On I wantedtohighlightoneoftheorganizersareSomeoftheorganizersarealumnioftheprogram I.
T.
P where I teachandtheyarehosting a processcommunityinIndia.
Lookstomelikeyou'rethere, actuallyfourinIndiaandwhat's probablyaroundmidnightMidnightlikeIndia, like 1/2 simplytheyneeded 1/2 houroffthemidnightor 12 30 inIndia, and I'llshowyouthisonehereinBangaloreandforprimary 2nd 2019 organizedbythesewonderfulfolks, RushAli, who I believeisprobablyactuallyjust a couplefloorsdowninthisbuildingrightnow, I kindofhave a fantasyofgoingtooneoftheseprostitutecommunitiesinIndiaWillseeif I couldmakethatup a littledifficulttotravel.
That's thattimeofyearsofar.
Butwhoknows?
Allright, sothat's onething I wanttomention.
Alsowantedtogive a shoutouttocodingtrain.
ViewerEliza, I justdiscovered, has a podcastwithkindoflikethemostawesomenameever.
Unicorns, fartpixels.
I havetoadmit, I discovereditbecausesomebodymentionedthat I wasmentionedand I listenedtotheepisodewhereshetalksaboutcodingtrain, amongstotherthings.
Sothere's a littlebitof, likevanityhereandmeshowingyouthispodcast.
But s o I didlistentoEpisodefiveFirstYesterday, whichtalksaboutBobRossandmeandPatricioandcodingTrainandBookofShade, Er's andallsortsofotherthings.
But I and I'm planningtogobackandlistentotheotherepisode.
So I encourageyoutocheckoutthiswonderfulpodcastaboutwhatdevelopmentandcreativecoating.
It's great.
Unlikemyfrantic, freneticbehaviour, I feellikeELISAhas a verynice, soothing, comfortingqualityinhervoice.
Inthis, I wanttomentionthat I alsowantedtomention a zaframeforreferencethat, uh, thecodingtrainhas a website.
Andforeverycodingchallengethat I do, I have a pageontheWebsite, so I'm tryingtogetthisbetter, butyoucouldseeSothisis, bytheway, somethingthatisanopportunityforthecommunitytocontribute.
Ifanybodywhileyou'rewatchingrightnow, I have a littlechallengeforyoutodowhileyou'rewatching.
I wasgoingtodothisaspartoftheLifestream, but I thinkitmightbemoreefficienttohavesomebodydothiswhile I'm doingotherstuffandthengivemetheresults.
So I wanttoatsomepointtodaybefore I leaveTrain a wordtoVektrainingwork, newwordTyvekmodelwith a sourcetext, and I thoughtmaybeitwouldbeinterestingtoseewhathappensif I usedmynatureofcodebookasthesourcetext.
It's prettydecentamountoftext.
It's nothuge, actuallyhave a robustmodeland I knowthecamerajustwentoff.
I wouldneed a muchlargerbodyoftext.
Butthere's something I thinkwecouldworkwithrightnow.
Sowhat I What I needistotakealloftheseHTMLfiles, stripoutmaybeusingredGregorexpressionorsomekindoflikefancypythonthingthatyouknowaboutstripoutalltheHTMLtagsandleaveitwithjustthetext, therawtextofthebook, andthencancannonadeoldtoeonefilesobviouslywantonefilenatureofcodedot t x t.
Thatisjusttherawtextofthebook.
And I'llusethattotrain a wordTyvekmodeled.
Italsocouldbeusefulbecause I coulduseitformyown l s t m.
Examplethat I wouldhopetomake.
Andwhyyouneed, bytheway, it's okay.
I dothis.
I don't wanttobedoingthis.
I don't wanttobesayingacronymsandhavingpeoplewatchingmegoingwhat?
That's a kindofneuralnetworkthatcouldbeusedforsequencedataandforgeneratingtext.
Andsoworthavecis a thing.
Maybeyouneverheardabout.
I don't claimtobeanexpertonwordsavecatall, but I amgoingtodotalkaboutitandshow a wonderfultutorialbyAlisonParish, whichcovers a lotabouthowweirdeffectworksandthentrytoeplayaround, makesomeexamplesofworkdefecttoday.
So I Oh, bytheway, Sogethubwhenyouhave a projecthostedonGetHubandoneofyourpackagesthatyourprojectisdependentonhas a securityvulnerabilityandwe'llgiveyouthisalert.
And I wasableto I hadthisalertforallthesenodepackages, and I know I dolikeNPMsomethingauditandyoufixthatupgradestuffanditworks.
