Sothat'llbetheSothezerowithelementisthe 33 by 33 data.
Andthenthefirstelementisthemove.
ThatwasassociateID.
Youknow?
Whatwhatmovedidtheagentmakebasedonthatdata?
Okay, sonow, um, dataequalsnpalluded.
Sonowwhatweneedtodoiswewantiteratedoverthatdata.
Sofour d indata, wewannatest x dotupend D zeroandthentest.
Why, uh, pen d one.
And, uh, that's allgood.
Andalsoprobably a pennparray.
Sothewisedon't actuallyneedtobe a numbpyre, Ray, But I'm prettysuretheECsdata I thinktensorflowisgonnagonnabarkatusififwedon't gowithHonoreforalltheECsdata.
Okay, once I stunmpdotsaveand, umand I hitmyinserttearsonthem.
Whathappened?
Whatishappening?
Therewego.
Uh, chest X, nonepie.
Thisislikethesametime.
Isthatprevioustutorialwhere I madeanegregiousmistake.
Now, inthecase, youknow, youmightalreadyhave a previousmodel, justlikebefore.
SoifloadprevemodelSoifthat's true, thenyoucansaymodelequals t f dotKeiraStottisn't models.
I thinkitstoppedmodelsyetdotmodelsdon't loadmodel, andthenpreviewmodelnameEasyisthatwedon't have a model, soit's else, andthenweneedtostartspecifyingourmodel.
Somodelequalssequentialwillgowith a sequentialtypeofmodel.
Um e I thinkwe'regonnastealfrommyselfduetodo, um, it's good.
Careusagain.
Andthenitwaswhatpart?
Threepartofmejustwantstojusttake a copyandpastethatcode.
Let's gotothebottom.
Thislooksgood, huh?
Wedon't wantthebinarycrossentropy.
Um, buttherestshouldbefine.
So I thinkwhat I'm gonnado, huh?
Uh, everythingdowntothefinaldenselayer, I think, iswhatwe'lltake.
Test X equalsNParraytest X becausewestillcan't get a shape.
It's a listofnumbpie.
Raiseupuntilwedothis.
Sowe'vegotthat.
Thenwegotthisrandomfunctioninthere.
That's fine.
Well, wecancleanthisstuffuplater.
Fornow, Uh, let's makeitwork.
I thinkwe're I thinkwe'realmostdonehere.
Um, sotheonlyotherthingthathasbeencominguprecentlypaddingequalssameis, forsomereasonandthenewerversionsoftensorflow I get a lotofthese, like, shapingissues, so I wouldrecommendaddingpattingsametoeveryconvolution a llinpoolinglayer.
Hopefully, thatiswhere I thinkthat's allwhereit's meanttobe, Um, becauseyouhavetospecifyandforsomereason, thatdefaultshavechanged.
UmThenwhatwe'regonnasayistrainingfilechunksthatisequaltothatchunksgeneratorthatwebuilthereandthenyoupass l n n.
Sotheentirelistisallofourfiles.
Sothat's trainingfilenamesafterthevalidation, or I'm sorry, whoalmostpissedupthere.
Okay, so, validationgamecount.
Onward.
Andthen, uh, training, trainingchunks, Eyesdone.
Okay.
I saidfudged, bytheway.
Okay.
Trainingfilechunks.
Um, probablyactually.
Justdefinethatuphere.
Thatshouldbefine.
Thio, putthatupthere.
Wedon't needtofindthateverypark.
Okay.
Great.
Fourindexandtrainingfiles, Uh, in a newandhumoreighttrainingfilechunksPrinceworkingondatachunkidee X plusoneoutof, um, thisoutofthisdividedbytrainingchunksize.
Sothatmightwindupgivingusquitelong.
Let's sayroundthat 22 decimalplaces.
Otherwise, thatmightbelike a superlongdecimalorsomething.
Andreally, whatweshouldsayismathdotseal, But I'm notgoingtodothat.
I justwantthegeneralgistofwhereweare.
Um, okay, sonowwhatwillsayisifloadifloadtrainedfilesfor E isgreaterthanzero.
Soifwe'vealreadydoneonewholeiPAQ, wehavethedata, sowecanactuallysay X equalsnpdotload.
Um, andthenwe'llcallthis, um, exdashsomethingdot n p y.
Andthiswouldbe I d x.
Asif I'm askingyouguys.
Um, butyeah, thatshouldThatshouldbewhatyouwant.
Copythat, paste.
Otherwise, whyequalsnpload?
Why?
Undertheidealscenario, thenelse, uh, X equalsemptylist.
Whyequalsemptylistfour f inetiquediatrainingfileyouNotrainingfiles.
Whatdowewanttodo?
We'regonnasay a dataequalsnpdotload f Really?
Thisisallprettysimilartowhatwehadbefore.
Um, also, westillhaven't balancedanything, huh?
Well, gettoit.
Dataequalsnpload F 40 indata.
Uh, let's sayexdothandnparray d zero.
Whydotupend D one?
Okay, SameKoda's weusedbefore.
Ifanything, probablymakethat a functionorsomething, but I'm notgonnaworryaboutit.
Okay?
Nowweneed, like, a balancingfunction.
Um, andthiscostsucks.
There's gottobe, like, a builtin a builtinbalancethatwecouldplausiblyuse.
Umsayunderscoresone.
Sowe'lldothisforeachoftheeachofthelist.
It's almostembarrassing.
Depressionpostthis.
Butlookatthat.
Wegottabalanceit.
There's gottabelike a builtinbalancingfunction, though.
Such a commontask.
Four x y inzip X y Basically, weneedtogettothepointwherealloftheselistsarethesamelength.
Ifyouhavequestions, comments, concerns, suggestions, a betterideaforthebalance.
Whatever.
Feelfreetow.
Leavethosebelow.
Otherwise, I willseeyouguysinthenextvideo, and I kindofwanttoletthisiPAQfinishdoesn't see, but I thinkit's hopefully I'm gonnabereallyshockedfor, like, beyond 30 percentages, Butokay, validation.
Andweactuallygot a bettervalidationaccuracythaninsample.
Interesting.
Okay, soyouguyslater, allright?
Itwouldn't be a centeggsvideowithout a mistakeinaneditattheveryend.
So, um, I wastryingtocompete, and I ranthefirstiPAQ, whichyouknow, really, wedid.
Westuckaroundtherangeof 24 to 28 Whichforvalidate.
Well, actually, I'm lookingatthewrongthing.
Let's look, E puckvalidationaccuracyGot a relative, uh, andwecouldseewherewewherewe'vestayedatsomewherebetween 24 a halfand 25% accurate, Which, inmyopinion, isactuallyprettygood, eh?
Sothosearetheresults.
But I caught.
I wastryingtodomultiplepox.
Um, and I stumbledon.
Hello?
Therewego.
Thissowhat's goingon?
Sothe 1st 1 I saidittorunforthree.
Fox, thisis a poxzeroandthenjustskipsiPAQone, too.
Itdoesn't doanything.
Theissueisourgenerator.
I moveditthinkingthatwasthesmartthingtodo, butno, thatwasthestupidthingtodo.