Isiqulatho[Fihla][Bonisa]
- 1. Cacisa umahluko phakathi kokufunda koomatshini, ubukrelekrele bokwenziwa, kunye nokufunda nzulu.
- 2. Nceda uchaze iindidi ezahlukeneyo zokufunda koomatshini.
- 3. Yintoni i-bias kunye nomahluko worhwebo?
- 4. Ii-algorithms zokufunda koomatshini ziye zavela kakhulu ngokuhamba kwexesha. Umntu ukhetha njani i-algorithm efanelekileyo yokusebenzisa isethi yedatha enikiweyo?
- 5. I-covariance kunye nokulungelelaniswa kwahluka njani?
- 6. Kukufunda koomatshini, kuthetha ukuthini ukuhlanganisana?
- 7. Yiyiphi ialgorithm yokufunda ngomatshini oyithandayo?
- 8. Ukunciphisa umgca kwiSifundo soMatshini: Yintoni na?
- 9. Chaza umahluko phakathi kwe-KNN kunye ne-k-ithetha ukuhlanganisana.
- 10. Kuthetha ntoni kuwe ukuthi “ukhetho lokukhetha”?
- 11. Yintoni kanye kanye iTheorem kaBayes?
- 12. KwiModeli yokuFunda ngoomatshini, yintoni 'iSeti yoqeqesho' kunye 'neseti yovavanyo'?
- 13. Yintoni i-Hypothesis kwi-Machine Learning?
- 14. Kuthetha ukuthini ukufakwa ngokugqithisileyo komatshini wokufunda, yaye kunokuthintelwa njani?
- 15. Yintoni kanye kanye abahluli baseNaive Bayes?
- 16. Iindleko zeMisebenzi kunye neMisebenzi yeLahleko ithetha ntoni?
- 17. Yintoni eyahlula imodeli yemveliso kumzekelo wocalucalulo?
- 18. Chaza iiyantlukwano phakathi kweempazamo zoHlobo loku-I kunye noHlobo lwe-II.
- 19. Kukufunda koomatshini, yintoni ubuchule bokufunda be-Ensemble?
- 20. Ziziphi iimodeli zeparametric? Nika umzekelo.
- 21. Chaza ukuhluza ngentsebenziswano. Kanye nokucoca okusekwe kumxholo?
- 22. Uthetha ukuthini kanye kanye ngongcelele lweXesha?
- 23. Chaza iiyantlukwano phakathi kwe-Gradient Boosting kunye ne-Random Forest algorithms.
- 24. Kutheni ufuna i-matrix yokudideka? Yintoni?
- 25. Yintoni kanye kanye uhlalutyo lwecandelo lomgaqo?
- 26. Kutheni ukujikeleziswa kwecandelo kubaluleke kakhulu kwi-PCA (uhlalutyo lwecandelo eliyintloko)?
- 27. Uhlengahlengiso kunye nohlengahlengiso lwahluka njani enye kwenye?
- 28. Ukwenziwa kwesiqhelo kunye nokubekwa emgangathweni kwahluke njani kwenye?
- 29. Ithetha ukuthini kanye kanye “i-variance factor inflation”?
- 30. Ngokusekelwe kubungakanani beseti yoqeqesho, ungamkhetha njani umdidi?
- 31. Yiyiphi i-algorithm ekufundeni komatshini ebizwa ngokuba "ngumfundi onobuvila" kwaye ngoba?
- 32. Yintoni i-ROC Curve kunye ne-AUC?
- 33. Yintoni i-hyperparameters? Yintoni eyenza ukuba bahluke kwiiparamitha zemodeli?
- 34. Lithetha ukuthini inqaku le-F1, khumbula, kunye nokuchaneka?
- 35. Yintoni kanye kanye ukuqinisekiswa okunqamlezileyo?
- 36. Masithi ufumanise ukuba imodeli yakho inomahluko obalulekileyo. Yeyiphi i-algorithm, ngokoluvo lwakho, ifaneleke kakhulu ukusingatha le meko?
- 37. Yintoni eyahlula ukuhlehla kwe-Ridge kwi-Lasso regression?
- 38. Yiyiphi eyona nto ibalulekileyo: ukusebenza kwemodeli okanye ukuchaneka kwemodeli? Yeyiphi kwaye kutheni ungayithanda?
- 39. Ungalulawula njani uluhlu lwedatha olunokungalingani?
- 40. Ungahlula njani phakathi kwe-boosting kunye nebhegi?