Abletodothat, I couldnotfigureouthowtofixthegemfiledotlockvulnerabilities.
Maybethat I wouldhavetotallyacceptablerequestonthat.
Ifthat's a thing, I thinkthat's thingthatsomebodycoulddowithoutmebeinganadministrator.
I don't knowenoughaboutRubystuff, andfrankly, I don't sayOhno, yeahOh, thisisForgethowpages.
That's whyit's inthere.
So, anyway, pleasedon't fixit.
O'Neill's websiteDon't fixit.
Neil's WebinthechatStop.
Stop A littlespool.
Stop.
TraintrainfullStop.
Probablyemergencybrake.
I don't videowebsitesdon't fixit.
I wasacceptedbythetime I saidthattherewouldbemoreinformationaboutwhynottoo.
Butthat's thatfornow.
Umallright.
Um, let's seehere.
I thinkthat's thatwasmyintroductiontothefactNow I havetostartdoingsomecontact.
I think I said I wasgonnadoworktoBeck, and I think I shoulddoworkdefecttoday.
But, youknow, I made a mistakebecause I'm goingtotalkaboutworkavecinmy n Y u Classtwo a weekand 1/2 fromnow.
Andtypicallyyoumightthinkthat I havenoideawhat I'm doingand I neverpracticed.
I'm justwingingit, whichismostlytrue.
But I dohave a policy, especiallywhen I'm trying a newtopicandtrytoactuallylookatitinmycoursefirst, where I havestudentsthat I canengagewithandtalkaboutandgetfeedbackfromandanswerquestionsandgetgettipsonhim.
Andthen I trytomake a videoaboutit.
So I reallyshouldprobablywait a coupleweeks, But I alreadyputitisthetitleofthisand I gotnothingelse.
I'vegotnothingelseforit.
Besides, I coulddolike a flockingcodingchallengesmeetingto d'oh.
Uhoh.
Actually, I wasthinkingofsubmitting a flockassomethingaboutflockingtoTomScott's guestthing.
I mean, maybemaybeTomScott.
Butanyway, I shouldmentionthisbecausethisis a coolIfyou're a youtuberorwantandwantThiosubmit A, uh, TomScottwas a wonderfulchannel.
I haveguestsforonemonthprior.
Doesn't makesenseforme.
It's beenanything.
I putthewrongpersonforthis, butbutand I'm I'llmentionitnow.
Allright?
Oh, no, no.
Thisisthisisvisible.
I donotwanttobuzz.
I did.
I havemyownspecialbyyou.
I don't wanttotalkaboutthis, thiscopoverhereanymore.
Don't evenmentionit.
Thankyou.
Sohe'llKateri.
I guessthesuperchatisbackonhisoffice.
Backon.
Okay.
Allright.
Neil's Webrightsbreakingnews, everyone, someonemakeme a breakingnews.
Like a customcodingtrainbreakingnewssoundeffectitshouldhave, like, didgiveyouthedignity, dignityandthen, like, youcouldusemyvoiceontheline.
A newBobMaroon I willbesendingoutifyou're a newmember, checkthecommunitytabforthepostthatyoucanviewthatHave a linkto a GoogleformThio, Getintheslackchannel.
Okay.
Neil's Webrights.
I haven't updatedversionlocally, butit's notpullrequestedyet.
I actuallyshouldkeepthisaswell, becausethisisanotherreferencethat I usedtocreatethattocreatethisthematerialsthatareinmyheadthat I'm gonnaworkontoday, andthenit's oneoftheplace.
What I wasgonnasaywas, I thinkthattodaymightactuallyjustbelargelyexploratoryandjusttry a bunchofthingsandnotactuallytakethecontentandmakingeditorialvideosoutofthem.
Buttherewasnooneelsetodoit, Sowhynot?
Uh, whynot?
JuststartwithAlison's tutorial, Okay.
Oh, gloveisnotwordtoBeck.
Globalvectorsforwordrepresentation.
Yes.
Sorry, butitdoeshave.
Sothisis a greatpoint.
ThatisthatSimonismaking.
Simonwritesinthechat.
Uh, itis, uhthankyou, Matt.
Youforremindingmeaboutthe l S t m interest.
I needtorecordgloveisperhapsusing a differentalgorithmwhereTyvekis a particularalgorithmtogeneratewordembeddingczarwordvectors.
WordtoBeckAward, Tyvekmodel.
Sotospeakonglove, I guess, uses a differentparticularalgorithm.
Butif I'm correct, theendresultisthesame.
It's a wordthatispairedwith a 300 dimensionalvectorword.
