Artificial Intelligence (AI) yakawana huwandu hwakakosha hwekuzivikanwa mumakore achangopfuura.
Kana iwe uri injinjini yesoftware, sainzi wekombuta, kana data sainzi anofarira kazhinji, saka iwe unogona kunge uchifadzwa neanoshamisa maapplication ekugadzirisa mifananidzo, kucherechedzwa kwepateni uye kuona chinhu chinopihwa nendima iyi.
Iyo inonyanya kukosha subfield yeAI yaungangonzwa nezvayo ndeye Kudzidza Kwakadzika. Iyi ndima inotarisa pane ane simba algorithms (mirayiridzo yepurogiramu yemakomputa) inoteedzerwa mushure mekushanda kwehuropi hwemunhu hunozivikanwa se Neural Networks.
Muchinyorwa chino, tichaenda pamusoro peiyo pfungwa yeNeural Networks uye maitiro ekuvaka, kuunganidza, kukwana uye kuongorora aya mamodheru uchishandisa. Python.
Neural Networks
Neural Networks, kana NNs, nhevedzano yealgorithms yakateedzerwa mushure mekuita kwebiological chiitiko chehuropi hwemunhu. Neural Networks ine nodes, inonziwo neurons.
Muunganidzwa wemanodhi akatwasuka anozivikanwa sema layer. Iyo modhi ine yekupinza imwe chete, imwe inobuda, uye akati wandei akavigwa maseru. Chikamu chega chega chine node, inonziwo neurons, uko kunoitika maverengero.
Mudhayagiramu rinotevera, madenderedzwa anomiririra zvipfundo uye muunganidzwa wemapfundo wakatwasuka unomiririra mitsara. Pane matatu akaturikidzana mune iyi modhi.
Manodhi eimwe layer akabatana kune inotevera layer kuburikidza nemitsara yekutapurirana sezvinoonekwa pazasi.
Dataset yedu ine data rakanyorwa. Izvi zvinoreva kuti chimwe nechimwe che data chakapihwa imwe kukosha kwezita.
Saka kune yemhando yedataset yemhuka tichava nemifananidzo yekatsi nembwa sedata redu, ine 'katsi' uye 'imbwa' semavara edu.
Zvakakosha kuziva kuti mavara anofanirwa kushandurwa kuita manhamba kuti modhi yedu iite zvine musoro, saka mavara edu emhuka anova '0' pakitsi uye '1' pambwa. Zvose data uye zvinyorwa zvinopfuudzwa kuburikidza nemuenzaniso.
Learning
Iyo data inopihwa kune modhi chinhu chimwe panguva. Iyi data yakaputswa kuita chunks uye yakapfuura nepakati imwe neimwe node yemuenzaniso. Node dzinoita mashandiro emasvomhu pane aya chunks.
Iwe haufanirwe kuziva masvomhu mabasa kana kuverenga kwechidzidzo ichi, asi zvakakosha kuve neruzivo rwakakwana rwekuti aya mamodheru anoshanda sei. Mushure mehuwandu hwekuverenga mune imwe nhanho, data inopfuudzwa pane inotevera dhizaini zvichingodaro.
Kana yapera, modhi yedu inofanotaura iyo data label pane inobuda layer (semuenzaniso, mudambudziko remhando yemhuka tinowana fungidziro '0' yekatsi).
Iyo modhi inozoenderera ichienzanisa kukosha kwakafanotaurwa neicho cheiyo chaiyo label kukosha.
Kana kukosha kwacho kuchienderana, modhi yedu inotora iyo inotevera yekuisa asi kana hunhu hwakasiyana iyo modhi ichaverenga mutsauko pakati pezvakakosha zvese, zvinonzi kurasikirwa, uye gadzirisa maverengero enode kuti abudise mavara anoenderana nguva inotevera.
Dzidzo Yakadzama Frameworks
Kuvaka Neural Networks mukodhi, isu tinofanirwa kuunza kunze Zvirongwa zveKudzidza Zvakadzika anozivikanwa semaraibhurari anoshandisa yedu Yakabatanidzwa Yekuvandudza Nzvimbo (IDE).
Mafuremu aya muunganidzwa wemabasa akanyorwa kare anozotibatsira muchidzidzo ichi. Tichange tichishandisa Keras chimiro kuvaka modhi yedu.
Keras iraibhurari yePython inoshandisa yakadzama yekudzidza uye yekunyepedzera hungwaru backend inonzi tensor flow kugadzira maNN muchimiro cheakareruka anoteedzana modhi zviri nyore.
Keras inouyawo neyayo preexisting modhi inogona kushandiswa zvakare. Pachidzidzo ichi, tichave tichigadzira yedu modhi tichishandisa Keras.
Iwe unogona kudzidza zvakawanda nezve iyi Deep Kudzidza chimiro kubva ku Keras webhusaiti.
Kuvaka Neural Network (Tutorial)
Ngatienderere mberi nekuvaka Neural Network tichishandisa Python.
Dambudziko Chirevo
Neural Networks imhando yemhinduro kuAI-yakavakirwa matambudziko. Pachidzidzo ichi tichave tichienda pamusoro pePima Indians Diabetes Data, inowanikwa pano.
ICU Machine Learning yakagadzira iyi dataset uye ine rekodhi yekurapa yevarwere veIndia. Muenzaniso wedu unofanirwa kufanotaura kana murwere aine chirwere cheshuga mukati memakore mashanu kana kwete.