- 41. Cacisa umahluko phakathi kwemfundo yokufundisa kunye neyokuxhuzulelwa imali.
- isiphelo
Amashishini asebenzisa itekhnoloji ye-cutting-edge, efana nobukrelekrele bokwenziwa (AI) kunye nokufunda koomatshini, ukwandisa ukufikeleleka kolwazi kunye neenkonzo kumntu ngamnye.
Obu buchwepheshe bamkelwa ngamashishini ahlukeneyo, kubandakanya ibhanki, imali, ukuthengisa, ukuvelisa, kunye nokhathalelo lwezempilo.
Enye yeendima zombutho ezifunwa kakhulu kusetyenziswa i-AI yeyenzululwazi yedatha, iinjineli zobukrelekrele bokwenziwa, iinjineli zokufunda ngoomatshini, kunye nabahlalutyi bedatha.
Esi sithuba siya kukukhokelela kwiindidi ezahlukeneyo yokufunda umatshini imibuzo yodliwano-ndlebe, ukusuka kwisiseko ukuya kubunzima, ukukunceda ukuba ulungele nayiphi na imibuzo onokuthi uyibuze xa ufuna umsebenzi wakho ofanelekileyo.
1. Cacisa umahluko phakathi kokufunda koomatshini, ubukrelekrele bokwenziwa, kunye nokufunda nzulu.
Ubukrelekrele bokwenziwa busebenzisa iindlela ezahlukeneyo zokufunda koomatshini kunye neendlela zokufunda ezinzulu ezivumela iinkqubo zekhompyuter ukuba zenze imisebenzi zisebenzisa ubukrelekrele obufana nomntu ngengqiqo kunye nemithetho.
Ukufunda ngoomatshini kusebenzisa iinkcukacha-manani ezahlukeneyo kunye neendlela zokuFunda ngokuNzulu ukwenza ukuba oomatshini bafunde ekusebenzeni kwabo kwangaphambili kwaye babe nobuchule ngakumbi bokwenza imisebenzi ethile bebodwa ngaphandle kweliso lomntu.
I-Deep Learning yingqokelela ye-algorithms evumela isoftware ukuba ifunde kuyo kwaye iqhube imisebenzi eyahlukeneyo yentengiso, efana nelizwi kunye nokuqaphela umfanekiso.
Iinkqubo eziveza ubuninzi bazo amanethiwekhi ukuya kwizixa ezikhulu zedatha yokufunda bayakwazi ukwenza ukufunda nzulu.
2. Nceda uchaze iindidi ezahlukeneyo zokufunda koomatshini.
Ukufunda ngoomatshini kukho kwiindidi ezintathu ezahlukeneyo ngokubanzi:
- IsiFundo esiLawulwayo: Imodeli yenza uqikelelo okanye izigwebo usebenzisa idatha ebhaliweyo okanye yembali ekufundeni koomatshini. Iiseti zedatha eziphawulweyo okanye ezilebhile ukuze kwandiswe intsingiselo yazo zibizwa ngokuba yidatha ephawulweyo.
- UkuFunda okungagadwanga: Asinayo idatha ebhaliweyo yokufunda ngokungajongwanga. Kwidatha engenayo, imodeli inokufumana iipateni, izinto ezingaqhelekanga, kunye nokulungelelaniswa.
- UkuFunda okomeleza: Imodeli inako funda ngokusebenzisa ukomeleza ukufunda kunye nemivuzo eyayifumanayo ngokuziphatha kwayo kwangaphambili.
3. Yintoni i-bias kunye nomahluko worhwebo?
Ukufakela ngokugqithisileyo kusisiphumo sokucalu-calucalulo, okuyinqanaba apho imodeli ihambelana nedatha. I-bias ibangelwa yingcinga engachanekanga okanye elula kakhulu kuwe umatshini wokufunda algorithm.
Ukwahluka kubhekiselele kwiimpazamo ezibangelwa ubunzima kwi-algorithm yakho ye-ML, evelisa ubuntununtunu kumanqanaba amakhulu okungafani kwidatha yoqeqesho kunye nokugqithisa.
Umahluko kukuba imodeli iyahluka kangakanani ngokuxhomekeke kumagalelo.
Ngamanye amazwi, iimodeli ezisisiseko zinomkhethe kakhulu kodwa zizinzile (umahluko ophantsi). Ukugqithisa kakhulu kuyingxaki ngeemodeli ezinzima, nangona kunjalo zibamba ubunyani bemodeli (i-bias ephantsi).
