MultiLinearLogistic Regr-ions

ivan gabrovski

Iqukethe izikhangiso

Mayelana nalolu hlelo lokusebenza

Ngezansi kunesiqondiso esisebenzayo se-Multiple (multivariate) Binary Logistic Regression —okungukuthi, ukubikezela umphumela we-binary (0/1) kusuka ezicini eziningi.

I-Binomial Logistic Regression (ngokuvamile ebizwa ngokuthi i-logistic regression) iyindlela yezibalo esetshenziselwa ukulingisa ubudlelwano phakathi kwe-variable eyodwa noma ngaphezulu ezimele kanye nomphumela we-binary (wezigaba ezimbili).

I-Binary: target y∈{0,1}
I-Multiple (multivariate): isici sokufaka esingaphezu kwesisodwa x_1, x_2, ..., x_n​
Imodeli:
p(y=1∣x)=1/(1+e^{−z}), lapho u-z=w_0+w_1*x_1+⋯+w_n*x_n

kanye no-w_0, w_1...w_n kuyisisindo esibalwa ngu-x_1, x_2, ..., x_n kanye namaphutha phakathi kuka-y kanye nokubikezela.
Esikhundleni sokubikezela amanani ngqo, i-logistic regression ibikezela ama-log-odds kusetshenziswa inhlanganisela eqondile yezibikezeli z. Amathuba okungena aguqulwa kusetshenziswa umsebenzi we-logistic (sigmoid) ukukhiqiza amathuba aphakathi kuka-0 no-1.
I-Binary Logistic Regression iyimodeli yokuhlukanisa okungenzeka esebenzisa umsebenzi we-sigmoid ukubikezela amathuba omunye wemiphumela emibili, okwenza isetshenziswe kabanzi kuzibalo, isayensi yedatha, kanye nokufunda komshini ukuze kwenziwe izinqumo ezimbili ezihunyushwayo.
Amapharamitha emodeli alinganiswa kusetshenziswa i-Maximum Likelihood Estimation (MLE). Inani lomkhawulo (ngokuvamile u-0.5) lisetshenziselwa ukuhlukanisa imiphumela (Uma u-P≥0.5 → ikilasi 1; Uma u-P<0.5 → ikilasi 0).
I-Multinomial logistic regression iyindlela yezibalo kanye nokufunda komshini esetshenziselwa ukulingisa ubudlelwano phakathi kwesethi yeziguquguquko ezizimele (izibikezelo) kanye ne-categorical dependent variable enemiphumela engaphezu kwemibili engenzeka, lapho izigaba zingenakho ukuhleleka kwemvelo.

Imodeli: Yekilasi k:
P(y=k∣x)=e^w_k⋅x / ∑e^w_j⋅x lapho j=1,2...K
Lapho: - x = isici sevektha
w_k = izisindo zekilasi k
K = inani lamakilasi
Kuhlelo lokusebenza into ngayinye i-Object_k(object_1, object_2 ... object_m)ichazwa ngeziguquguquko ezizimele(X_ki – izici, i = 1...n ) kanye ne-variable eyodwa exhomeke kuyo(Y_k -target). Indlela efana nezikwele ezijwayelekile ezincane (OLS) isetshenziswa ukubala amanani afanele kakhulu ama-coefficients (beta_0, beta_1, beta_2, ..., beta_n). Inani eliqondiwe libalwa ngo:
Y = beta_0 + beta_01* P_1 + beta_2 *P_2 + ... + beta_n* P_n
Lapho: P_1, P_2...P_n ziyizibikezelo zethagethi.
Uhlelo lokusebenza lugcina idatha yamamodeli amaningi okubuyisela emuva kwe-logistic ku-database (DB) uhlobo lwe-SQLite olubizwa nge-AppMultiNomialLogisticRegression.db. Amamodeli okubuyisela emuva ahlukaniswa ngamagama.

Isikrini sokuqala sohlelo lokusebenza (i-App Multinomial Linear Logistic Regression Solver) sibonisa uhlu lwamasampula amamodeli okubuyisela emuva (ohlwini lwe-spinner) kanye nezinkinobho zokuvumela imisebenzi ukuthi idale (Isampula entsha), ilayishe (Layisha), igcine (Londoloza), igcine (Londoloza njenge), ibale (Bala), futhi isuse (Susa) amasampula amamodeli okubuyisela emuva. Kusukela esikrinini esiyinhloko, ngezakhi zemenyu, ungafinyelela nemisebenzi efana nokukhetha ulimi, ukulondoloza nokukopisha i-database, ukuqalisa i-database ngedatha yesampula, kanye nemisebenzi yokusiza njengosizo lohlelo lokusebenza, izilungiselelo, kanye nesixhumanisi sewebhusayithi enencazelo yazo zonke izinhlelo zokusebenza ngababhali.

Imisebenzi yokudala (Isampula entsha) ifaka ibhokisi lokufaka usayizi we-matrix lapho kufakwa khona idatha yesampula entsha - inani lemigqa (inombolo ifaka umugqa wedatha ebikezelwe P_1, P_2...P_n– umugqa wokugcina) kanye nenani lamakholomu (inombolo ifaka ikholomu yedatha encike ku-Y_1, Y_2,...Y_k– ikholomu yokugcina). Bese kukhiqiza ithebula lokufaka idatha efanele. Ithebula eligcwele abantu kumele liqanjwe ngaphambi kokulondolozwa. Umsebenzi Layisha usule ithebula.

Ithebula elidala eligciniwe lingase liboniswe ngokukhethwa ohlwini lwe-spinner. Ithebula eliboniswayo lingabalwa futhi isixazululo sivele ebhokisini Imiphumela yohlelo lokusebenza. Umsebenzi Phrinta ungasetshenziswa kusukela kuleli bhokisi lombhalo kufayela le-AppMultipleLogisticRegressionSolver.txt. Umsebenzi Phrinta ufaka ifayela le-Db/Save ngalo kukhethwe ifolda lapho ulondoloza khona ifayela. Ngemuva kokukhetha ifolda kuvela inkinobho yokugcina. Kusukela emsebenzini ofanayo kungaboniswa okuqukethwe kwefayela elikhethiwe, kanye nokususa ifayela elikhethiwe.