Loading Dataset
Dataset yedu ifaira rimwe chete reCSV rinonzi 'diabetes.csv' rinogona kushandiswa zviri nyore uchishandisa Microsoft Excel.
Tisati tagadzira modhi yedu, isu tinofanirwa kuendesa kunze dataset yedu. Uchishandisa kodhi inotevera unogona kuita izvi:
kupinza pandas se pd
data = pd.read_csv('diabetes.csv')
x = data.drop("Mugumisiro")
y = data["Mugumisiro"]
Pano isu tiri kushandisa iyo pandas raibhurari kuti ikwanise kushandura yedu CSV faira data, read_csv() ibasa rakavakirwa-mukati rePandas rinotibvumira kuchengetedza kukosha mufaira redu kune inosiyana inonzi 'data'.
Iyo inosiyana x ine yedu dataset isina mhedzisiro (mavara) data. Isu tinowana izvi neiyo data.drop() basa rinobvisa mavara e x, nepo y iine chete mhinduro (label) data.
Kuvaka Sequential Model
Danho 1: Kupinza Maraibhurari
Chekutanga, isu tinofanirwa kupinza TensorFlow neKeras, pamwe nemamwe maparamita anodiwa kune yedu modhi. Iyo inotevera kodhi inotibvumira kuita izvi:
import tensorflow se tf
kubva tensorflow import keras
kubva tensorflow.keras.models import Sequential
kubva tensorflow.keras.layers import Activation, Dense
kubva tensorflow.keras.optimizers pinza Adamu
kubva tensorflow.keras.metrics import categorical_crossentropy
Kune yedu modhi tiri kuunza dense layers. Aya ndiwo mitsara yakanyatsobatanidzwa; kureva kuti, nodi imwe neimwe mudungwe yakanyatsobatanidzwa neimwe nodi muchikamu chinotevera.
Tiri kuunza zvakare kushandiswa basa rinodiwa pakuyera data rinotumirwa kumanodhi. Optimizers zvakare akaunzwa kunze kuti aderedze kurasikirwa.
Adamu ane mukurumbira optimizer inoita kuti yedu yemodhi yekuvandudza node kuverenga zvakanyanya, pamwe chete categorical_crossentropy inova rudzi rwebasa rekurasikirwa (rinoverenga mutsauko pakati pechaiyo uye yakafanotaurwa yemhando yemavara) yatichange tichishandisa.
Danho 2: Kugadzira Muenzaniso Wedu
Modhi yandiri kugadzira ine imwe yekupinda (ine gumi nematanhatu mayunitsi), imwe yakavanzika (ine makumi matatu nemaviri mayunitsi) uye imwe inobuda (ine 16 maunitsi) layer. Nhamba idzi hadzina kugadziriswa uye zvinoenderana zvachose nedambudziko rakapihwa.
Kuisa nhamba chaiyo yemayuniti uye akaturikidzana inzira inogona kuvandudzwa overtime kuburikidza nekudzidzira. Activation inoenderana nerudzi rwekuyera yatichange tichiita pane yedu data tisati tapfuura nepa node.
Relu neSoftmax vane mukurumbira activation mabasa ebasa iri.
muenzaniso = Sequential ([
Dense(mayunitsi = 16, input_shape = (1,), activation = 'relu'),
Dense(mayuniti = 32, activation = 'relu'),
Dense(mayunitsi = 2, activation = 'softmax')
])
Hezvino izvo chidimbu chemodhi chinofanira kutaridzika:
Kudzidzisa Muenzaniso
Modhi yedu ichadzidziswa mumatanho maviri, yekutanga kuve kugadzira modhi (kuisa modhi pamwe chete) uye inotevera ichikodzera iyo modhi pane yakapihwa dataset.
Izvi zvinogona kuitwa uchishandisa iyo model.compile() function inoteverwa neiyo model.fit() function.
model.compile(optimizer = Adam(learning_rate = 0.0001), loss ='binary_crossentropy', metrics = ['acuracy'])
model.fit(x, y, epochs = 30, batch_size = 10)
Kutsanangura iyo 'accuracy' metric inotibvumira kuona chokwadi chemodhi yedu panguva yekudzidziswa.
Sezvo mavara edu ari muchimiro che 1's na0's, tichange tichishandisa bhinari kurasikirwa basa kuverengera mutsauko uripo pakati pemazita chaiwo uye akafanotaurwa.
Iyo dataset iri kupatsanurwa kuita mabhechi egumi (batch_size) uye ichapfuudzwa nemuenzaniso kamakumi matatu (epochs). Kune dataset rakapihwa, x angave data uye y angave mazita anoenderana nedata.
Kuedza Model Kushandisa Mafungidziro
Kuti tiongorore modhi yedu, tinoita fungidziro pane data rebvunzo tichishandisa kufanotaura () basa.
kufanotaura = muenzaniso.predict(x)
Uye ndizvo!
Iwe zvino unofanirwa kuve nekunzwisisa kwakanaka kweiyo Deep Learning application, Neural Networks, mashandiro avanoita mune zvese uye maitiro ekuvaka, kudzidzisa uye kuyedza modhi muPython kodhi.
Ndinovimba chidzidzo ichi chinokupa iwe kickstart kugadzira uye kuendesa yako Yakadzika Kudzidza modhi.
Tizivise mumashoko kana chinyorwa chacho chaibatsira.
Leave a Reply