Ukuze kuthintelwe zombini ukuhluka okuphezulu kunye nokuthambekela okuphezulu, ukurhweba phakathi kokungakhethi kunye nokungafani kuyafuneka ukuze kuncitshiswe impazamo.
4. Ii-algorithms zokufunda koomatshini ziye zavela kakhulu ngokuhamba kwexesha. Umntu ukhetha njani i-algorithm efanelekileyo yokusebenzisa isethi yedatha enikiweyo?
Ubuchule bokufunda komatshini ekufuneka busetyenziswe buxhomekeke kuphela kuhlobo lwedatha kwiseti yedatha ethile.
Xa idatha ingumgca, ukubuyisela umgca kusetyenziswa. Indlela yokufaka ibhegi ingenza ngcono ukuba idatha ibonise ukungahambelani. Singasebenzisa imithi yesigqibo okanye i-SVM ukuba idatha kufuneka ivavanywe okanye itolikwe ngeenjongo zorhwebo.
Uthungelwano lweNeural lunokuba luncedo ukufumana impendulo echanekileyo ukuba iseti yedatha ibandakanya iifoto, iividiyo, kunye nesandi.
Ukukhethwa kwe-algorithm kwimeko ethile okanye ukuqokelelwa kwedatha akunakukwenziwa nje ngomlinganiselo omnye.
Ngenjongo yokuphuhlisa eyona ndlela ifanelekileyo, kufuneka siqale sihlolisise idatha ngokusebenzisa uhlalutyo lwedatha yokuhlola (EDA) kwaye siqonde injongo yokusebenzisa idatha.
5. I-covariance kunye nokulungelelaniswa kwahluka njani?
I-Covariance ivavanya indlela iinguqu ezimbini ezidityaniswe ngayo omnye komnye kunye nendlela umntu anokutshintsha ngayo ekuphenduleni utshintsho kwelinye.
Ukuba isiphumo silungile, sibonisa ukuba kukho unxibelelwano oluthe ngqo phakathi kwezinto eziguquguqukayo kunye nokuba umntu uya kunyuka okanye anciphise ngokunyuka okanye ukuncipha kwi-variable base, ecinga ukuba zonke ezinye iimeko zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala zihlala khona.
Ulungelelwaniso lulinganisa ikhonkco phakathi kwezinto ezimbini eziguquguqukayo kwaye zinamaxabiso amathathu kuphela ahlukeneyo: 1, 0, kunye -1.
6. Kukufunda koomatshini, kuthetha ukuthini ukuhlanganisana?
Iindlela zokufunda ezingajongwanga ezidibanisa amanqaku edatha zibizwa ngokuba yi-clustering. Ngengqokelela yamanqaku edatha, ubuchule bokudibanisa bunokusetyenziswa.
Ungadibanisa onke amanqaku edatha ngokwemisebenzi yawo usebenzisa esi sicwangciso.
Iimpawu kunye neempawu zamanqaku edatha awela kudidi olufanayo ziyafana, ngelixa ezo zamanqaku edatha awela kumaqela ahlukeneyo ahlukile.
Le ndlela ingasetyenziselwa ukuhlalutya idatha yamanani.
7. Yiyiphi ialgorithm yokufunda ngomatshini oyithandayo?
Unethuba lokubonisa izinto ozikhethayo kunye neetalente ezizodwa kulo mbuzo, kunye nolwazi lwakho olubanzi lweendlela ezininzi zokufunda koomatshini.
Nazi ezimbalwa ii-algorithms zokufunda koomatshini onokucinga ngazo:
- Ukulungiswa komgca kwakhona
- Ukuhlengahlengiswa kwezinto
- Naive Bayes
- Imithi yezigqibo
- K kuthetha
- Random ihlathi algorithm
- Oyena mmelwane u-K (KNN)
8. Ukunciphisa umgca kwiSifundo soMatshini: Yintoni na?
I-algorithm yokufunda koomatshini egadiweyo kukuhlehla ngomgca.
Iqeshwe kuhlalutyo oluqikelelweyo ukumisela uxhulumaniso lomgca phakathi kwezinto ezixhomekeke kunye nezizimeleyo.
Inxaki yobuyiselo ngomgca ngolu hlobo lulandelayo:
Y = A + BX
apho:
- Igalelo okanye umahluko ozimeleyo ubizwa ngokuba ngu-X.
- Ukuxhomekeka okanye ukwahluka kwemveliso nguY.
