I-Artificial Intelligence (AI) ifumene inani elibalulekileyo lokuthandwa kwiminyaka yakutshanje.
Ukuba uyinjineli yesoftware, inzululwazi yekhompyuter, okanye umthandi wesayensi yedatha ngokubanzi, ngoko mhlawumbi unomdla kwizicelo ezimangalisayo zokusetyenzwa kwemifanekiso, ukuqondwa kwepateni kunye nobhaqo lwento olubonelelwa ngulo mmandla.
Eyona ndawo ibalulekileyo ye-AI okhe wayiva malunga nokuFunda okuNzulu. Lo mmandla ugxile kwii-algorithms ezinamandla (imiyalelo yeprogram yekhompyutha) imodeli emva kokusebenza kwengqondo yomntu eyaziwa ngokuba Inethiwekhi yeNeural.
Kweli nqaku, siza kudlula ingqikelelo yeNeural Networks kunye nendlela yokwakha, ukuqokelela, ukulungelelanisa kunye nokuvavanya ezi modeli zisebenzisa. Python.
Inethiwekhi yeNeural
IiNeural Networks, okanye ii-NNs, luluhlu lwe-algorithms oluyimodeli emva komsebenzi webhayoloji wobuchopho bomntu. IiNeural Networks ziquka iinodi, ezikwabizwa ngokuba zii-neurons.
Ingqokelela yeenodi ezithe nkqo zaziwa njenge maleya. Imodeli iqukethe igalelo elinye, imveliso enye, kunye nenani leeleya ezifihliweyo. Umaleko ngamnye uneenodi, ezikwabizwa ngokuba ziineurons, apho ubalo lwenzeka khona.
Kulo mzobo ulandelayo, izangqa zimele iindawo kwaye ingqokelela ethe nkqo yeenodi imele iileya. Kukho iileya ezintathu kule modeli.
Iinodi zomaleko omnye ziqhagamshelwe kumaleko olandelayo ngeentambo zothumelo njengoko kubonwa ngezantsi.
Uluhlu lwethu lwedatha luqulathe idatha ephawulweyo. Oku kuthetha ukuba iqumrhu ngalinye ledatha linikwe ixabiso legama elithile.
Ke kwidatha yokuhlelwa kwezilwanyana siya kuba nemifanekiso yeekati nezinja njengedatha yethu, kunye 'nekati' kunye 'nenja' njengeeleyibhile zethu.
Kubalulekile ukuqaphela ukuba iileyibhile kufuneka ziguqulelwe kumanani amanani ukuze imodeli yethu ivakale, ngoko ke iileyibhile zethu zezilwanyana ziba ngu-'0' kwikati kunye no-'1' kwinja. Zombini idatha kunye neelebhile zigqithiswa kwimodeli.
nokufunda
Idatha inikwa imodeli yequmrhu elinye ngexesha. Le datha yahlulahlulwe ibe ngamaqhekeza kwaye idluliselwe kwindawo nganye yemodeli. IiNodi zenza imisebenzi yezibalo kwezi ziqendwana.
Awudingi ukwazi imisebenzi yemathematika okanye izibalo zesi sifundo, kodwa kubalulekile ukuba nombono jikelele wendlela ezi modeli zisebenza ngayo. Emva koluhlu lwezibalo kuluhlu olunye, idatha idluliselwa kuluhlu olulandelayo njalo njalo.
Yakuba igqityiwe, imodeli yethu iqikelela ileyibhile yedatha kwinqanaba lemveliso (umzekelo, kwingxaki yokuhlelwa kwezilwanyana sifumana uqikelelo '0' lwekati).
Imodeli ke iqhubela phambili ngokuthelekisa eli xabiso liqikelelweyo kunye nelona xabiso lelebhile yokwenyani.
Ukuba amaxabiso ahambelana, imodeli yethu iya kuthatha igalelo elilandelayo kodwa ukuba amaxabiso ayahluka imodeli iya kubala umahluko phakathi kwamaxabiso omabini, ebizwa ngokuba yilahleko, kwaye uhlengahlengise izibalo zenodi ukuvelisa iilebhile ezihambelanayo kwixesha elizayo.
