Abashakashatsi Bashyizeho Amayeri Yimibare Yuburyo Bwiza Utubuto duto Utarinze GPU yawe

Imbonerahamwe

• Ibisobanuro na 1 Intangiriro
• Uburyo 2
  • 2.1 Igisekuru cya PV-gishingiye ku gaciro kamwe kangirika
  • 2.2 Igisekuru cya PV-gishingiye ku kubora kwa PLS
• 3 Ibisubizo
  • 3.1 Imibare
  • 3.2 ANN gusubira inyuma kwamakuru ya Tecator
  • 3.3 ANN itondekanya amakuru yumutima
• 4 Ikiganiro
  • 5 Imyanzuro

Ibisobanuro

Muri iyi nyandiko, turasaba uburyo bushya bwo kongera imibare namakuru avanze. Uburyo butanga amakuru yinyongera ukoresheje kwambukiranya-kwemeza no guhinduranya ibintu byihishe. Irakora cyane cyane kuri datasets zifite urwego ruciriritse kandi ruri hejuru ya collinearite, kuko ikoresha neza iyi mitungo ibisekuruza. Uburyo buroroshye, bwihuse, kandi bufite ibipimo bike cyane, nkuko bigaragara mu mpapuro, bidasaba guhuza neza. Yageragejwe kuri datasets nyinshi zifatika; hano, turatanga raporo irambuye kubibazo bibiri, guhanura poroteyine mu nyama zacishijwe bugufi hashingiwe hafi ya infrarafarike (imibare yuzuye yuzuye hamwe na collinearite) hamwe n'ivangura ry'abarwayi boherejwe na coronary angiography (amakuru avanze, hamwe numubare uhindagurika, no gushyira mu gaciro). Muri ibyo bihugu byombi, imiyoboro y’imitsi ikoreshwa mu guteza imbere gusubira inyuma no kuvangura. Ibisubizo byerekana iterambere rigaragara mumikorere yicyitegererezo; kubwibyo guhanura poroteyine yinyama, guhuza icyitegererezo namakuru yiyongereye byatumye igabanuka ryumuzi bivuze ikosa rya kwaduka ryabazwe kubizamini byigenga byashyizweho inshuro 1.5 kugeza kuri 3.

Ijambo ryibanze : kongera amakuru, imiyoboro yubukorikori yubukorikori, Procrustes yambukiranya imipaka, impinduka zihishe, guhuza

1 Intangiriro

Uburyo bwa kijyambere bwo kwiga imashini bushingiye kubintu bigoye cyane, nkurusobe rwimitsi (ANN), bisaba amakuru menshi yo guhugura no kunoza imiterere. Amahugurwa adahagije akenshi atera ibibazo bikabije, kuko umubare wa hyperparameter ntangarugero kugirango uhuze ni munini cyane kuruta umubare wubwisanzure muri dataset.

Ikindi kibazo gikunze kugaragara muri uru rubanza ni ukubura imyororokere kuko inzira ya ANN yo guhugura ntabwo igenwa, urebye guhitamo byimazeyo ibipimo byambere byerekana imiterere nuburyo bukomeye bwo kubitezimbere. Kubwibyo, ntabwo biganisha ku cyitegererezo gifite ibipimo bimwe nibikorwa, kuko ibigeragezo bitandukanye byamahugurwa bishobora kuvamo imiterere itandukanye. Ihinduka riba rinini niba imyitozo yashizweho ari nto cyane.

Iki kibazo cyihutirwa cyane cyane kubijyanye no guhuza amakuru yubushakashatsi, kuko akenshi bihenze kandi bitwara igihe cyo gukora ibigeragezo byinshi byubushakashatsi, bigatuma bidashoboka gusa gukusanya ibipimo ibihumbi nibihumbi bikenewe mumahugurwa akwiye kandi neza. Hashobora kubaho izindi mbogamizi, nkimpapuro zijyanye nimpushya mubushakashatsi bwubuvuzi.

Bumwe mu buryo bwo gutsinda ikibazo cyamahugurwa adahagije nukuyongera muburyo bwo kwigana ingingo nshya cyangwa guhindura bike kubiriho. Ubu buhanga bukunze kwitwa "kongera amakuru". Kongera amakuru byagaragaye ko ari byiza cyane mu gusesengura amashusho no gushyira mu byiciro, hamwe n’ubushakashatsi bunini bwerekana uburyo bwo kongera uburyo butandukanye [1] [2], [3] nuburyo bukoreshwa cyane cyane kubibazo byihariye [4] [5] . Uburyo bwo kuzamura ibihe byuruhererekane nabwo bwateye imbere neza [6].