- I-coefficient ka-X ngu-b, kwaye uqhawulo lwayo ngu-a.
9. Chaza umahluko phakathi kwe-KNN kunye ne-k-ithetha ukuhlanganisana.
Owona mahluko uphambili ngowokuba i-KNN (indlela yokuhlelwa, ukufunda okugadiweyo) ifuna amanqaku alebhileweyo ngelixa u-k-means engafuni (i-algorithm yokuhlanganisana, ukufunda okungagadwanga).
Ungahlela idatha ephawulweyo kwindawo engabhalwanga ngokusebenzisa abamelwane be K-Ekufuphi. I-K-ithetha ukuhlanganisana isebenzisa umgama ophakathi phakathi kwamanqaku ukufunda indlela yokuhlanganisa amanqaku angabhalwanga.
10. Kuthetha ntoni kuwe ukuthi “ukhetho lokukhetha”?
Utyekelo kwisigaba sovavanyo lwesampulu kungenxa yokungachaneki kweenkcukacha-manani.
Elinye iqela lesampulu likhethwa rhoqo kunamanye amaqela kuvavanyo ngenxa yokungachaneki.
Ukuba ukuthambekela kokukhetha akuvunywanga, oko kunokubangela isigqibo esingachanekanga.
11. Yintoni kanye kanye iTheorem kaBayes?
Xa sizazi ezinye izinto ezinokwenzeka, sinokubona ukuba kunokwenzeka sisebenzisa iTheorem yeBayes. Ibonelela ngasemva kokunokwenzeka kwesehlo ngokusekelwe kulwazi lwangaphambili, ngamanye amagama.
Indlela evakalayo yokuqikelela okunokubakho ngokwemiqathango ibonelelwa yile theory.
Xa kuphuhliswa udidi lweengxaki zemodeli yokuxela kwangaphambili kunye nokufaka imodeli kuqeqesho isethi yedatha ekufundeni koomatshini, kusetyenziswe ithiyori yeBayes (okt. Naive Bayes, Bayes Optimal Classifier).
12. KwiModeli yokuFunda ngoomatshini, yintoni 'iSeti yoqeqesho' kunye 'neseti yovavanyo'?
Iseti yoqeqesho:
- Iseti yoqeqesho ineemeko ezithunyelwa kumzekelo ukuze kuhlalutywe kwaye kufundwe.
- Le yidatha ebhaliweyo eya kusetyenziswa ukuqeqesha imodeli.
- Ngokuqhelekileyo, i-70% yedatha iyonke isetyenziswa njengeseti yoqeqesho.
Iseti yoVavanyo:
- Isethi yovavanyo isetyenziselwa ukuvavanya ukuchaneka kwesizukulwana se-hypothesis yemodeli.
- Sivavanya ngaphandle kwedatha ebhaliweyo kwaye sisebenzise iilebhile ukuqinisekisa iziphumo.
- I-30% eseleyo isetyenziswa njengeseti yovavanyo.
13. Yintoni i-Hypothesis kwi-Machine Learning?
Ukufunda ngoomatshini kuvumela ukusetyenziswa kweeseti zedatha esele zikho ukuqonda ngcono umsebenzi onikiweyo odibanisa igalelo kwimveliso. Oku kwaziwa njengoqikelelo lomsebenzi.
Kule meko, uqikelelo kufuneka luqeshelwe umsebenzi ongaziwayo ekujoliswe kuwo ukuze kudluliselwe zonke iingqikelelo ezinokucingeleka ngokusekelwe kwimeko enikiweyo ngeyona ndlela ingcono kakhulu.
Ekufundeni komatshini, i-hypothesis ngumzekelo onceda ekuqikeleleni umsebenzi ojoliswe kuyo kunye nokugqiba i-input-to-output mappings efanelekileyo.
Ukukhethwa kunye noyilo lwe-algorithms luvumela ukuchazwa kwendawo yeengcamango ezinokuthi zimelwe yimodeli.
Kwingqikelelo enye, unobumba omncinci u-h (h) usetyenziswa, kodwa i-capital h (H) isetyenziselwa yonke isithuba sengqikelelo esiphengululwayo. Siza kuphonononga ngokufutshane ezi nkcazo:
- I-hypothesis (h) yimodeli ethile eququzelela ukuhanjiswa kwegalelo kwimveliso, enokuthi emva koko isetyenziswe kuvavanyo kunye noqikelelo.