IiNkqubo zokuFunda nzulu
Ukwakha iiNeural Networks kwikhowudi, kufuneka singenise ngaphandle Izikhokelo zokuFunda nzulu eyaziwa ngokuba ngamathala eencwadi asebenzisa i-Integrated Development Environment (IDE).
Ezi zikhokelo ziyingqokelela yemisebenzi ebhalwe kwangaphambili eya kusinceda kwesi sifundo. Siza kusebenzisa isakhelo seKeras ukwakha imodeli yethu.
I-Keras yilayibrari yePython esebenzisa ukufunda okunzulu kunye ne-artificial intelligence backend ebizwa I-Tensorflow ukwenza iiNNs ngokohlobo lweemodeli ezilula ezilandelelanayo ngokulula.
I-Keras iza neemodeli zayo zangaphambili ezinokuthi zisetyenziswe nazo. Kwesi sifundo, siza kube sisenza eyethu imodeli sisebenzisa iiKeras.
Unokufunda ngakumbi malunga nesi sikhokelo seSifundo esiNzulu kwi Iwebhusayithi yeKeras.
Ukwakha iNeural Network (Isifundo)
Masiqhubele phambili ekwakheni iNeural Network sisebenzisa iPython.
Ingxelo yeNgxaki
IiNeural Networks luhlobo lwesisombululo kwiingxaki ezisekelwe kwi-AI. Kwesi sifundo siza kuba sihamba ngePima Indians Diabetes Data, ekhoyo Apha.
I-ICU UkuFunda ngoomatshini kuqulunqe le datha kwaye iqulethe irekhodi yonyango yezigulane zaseIndiya. Imodeli yethu kufuneka iqikelele ukuba isigulana sinesifo seswekile phakathi kweminyaka emi-5 okanye hayi.
Ilayisha iSeti yedatha
Iseti yethu yedatha yifayile enye ye-CSV ebizwa ngokuba yi-'diabetes.csv' enokusetyenziswa ngokulula kusetyenziswa i-Microsoft Excel.
Ngaphambi kokudala imodeli yethu, kufuneka singenise idatha yedatha yethu. Usebenzisa le khowudi ilandelayo ungenza oku:
ukungenisa iipandas njenge-pd
idatha = pd.read_csv('diabetes.csv')
x = data.drop("Isiphumo")
y = idatha[“Isiphumo”]
Apha sisebenzisa i Iipandas ilayibrari ukuze ikwazi ukukhohlisa idatha yethu yefayile ye-CSV, read_csv () ngumsebenzi owakhelwe-ngaphakathi wePandas osivumela ukuba sigcine amaxabiso kwifayile yethu kuguquko olubizwa ngokuba 'yidatha'.
I-variable x iqulethe isethi yedatha yethu ngaphandle kwesiphumo (ielebhile) idatha. Sifezekisa oku nge data.drop() umsebenzi osusa iilebhile ze x, ngelixa u-y equlathe kuphela isiphumo (ileyibhile) idatha.
Ukwakha iModeli yolandelelwano
Inyathelo loku-1: Ukuthathwa ngaphandle kwamaThala eencwadi
Okokuqala, kufuneka singenise iTensorFlow kunye neeKeras, kunye neeparamitha ezithile ezifunekayo kwimodeli yethu. Le khowudi ilandelayo isivumela ukuba senze oku:
ngenisa i-tensorflow njenge-tf
ukusuka kwi keras yokungenisa tensorflow
ukusuka tensorflow.keras.models ukungenisa ngokulandelelana
ukusuka kwi-tensorflow.keras.layers yokungenisa Ukuqalisa, iDense
ukusuka tensorflow.keras.optimizers ukungenisa Adam
ukusuka kwi-tensorflow.keras.metrics ngenisa ngokwecandelo_crossentropy
Kwimodeli yethu singenisa iileya ezixineneyo. Ezi zimaleko ezidityaniswe ngokupheleleyo; oko kukuthi, indawo nganye kumaleko idityaniswe ngokupheleleyo nenye indawo kumaleko olandelayo.