Embeddingisalsoanotherwayofsayingit.
Somaybe I letmeputtheglovestuffasidesothatthatdoesn't become a confusingthing, butletmejustbegin.
Ah, vectorthisweekendbasicallysomehowcomeupwiththissortof a numericmathematicalessenceofthisword, assomearrayofnumberslike 0.7 and 1.2 innegative 0.0 point 345 etcetera, etcetera.
Andthere's gonnabesomeamountofnumbersinhere.
Thisseemslike a crazything.
Whywould I everwanttohave a wordassociatedwithanarrayofnumbers?
Well, oneofthethingsthatonecoulddowith a raiseofnumbersismath, linear, algebra, multiplying, subtracting, averaging, adding.
Soweknowwecandothatwith a raiseofnumbers, andthisisthekindofthingthathappensinlotsofmyothertutorialswithprogramming, graphicsandpixelprocessingandmachinelearning.
Could I getthosesortofmathematicalessenceofthesewords?
Addthemtogetherandget a newword.
Well, thetheory, theprompt, theideaherethattheargumentthat I that I ammakingtousethatwordtoBeckis a mechanismbywhichyoucoulddostufflikethisrightthereinyourcode.
If I couldquantifytheword.
Appleis a seriesofnumbersand I couldquantifythewordpurpleas a seriesofnumbers.
Thencouldn't Ijustaddallthosenumberstogether?
I wouldget a newseriesofnumbers, andthen I mightlookandfindwhichwordorhas a setofnumbersthatismostclosetothese.
Setthissetofnumbers.
Howcould I findthesimilarity?
I couldcalculate a similarityscorebetweenanytwosetsofnumbers.
I couldfindthewordthathasthemostsimilartothisPlusthis.
I mean, I didif I didmentioneasierthanonementiononementionsweirdsometimes, butAndwecouldseeherethattherelationshipbetweenTarango, DjulaandHamsterisjustlikechickenand, uh, kitten, maybeclosetopuppy.
Right?
Andsothisatthepausefor a secondtalkaboutimprovising.
I havenoideaofthisismakinganysense.
Uh, allright.
Whilespongemenalreadymademynatureofcodedot t x t Thankyou.
Butinordertodothat, it's usefultolookat a singleexamplethattieswordstonumbersin a lowdimensionalspacethatwecaneithervisualizeorifwe're, like, putintoourbrainsandsokindofdescribesyou.
GethubslashShipmanslash p fivedashworkedavecand I'm lookingatthewordcolorvectors.
Butthenwhat I wanttogetisthecolordata.
Oh, comeon.
Thisissocrazy.
Howitwon't, um, controlopening a newtab.
Okay, sowe'llstartwiththis, and I'm going.
I'm notgonnadothisinthe p fiveeditor, since I thinkisrequires a bitmoresothat I'm gonnagotothedesktopanddowordtiuvek, uh, looks, uhitdoesn't autofillifitdoesn't exist.
But I'm sousedtolikegmailnow, justlikefinishes.
Mightsaidit's wheremeandlike, youknow, thedirectorynamethat I wanttomake.
Justtab.
Finishitforme.
Umallright.
Nowwhatam I doing?
I'm gonnagointothatdirectory.
Oh, andthen I wanttodo p five g dash B.
Andwhat's so?
I wouldsay, uh, color, That's a colorvectors.
Oh, I didn't I didn't solvethisin p m Dash g er 10 p.m. install p five.
ManagerDusty.
So I use a toolcalled P fivemanager, whichwillspinupallthefiles I needfor a P fiveprojectreallyquickly.
Andofcourse, itdidn't work.
Oh, what's funnyis I mean, alltheseworkflowvideossettingupallofmysettings, and I made a like a freshuseronthiscomputer.
Wentbacktothisuser, andnoneofthestuffissetup.
Howdo I fixthatthingagain?
It's fixed.
Ofcourse.
I wentoverthisinmyvideo, but I can't rememberitfixedNPMpermissionsonThisisthewaythat I liketodoit.
Um, I liketomake a directorycalledNPMGlobalinthissetthatisthepathon.
Andthen, uhexportthatand a source.
But I'm using Z s h.
Sosource z r s areZs R c h c r a z c R s Uhhuh.
Z s h r c.
That's whatitis.
Zs, HRC.
Herewego.
Andnow I shouldfeelinstall.
Thatapologyisthat I'm notexplainingwhat I'm doing, but I doexplainthisinmywork.
Look, getbacktotheshothere.
Uh, who's somebodyalreadymade?
