I-Artificial Intelligence (AI) ithole inani elikhulu lokuduma eminyakeni yamuva nje.
Uma ungunjiniyela wesofthiwe, usosayensi wekhompiyutha, noma umthandi wesayensi yedatha ngokujwayelekile, khona-ke cishe uyathakaselwa izinhlelo zokusebenza ezimangalisayo zokucubungula izithombe, ukubonwa kwephethini nokutholwa kwento okunikezwa yilo mkhakha.
Indawo engaphansi ebaluleke kakhulu ye-AI okungenzeka ukuthi uzwe ngayo Ukufunda Okujulile. Lo mkhakha ugxile kuma-algorithms anamandla (iziqondiso zohlelo lwekhompyutha) olufanekiswe ukusebenza kobuchopho bomuntu okwaziwa ngokuthi Amanethiwekhi eNeural.
Kulesi sihloko, sizodlula umqondo weNeural Networks nendlela yokwakha, ukuhlanganisa, ukulinganisa nokuhlola lawa mamodeli kusetshenziswa. Python.
Amanethiwekhi eNeural
I-Neural Networks, noma ama-NNs, awuchungechunge lwama-algorithms amodelwe ngemuva komsebenzi webhayoloji wobuchopho bomuntu. I-Neural Networks iqukethe ama-node, abizwa nangokuthi ama-neurons.
Iqoqo lamanodi aqondile aziwa ngokuthi izendlalelo. Imodeli iqukethe okokufaka okukodwa, okuphumayo okukodwa, kanye nenani lezendlalelo ezifihliwe. Ungqimba ngalunye luqukethe ama-node, abizwa nangokuthi ama-neurons, lapho izibalo zenzeka khona.
Kumdwebo olandelayo, imibuthano imelela ama-node futhi iqoqo eliqondile lamanodi limelela izendlalelo. Kunezendlalelo ezintathu kule modeli.
Amanodi ongqimba olulodwa axhunywe kwesendlalelo esilandelayo ngemigqa yokudlulisela njengoba kubonakala ngezansi.
Idathasethi yethu iqukethe idatha enelebula. Lokhu kusho ukuthi ibhizinisi ledatha ngalinye linikezwe inani elithile legama.
Ngakho-ke kudathasethi yezigaba zezilwane sizoba nezithombe zamakati nezinja njengedatha yethu, kanye 'nekati' kanye 'nenja' njengamalebula ethu.
Kubalulekile ukuqaphela ukuthi amalebula adinga ukuguqulwa abe amanani ezinombolo ukuze imodeli yethu ibe nengqondo, ngakho amalebula ezilwane zethu aba ngu-'0' wekati futhi '1' wenja. Kokubili idatha namalebula kudluliselwa kumodeli.
Learning
Idatha inikezwa imodeli yebhizinisi elilodwa ngesikhathi. Le datha ihlukaniswa ibe izingcezu futhi idluliselwe ku-node ngayinye yemodeli. Amanodi enza imisebenzi yezibalo kulezi ziqephu.
Awudingi ukwazi imisebenzi yezibalo noma izibalo zalesi sifundo, kodwa kubalulekile ukuba nombono ojwayelekile wokuthi la mamodeli asebenza kanjani. Ngemva kochungechunge lwezibalo kungqimba olulodwa, idatha idluliselwa kungqimba olulandelayo njalonjalo.
Uma isiqediwe, imodeli yethu ibikezela ilebula yedatha kusendlalelo sokuphumayo (isibonelo, enkingeni yokuhlukanisa izilwane sithola isibikezelo esingu-'0' sekati).
Imodeli ibe isiqhubeka nokuqhathanisa leli nani elibikezelwe nenani lelebula langempela.
Uma amanani ehambisana, imodeli yethu izothatha okokufaka okulandelayo kodwa uma amanani ehluka imodeli izobala umehluko phakathi kwamanani womabili, abizwa ngokuthi ukulahlekelwa, futhi ilungise izibalo zenodi ukuze ikhiqize amalebula afanayo ngokuzayo.