Sikwangenisa i ku sebenza umsebenzi ofunekayo kuluhlu lwedatha oluthunyelwe kwiindawo. Izilungisi nazo zithengwe kumazwe angaphandle ukuze kuncitshiswe ilahleko.
UAdam sisixhasi esaziwayo esenza ukuba imodeli yethu yohlaziyo lwezibalo iphumelele ngakumbi, kunye categorical_crossentropy eyi uhlobo lomsebenzi welahleko (ubala umahluko phakathi kwamaxabiso eileyibhile angawo kanye aqikelelweyo) esiya kuwasebenzisa.
Inyathelo lesi-2: Ukuyila iModeli yethu
Imodeli endiyenzayo inegalelo elinye (kunye neeyunithi ezili-16), enye efihliweyo (eneeyunithi ezingama-32) kunye nemveliso enye (eneeyunithi ezi-2) umaleko. La manani awalungiswanga kwaye aya kuxhomekeka ngokupheleleyo kwingxaki enikiweyo.
Ukuseta inani elichanekileyo leeyunithi kunye neeleya yinkqubo enokuthi iphuculwe ixesha elongezelelweyo ngokuziqhelanisa. Ukusebenza kuhambelana nohlobo lokulinganisa esiza kukwenza kwidatha yethu ngaphambi kokuba siyidlulise kwindawo.
I-Relu kunye neSoftmax yimisebenzi eyaziwayo yokuvula lo msebenzi.
imodeli = Ulandelelwano([
Ixinene(iiyunithi = 16, input_shape = (1,), isebenze = 'relu'),
Ixinene(iiyunithi = 32, ukusebenza = 'relu'),
Ixinene(iiyunithi = 2, ukusebenza = 'softmax')
])
Nantsi indlela isishwankathelo semodeli ekufuneka sijonge ngayo:
Ukuqeqesha uMzekelo
Imodeli yethu iya kuqeqeshwa ngamanyathelo amabini, elokuqala libe kukuqulunqa imodeli (ukubeka imodeli kunye) kwaye elilandelayo lifanele imodeli kwidathasethi enikiweyo.
Oku kunokwenziwa kusetyenziswa imodeli.compile() umsebenzi olandelwa yi model.fit() umsebenzi.
model.compile(optimizer = Adam(learning_rate = 0.0001), ilahleko = 'binary_crossentropy', metrics = ['ukuchaneka'])
imodeli.fit(x, y, epochs = 30, batch_size = 10)
Ukuchaza i-'acuracy' metric kusivumela ukuba sijonge ukuchaneka kwemodeli yethu ngexesha loqeqesho.
Kuba iileyibhile zethu zikuhlobo luka-1 kunye no-0, sizakusebenzisa ilahleko yokubini ukubala umahluko phakathi kweelebhile ezizizo neziqikelelweyo.
I-dataset iphinda ihlulwe ibe ziibhetshi ze-10 (i-batch_size) kwaye iya kudluliselwa kwimodeli amaxesha angama-30 (ii-epochs). Kwiseti yedatha enikiweyo, x iyakuba yidatha kwaye u-y abe ziileyibhile ezihambelana nedatha.
UVavanyo lweModeli usebenzisa iiPredictions
Ukuvavanya imodeli yethu, senza uqikelelo kwidatha yovavanyo sisebenzisa uqikelelo () umsebenzi.
uqikelelo = imodeli.predict(x)
Kwaye kunjalo!
Kuya kufuneka ngoku uqonde kakuhle Ukufunda nzulu isicelo, iNeural Networks, indlela abasebenza ngayo ngokubanzi kunye nendlela yokwakha, ukuqeqesha kunye nokuvavanya imodeli kwikhowudi yePython.
Ndiyathemba ukuba esi sifundo sikunika isiqalo sokwenza kunye nokusebenzisa iimodeli zakho zokuFunda ngokuNzulu.
Sazise kwizimvo ukuba inqaku beliluncedo.
Shiya iMpendulo