A breakingnews.
Soundeffect.
What's allthis?
Uh, okay.
Whips.
I havemymutethatsomebodymade a breakingnewssoundeffect, which I amthrowingcautionto.
Andnowinthisvideo, I'm goingtodo a directportofwithsomewhere.
It's somewherehere.
I wenttoofar.
I wentAh, therewego.
languagewithvectorscolor.
So a lotofpeoplewhoarewatchingthisliveatthemomentwe'reaskingabout, I don't getit.
I sortofgetthat.
That's likewordwordinVictor's a wordinbeddings, butlikewhat?
Thatmadenosense.
But I wouldjustwentoffon a tangenthere.
Sopeoplewereaskinginthepreviousvideo, I don't getit.
Like I thoughtthatwordvectorsarejustkindoflikeaboutthemeaningofthewordorthemeaningofthewordinitscontextorsimilaritybetweenwords.
And I don't wantwhatiscutenousthisexamplebefore I lookedatAlisonParisexamplewith, like, acutenessinsides.
Whatdoesthathavetodo?
Anything.
Soletmebeclear.
Mygoalistogettothepointwherewe'retalkingaboutthismoregeneralizedideathatends, thatwegetword M beddingswordspairedwithnumbersfrom a machinelearningprocessthatgeneratesthesenumbersaccordingtothewaywordsappearin a verylargebodyoftext.
Butwe'renotdoingthatyet.
I justwannalookatthisideaof a vectorspaceandsomeofthemathassociatewithitandhowthatcantransfertotext.
Yeah, um I guess I willjustsortofexplainthatagain.
I don't evenrememberwhat I said, soallright.
Okay.
Allright.
Um, emquedemo 88 asked.
Whydoyoualwayshavetoclickthecameraonandoff?
I willreferyoutothetelevisionprogram.
Lostfromthe 19 fromtheswindles.
LostDunn's inthenineties.
Noteearlytwothousands.
Whatever.
I'llreferyoutothatprogramwhere a buttonhadtobepressed, lestandthatbutitwasnotpressed.
Theworldwouldceasetoexist.
And I willmakethecasethatif I don't resetthatcameraevery 30 minutes, kaput.
Okay, I reallylostmymomentum.
Sowhat I wanttodoisdo a directportofthisexamplefromAlisonParishaboutlanguagewithvectorscolors.
NowpeoplewereaskingInthepreviousvideo, I kindoflookedatAlison's exampleoflike a wordlikeananimalcouldbepairedwithdatalikeitssizeanditsacuteness.
Whatdoesthathavetodowithwordvetted?
Embedded, weren't you?
Oh, thiscamerawentoff.
Wait, thiscameradidnotgooff.
It's cableislooseorit's plugisout.
It's likeMerrygo.
It's a kids.
It's Ah, connectionissue.
Jeez, howmanytimes, allright?
I don't evenknowwhere I waasSoSowhat I wanttodoisgothroughthisexactexamplewhichisinMrTurtleandPythonandpouredintoJavascriptonwhatthisexampleisgoingtodo.
It's goingtogiveustheopportunitytolookattheideaof a wordpairedwith a particular.
Isthiswhere I was?
Oh, and I haven't evenrecordingthistwodisceverything.
What a mess.
I coulddothis.
I cangetpastthispartwhere I'm stuck.
Okay.
Thankyou, Siddharth.
I don't understandwhyanyonewouldpossiblygiveme a superchatrightnowwithwhat's goingonintheprevious.
I think I'm gonnatrytojustgofromwhere I walkedoverthere.
Mata, Thankyouforallofeverythingthatyou D'oh!
Um, Sowhat I wannadois a directportofthisexamplefromAlisonParishthatlooksatlanguagewithvectorscolors.
Now, letmeletmebeclearaboutsomethingfor a second.
Inthepreviousvideo, I talked a littlebitaboutthisexamplewhichwaslookingatanimalsin a twodimensionalspacewiththe X axisisliketheirsizeandthe Y axisofthesizeand x axisacuteness.
AndisthatwordTyvek?
Well, notexactly.
Thistheseairword M beddingsmeaningthereis a wordembeddedwithsomenumericdata.
That's thebestwaytodescribeit.
Butit's notexactlywhatyouwillfindifyoulookupthetermwordTyvek, wheredefectis a particularalgorithm, a machinelearningalgorithmtotake a largebodyoftextandproducedthesewordinbeddingshighdimensionalraysofnumbersbasedonhowthewordsappearincontext.
Inthattext, I'm gonnagettothat a littlebitlatertosort.