Izinhlaka Zokufunda Ezijulile
Ukuze sakhe Amanethiwekhi E-Neural ngekhodi, sidinga ukungenisa Izinhlaka zokufunda ezijulile eyaziwa ngokuthi imitapo yolwazi esebenzisa Indawo yethu Yokuthuthukisa Edidiyelwe (IDE).
Lezi zinhlaka ziyiqoqo lemisebenzi ebhalwe ngaphambilini ezosisiza kulesi sifundo. Sizosebenzisa uhlaka lwe-Keras ukwakha imodeli yethu.
I-Keras iwumtapo wezincwadi wePython osebenzisa ukufunda okujulile kanye ne-backend yobuhlakani bokwenziwa ebizwa I-Tensorflow ukudala ama-NN ngendlela yamamodeli alandelanayo alula kalula.
I-Keras nayo iza namamodeli ayo asevele akhona angasetshenziswa nawo. Kulesi sifundo, sizobe sidala imodeli yethu sisebenzisa i-Keras.
Ungafunda kabanzi ngalolu hlaka Lwezifundo Ezijulile ku Iwebhusayithi yeKeras.
Ukwakha Inethiwekhi Yezinzwa (Okokufundisa)
Masiqhubekele ekwakheni iNeural Network sisebenzisa iPython.
Isitatimende Sezinkinga
I-Neural Networks iwuhlobo lwesixazululo sezinkinga ezisekelwe ku-AI. Kulesi sifundo sizobe sibheka Idatha ye-Pima Indians Diabetes, etholakalayo lapha.
ICU Ukufunda Ngomshini kuhlanganise le dathasethi futhi iqukethe irekhodi lezokwelapha leziguli zaseNdiya. Imodeli yethu kufanele ibikezele ukuthi isiguli sinokuqala kwesifo sikashukela phakathi neminyaka emi-5 noma cha.
Ilayisha Isethi Yedatha
Idathasethi yethu iyifayela elilodwa le-CSV elibizwa ngokuthi 'diabetes.csv' elingakhohliswa kalula kusetshenziswa i-Microsoft Excel.
Ngaphambi kokudala imodeli yethu, sidinga ukungenisa idathasethi yethu. Usebenzisa ikhodi elandelayo ungenza lokhu:
ngenisa ama-pandas njenge-pd
idatha = pd.read_csv('diabetes.csv')
x = data.drop(“Umphumela”)
y = idatha[“Umphumela”]
Lapha sisebenzisa i- AmaPandas umtapo wolwazi ukuze ukwazi ukukhohlisa idatha yefayela lethu le-CSV, i-read_csv() iwumsebenzi owakhelwe ngaphakathi we-Pandas osivumela ukuthi sigcine amanani kufayela lethu kokuguquguqukayo okubizwa ngokuthi 'idatha'.
I-variable x iqukethe idathasethi yethu ngaphandle kwedatha yomphumela (amalebula). Sifinyelela lokhu ngomsebenzi wedatha.drop() osusa amalebula okuthi x, kuyilapho u-y equkethe kuphela idatha yomphumela (ilebula).
Ukwakha Imodeli Elandelanayo
Isinyathelo 1: Ukungenisa Amalabhulali
Okokuqala, sidinga ukungenisa i-TensorFlow nama-Keras, kanye namapharamitha athile adingekayo kumodeli yethu. Ikhodi elandelayo isivumela ukuthi senze lokhu:
ngenisa i-tensorflow njenge-tf
kusuka kuma-keras wokungenisa we-tensorflow
kusuka ku-tensorflow.keras.models ingenisa Okulandelanayo
kusuka ku-tensorflow.keras.layers import Activation, Dense
kusuka ku-tensorflow.keras.optimizers ngenisa u-Adam
kusuka ku-tensorflow.keras.metrics ngenisa ngokwezigaba_crossentropy
Kumodeli yethu singenisa izendlalelo eziminyene. Lezi izingqimba ezixhumeke ngokugcwele; okungukuthi, i-node ngayinye kusendlalelo ixhunywe ngokugcwele nenye i-node kusendlalelo esilandelayo.
