Andwelcometopartfiveofourshenaniganswiththeneuralnetworkstutorialseriesinthispart, whatwe'regonnatalkaboutistheresultsfromtrainingthegeneralthiscallit a classificationsgenerator.
So, uh, let's talkaboutitsoyoucouldseewhat I'm runningrighthere.
Isactually, umtheLet's see a key.
I guessyoucan't seethisonthescreen.
Wepullusup a littlebitif I can.
Comeon.
Therewego.
So, basically, whatit's doingiseverytimeitgetstheclassification, right, Likethatwastheonepredicteditwas a one.
Thiswasseven.
I'm guessingthisis a 20 andsoon.
Sotheseareallincorrect.
Incorrect, incorrect.
Andtheneventually, itmightget a correctone.
Likehere.
Zerocorrect.
Sonot a seven.
Anyways, um, I'vebeenrunningthisfor a while.
About 316 tests.
3 17 now, andit's about 22% accurate.
Also, I'm stilltrainingthismodel.
Thisisactuallystillveryyoungmodel.
Somovingoverhere.
Thisisthemodelstillintraining.
Asyoucansee, ithasleveledoutquite a bit.
Butif I dosmooththisout, youshouldseethatit's stillslowlytaperingdownstill, so I'd liketoactuallyletitkeepgoing.
Infact, it's stillit's likelessthan 10% donewithatleastthe 50 e poxthatwewantedtogothrough.
So I kindofwantedtoletitcontinuegoing.
But I didn't wanttodothatifitwasjust a wasteoftime.
So I actuallydidnotknowifthiswasgoingtobesuccessful.
Myfirstattemptwasunsuccessfulcompletely.
Andthen I wonderedifwegaveit a littlemoreofanopportunitytogetthingsright.
Ifthatwouldhelpin, sureenough, ithas.
Sowhat I'd liketodonowisgoover.
Um, atleastthecodethat I usedtogetthis I don't thinkit's wise.
There's reallynopointforustowriteit.
So I justkindofwanttoshowyouthelogicthat I'm doingtodetectwhetherornotwegotsomethingright.
Butanyways, as I wassaying, um, sointhecaseoflike a zero, forexample, there's frequentlyzeros.
Um, so a lotoftimestojuststartspredictinganothernumberlikethisis a perfectexamplewhere I mean, itgottheCoghlan's anditstartedtosortofmakethatprediction.
Generally, itdoesn't reallygoodjobof, umbecauseintheonecasewe'redoing a general, we'redoingclassifications, andthentheotheroneisdoinggeneration, whichislikestuffit's seenbefore, s soitshouldbe.
Ah, it's a slightlyeasiertasktogetthenumbersrightthatithasgenerated.
It's a muchhardertasktogetthenumbersright.
That, umit's notgenerating, so I think I might I don't know, thisshouldgoprettyquick, but I mighthavetopausethisone.
Well, whilewewait, youalsojustcheckoutthislookslikeit's startingtotickuphere, buton a longenoughtimeline, itshouldjustcontinuetotoeatleastlevelout.
I don't thinkwe'regonnamakeanymorebigchanges, but, um, I wouldn't mindlettingititeratedfewmoretimesthroughitsowndataset, though, Stillgoin, andmaybe I'llpauseit.
I'm justgonnapausethisuntilit's done.
Socoredumped.
I can't decideit.
Isthatmyfault?
Isthatlikemehitting a keythatdoesthat?
Ordoesthatjustdothat?
Uh, see, Dowehaveanouttext?
We, atleasthow?
It's empty.
Damnit.
Okay, let's dojustdo.
25,000 5 25,000 Okay.
Pausing.
Okay.
It's definitelyme.
Hittingcontrolforthatismypaws, butinthatiscausing a coredump.
Interestingnoted.
Thankyou, sir.
Allright, I'm gonnapausenow, and I'm gonnarunthis.
Allright?
Sonowitisdone.
Let's seehowwe'vedoneourselves.
SeeoutdotTextsshouldbethisone, sohopefullythiswilllook a littlebetterthantheotherone.
Sothreecorrect.
Not a 51 shouldbe a 61 Shouldbe a nine 744 Good, good, good.
I'm notactuallysureifthat's a sevenor a one.
I'm gonnagiveittohim, though.
Ah, Sointerestinglyenough, thisoneactuallyisdoingworse, uh, atgeneratingthedeclassificationpartduringjust a straightupbig, um, bigsamplethanwhenitwasjust a onehotarray.
Itdidmuchbetter.
Theotherone, but I'm notreallyinterestedinindoingthat.
I'm notinterestedinclassificationsbeingrightinthegenerationwhereitjustgenerates a bunchmoreinterestedinhowaccurateitcanbeonthingsthat's neverseenstuffanyways, Butnotbad.
It's It's definitelybetterthanitisonnumbers.
It's neverseen.
But, um, beforeitwas, like, perfect.
Anyways, Um, yes.
Sothat's justthat's justgenerationthat's a littleeasierthangeneratingclassificationsonthevalidationsetjusttobeclear.