Ariko, harabura uburyo bunoze bushobora gutanga amakuru meza yo kongera imibare yimibare hamwe na dogere igereranije kugeza murwego rwo hejuru. Imibare nkiyi ikwirakwira mubushakashatsi bwubushakashatsi, harimo ubwoko butandukanye bwamakuru ya spekitroscopique, ibisubizo bikurikirana genome (urugero, 16S RNA), nibindi byinshi. Imibare myinshi yatanzwe kandi yerekana imiterere yimbere aho impinduka zifitanye isano. Kugeza ubu uburyo buboneka bwo kongera amakuru nkaya ahanini bushingiye ku kongeramo urusaku rutandukanye [7] kubipimo bihari, ntabwo buri gihe bihagije. Hariho kandi uburyo butanga ikizere bukoresha autoencoders zinyuranye ukoresheje icyitegererezo cyaturutse kumwanya wabo uhindagurika [8], cyangwa uburyo bushingiye kumurongo wibyara inyungu [4]. Ikibi ni uko ubwo buryo bwombi busaba kubaka no guhuza imiterere yihariye y'urusobe rw'imitsi kugirango hongerwe amakuru bityo bikenera uburyo bunoze kandi busaba inzira yo gutezimbere hamwe n'amahugurwa manini yatangijwe.

Muri iyi nyandiko, turasaba uburyo bworoshye, bwihuse, butandukanye, ariko bukora neza kugirango twongere imibare kandi ivanze na collinear datasets. Uburyo bushingiye ku buryo bwatangiye gukorwa ku zindi ntego, cyane cyane mu gutanga ibyemezo byemewe, bityo bikaba bizwi nka Procrustes cross-validation [9] [10]. Nyamara, nkuko bigaragara muriyi mpapuro, ikemura neza ikibazo cyo kongera amakuru, bikavamo moderi zitezimbere cyane guhanura cyangwa gukora ibyiciro.

Uburyo bwacu bukoresha neza ubufatanye muburyo bwo kubyara. Ihuza amahugurwa yamahugurwa hamwe nuruhererekane rwibihinduka byihishe hanyuma ikoresha cross-kwemeza gusubiramo kugirango ipime itandukaniro mubyerekezo byimpinduka. Iri tandukaniro noneho ritangizwa mumahugurwa yashizweho nkikosa ryikitegererezo, bikavamo urutonde rushya rwamakuru.

Moderi ebyiri zibereye zirashobora gukoreshwa - agaciro kamwe kubora (SVD) hamwe nibice byibuze kwaduka (PLS) kubora. Guhitamo icyitegererezo gikwiye bituma uyikoresha ashyira imbere igice cyimiterere ya covariance, izakoreshwa mugusohora amakuru mashya.

Moderi zombi zikwiranye zifite ibipimo bibiri - umubare wibihinduka byihishe hamwe numubare wibice bikoreshwa muguhuza-kwemeza. Ubushakashatsi bwerekana nubwo ibipimo bidasaba guhuza neza. Umubare uwo ariwo wose wihishe munini bihagije kugirango ufate gahunda ihindagurika yimyitozo yashizweho indangagaciro ikora neza. Nkumubare uwo ariwo wose wibice bitangirira kuri bitatu.

Uburyo buteganijwe burahuriweho kandi burashobora gukoreshwa muburyo bwuzuye bwimibare kimwe no kubisobanuro byatanzwe aho kimwe cyangwa byinshi bihinduka byujuje ubuziranenge. Ibi bifungura ikindi cyerekezo, aricyo gusebanya amakuru, bishobora kuba ingirakamaro, urugero, mugupima sisitemu ya software yuzuye, nubwo tutareba iyi ngingo hano.

Urupapuro rusobanura ishingiro ryuburyo bwuburyo kandi rugaragaza uburyo bukoreshwa nuburyo bukora bishingiye ku mibare ibiri yimiterere itandukanye. Itanga ibisobanuro birambuye byukuntu uburyo bwakoreshwa neza muburyo butandukanye bwamakuru atandukanye.

Twashyize mubikorwa uburyo mundimi nyinshi zo gutangiza gahunda, harimo Python, R, MATLAB, na JavaScript, kandi ibyashyizwe mubikorwa byose biraboneka kubuntu mububiko bwa GitHub (https://github.com/svkucheryavski/pcv). Byongeye kandi, dutanga verisiyo yo kumurongo aho umuntu ashobora kubyara amakuru mashya muri mushakisha (https://mda.tools/pcv).