Siphinde singenise i ukusebenza umsebenzi odingekayo wokukala idatha othunyelwe kumanodi. Izithuthukisi nazo zingenisiwe ukuze kuncishiswe ukulahlekelwa.
U-Adam uyisilungisi esidumile esenza ukubalwa kwe-node yemodeli yethu ngempumelelo kakhudlwana, kanye categorical_crossentropy okuyi uhlobo lomsebenzi wokulahlekelwa (ibala umehluko phakathi kwamanani amalebula angempela naqagelwe) esizowasebenzisa.
Isinyathelo sesi-2: Ukuklama Imodeli Yethu
Imodeli engiyidalayo inokokufaka okukodwa (okunamayunithi angu-16), okufihliwe (okunamayunithi angu-32) kanye nosendlalelo esisodwa (namayunithi angu-2). Lezi zinombolo azilungisiwe futhi zizoncika ngokuphelele enkingeni enikeziwe.
Ukusetha inani elilungile lamayunithi kanye nezendlalelo kuyinqubo engathuthukiswa isikhathi esengeziwe ngokuzijwayeza. Ukwenza kusebenze kuhambisana nohlobo lokukala esizobe sikwenza kudatha yethu ngaphambi kokuyidlulisela endaweni.
I-Relu ne-Softmax yimisebenzi edumile yokwenza kusebenze lo msebenzi.
imodeli = Okulandelanayo([
Kuminyene(amayunithi = 16, input_shape = (1,), ukwenza kusebenze = 'relu'),
Kuminyene(amayunithi = 32, ukwenza kusebenze = 'relu'),
Kuminyene(amayunithi = 2, ukwenza kusebenze = 'softmax')
])
Nakhu ukuthi isifinyezo semodeli kufanele sibukeke kanjani:
Ukuqeqesha Imodeli
Imodeli yethu izoqeqeshwa ngezinyathelo ezimbili, esokuqala kube ukuhlanganisa imodeli (ukubeka imodeli ndawonye) bese okulandelayo kube ukufaka imodeli kudathasethi ethile.
Lokhu kungenziwa kusetshenziswa umsebenzi we-model.compile() olandelwa umsebenzi we-model.fit().
model.compile(optimizer = Adam(learning_rate = 0.0001), loss = 'binary_crossentropy', metrics = ['ukunemba'])
model.fit(x, y, epochs = 30, batch_size = 10)
Ukucacisa imethrikhi 'yokunemba' kusivumela ukuthi sibheke ukunemba kwemodeli yethu phakathi nokuqeqeshwa.
Njengoba amalebula ethu esesimweni sika-1 kanye no-0, sizobe sisebenzisa umsebenzi wokulahlekelwa kanambambili ukuze sibale umehluko phakathi kwamalebula angempela nabikezelwe.
Idathasethi nayo ihlukaniswa ibe amaqoqo angu-10 (i-batch_size) futhi izodluliswa kumodeli izikhathi ezingu-30 (izinkathi). Kudathasethi enikeziwe, u-x kuzoba idatha futhi u-y kuzoba amalebula ahambisana nedatha.
Imodeli Yokuhlola Ukusebenzisa Izibikezelo
Ukuze sihlole imodeli yethu, senza izibikezelo kudatha yokuhlola sisebenzisa umsebenzi wokubikezela().
izibikezelo = imodeli.predict(x)
Futhi yilokho!
Manje kufanele ube nokuqonda okuhle kwe Ukufunda Okujulile uhlelo lokusebenza, amaNeural Networks, asebenza kanjani ngokujwayelekile kanye nendlela yokwakha, ukuqeqesha nokuhlola imodeli ngekhodi yePython.
Ngethemba ukuthi lesi sifundo sikunikeza isiqalo sokudala nokusebenzisa amamodeli akho e-Deep Learning.
Sazise kumazwana uma lesi sihloko besiwusizo.
shiya impendulo