovember16,2016at7:48pm#
GetYourStartinMachineLearning
Jasonthankyouforsuchnicepost.

Rightnowiaminthesamephaseofthinkingandgotthelightbyyourarticle.

Reply

JasonBrownleeNovember17,2016at9:53am#

ImgladtohearitGanesh.
http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 12/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

Reply

9.
HassanDecember31,2016at12:32am#

Iagreethattheteachinginthesesubjectsisshit.IhaveabackgroundinEngineeringPhysicsandhave
absolutelynoproblemwithgraduatestatistics,butstillthetheorytaughtinMachineLearning(Ididcourses
inNeuralNetworksandArtificialIntelligence)wasimpossibletofollowduringthelecturesandapaininthe
ass.Intheend,Ibarelypassed,eventhoughIoftengetpraiseformymathskillsfrommyfriendsat
university.

Absolutelyterrible.

Reply
Thankyouforsigningup!
Pleasecheckyouremailandclickthelink
JasonBrownleeDecember31,2016at7:03am#
providedtoconfirmyoursubscription.
ThanksHassan,itsgreattohaveyouhere.

ImheretohelpifIcan.

Reply

10.
VinodRalhFebruary17,2017at9:03am#

Greatarticle Thatswhywecodersstartwithahelloworldprogramtogetsomethingpositiveoutto
spuruson.

Reply

JasonBrownleeFebruary17,2017at10:07am#

IcouldnotagreemoreVinod!

Reply

11. GetYourStartinMachineLearning
TylerRenelleFebruary24,2017at7:57am#

Imadeapodcastepisodeonthemathyouneedformachinelearning,andtheresourcesforlearning
http://ocdevel.com/podcasts/machinelearning/8.ThereIagreeentirelywithyourtopdownlearning
approach.Buildfirst,theorylater!

Reply

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 13/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

JasonBrownleeFebruary24,2017at10:18am#

ThanksforsharingTyler.

Reply

12.
SachinShelarMarch24,2017at4:19pm#

HiJason,Thanksforarticle.Itisveryuseful.Fordeveloperandstarterinmachinelearning,thisisvery
usefulinformation.Ilikeyourtopdownapproach.

Reply

Thankyouforsigningup!
JasonBrownleeMarch25,2017at7:33am#
Pleasecheckyouremailandclickthelink
ThanksSachin. providedtoconfirmyoursubscription.

Reply

LeaveaReply

Comment

Name(required)

Email(willnotbepublished)(required)

Website

SubmitComment

GetYourStartinMachineLearning
WelcometoMachineLearningMastery

Hi,I'mDr.JasonBrownlee.
MygoalistomakepractitionerslikeYOUawesomeatappliedmachinelearning.
http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 14/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

ReadMore

UnderstandMachineLearningAlgorithms!
Sickofalltheadvancedmath?
Needstepbystepexplainationsfortopalgorithms?
Wantworkedexamplesinspreadsheets?

Thankyouforsigningup!
Pleasecheckyouremailandclickthelink
providedtoconfirmyoursubscription.

FinallyUnderstandMachineLearningAlgorithms!

Popular

TimeSeriesPredictionwithLSTMRecurrentNeuralNetworksinPythonwithKerasJuly21,2016

DevelopYourFirstNeuralNetworkinPythonWithKerasStepByStepMay24,2016

YourFirstMachineLearningProjectinPythonStepByStepJune10,2016

HowtoRunYourFirstClassifierinWekaFebruary17,2014

SequenceClassificationwithLSTMRecurrentNeuralNetworksinPythonwithKerasJuly26,2016
GetYourStartinMachineLearning

ATourofMachineLearningAlgorithmsNovember25,2013

TutorialToImplementkNearestNeighborsinPythonFromScratchSeptember12,2014

MultiClassClassificationTutorialwiththeKerasDeepLearningLibraryJune2,2016

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 15/16
4/13/2017 5TechniquesToUnderstandMachineLearningAlgorithmsWithouttheBackgroundinMathematicsMachineLearningMastery

RegressionTutorialwiththeKerasDeepLearningLibraryinPythonJune9,2016

HowToImplementNaiveBayesFromScratchinPythonDecember8,2014

2017MachineLearningMastery.AllRightsReserved.

Privacy|Contact|About

Thankyouforsigningup!
Pleasecheckyouremailandclickthelink
providedtoconfirmyoursubscription.

GetYourStartinMachineLearning

http://machinelearningmastery.com/techniquestounderstandmachinelearningalgorithmswithoutthebackgroundinmathematics/ 16/16