- Iseti ye-hypothesis (H) yindawo enokukhangelwa yeengqikelelo ezinokuthi zisetyenziswe ukwenza imephu yegalelo kwiziphumo. Ukuqulunqa umba, imodeli, kunye noqwalaselo lwemodeli yimizekelo embalwa yemida yegeneric.
14. Kuthetha ukuthini ukufakwa ngokugqithisileyo komatshini wokufunda, yaye kunokuthintelwa njani?
Xa umatshini uzama ukufunda kwisethi yedatha engonelanga, ukufakwa ngokugqithisileyo kwenzeka.
Ngenxa yoko, i-overfitting ihambelana ngokungafaniyo nomthamo wedatha. Indlela yokuqinisekisa i-cross-validation ivumela ukugqithisa ukuba kuphetshwe kwiiseti zedatha ezincinci. Iseti yedatha yahlulwe yangamacandelo amabini kule ndlela.
Idathasethi yovavanyo noqeqesho iya kuba nala macandelo mabini. I-dataset yoqeqesho isetyenziselwa ukudala imodeli, ngelixa i-dataset yokuvavanya isetyenziselwa ukuvavanya imodeli isebenzisa iigalelo ezahlukeneyo.
Le yindlela yokuthintela ukugqithisa.
15. Yintoni kanye kanye abahluli baseNaive Bayes?
Iindlela ezahlukeneyo zokuhlela zenza abahlela beNaive Bayes. Iseti yomgaqo-nkqubo owaziwa ngokuba ngaba bahluli bonke basebenza kwingcinga enye esisiseko.
Ingqikelelo eyenziwe ngabahluli beBayes abangenalwazi kukuba ubukho okanye ukungabikho kwenqaku elithile akunanto yakwenza nokubakho okanye ukungabikho kolunye uphawu.
Ngamanye amazwi, le nto sibhekisa kuyo “njengesingenangqondo” kuba isenza uqikelelo lokuba uphawu ngalunye lweseti yedatha lubaluleke ngokulinganayo kwaye luzimele.
Ukuhlelwa kwenziwa kusetyenziswa abahluli beBayes abangenalwazi. Zilula ukuzisebenzisa kwaye zivelise iziphumo ezingcono kuneziqikelelo ezinzima ngakumbi xa indawo yokuzimela iyinyani.
Kucazululo lokubhaliweyo, ukuhluzwa kwe-spam, kunye neenkqubo zokuncoma, ziqeshwe.
16. Iindleko zeMisebenzi kunye neMisebenzi yeLahleko ithetha ntoni?
Ibinzana elithi "umsebenzi welahleko" libhekiselele kwinkqubo yokulahleka kwekhompyuter xa icandelo elinye ledatha lithathelwa ingqalelo.
Ngokuchaseneyo, sisebenzisa umsebenzi weendleko ukumisela isixa esipheleleyo seempazamo kwiidatha ezininzi. Akukho mahluko ubalulekileyo okhoyo.
Ngamanye amazwi, ngexa imisebenzi yeendleko idibanisa umahluko wesethi yedatha yoqeqesho iyonke, imisebenzi yelahleko iyilelwe ukubamba umahluko phakathi kwawona maxabiso aqikelelweyo nerekhodi enye.
17. Yintoni eyahlula imodeli yemveliso kumzekelo wocalucalulo?
Imodeli ecalulayo ifunda umahluko phakathi kweendidi ezininzi zedatha. Imodeli evelisayo ithatha kwiintlobo ezahlukeneyo zedatha.
Kwiingxaki zokuhlela, iimodeli zocalucalulo zihlala zigqwesa ezinye iimodeli.
18. Chaza iiyantlukwano phakathi kweempazamo zoHlobo loku-I kunye noHlobo lwe-II.
Iipositi zobuxoki ziwela phantsi kodidi lweempazamo zoHlobo lwe-I, ngelixa ii-negatives ezibubuxoki zingena phantsi kweempazamo zoHlobo lwe-II (ukubanga ukuba akukho nto yenzekileyo xa yenzekile ngokwenene).
19. Kukufunda koomatshini, yintoni ubuchule bokufunda be-Ensemble?
Indlela ebizwa ngokuba yi-ensemble learning idibanisa iimodeli ezininzi zokufunda koomatshini ukuvelisa iimodeli ezinamandla ngakumbi.
Imodeli inokuhluka ngenxa yezizathu ezahlukeneyo. Ziliqela izizathu ezi:
- Abemi abahlukeneyo
- Iingcamango ezahlukeneyo
- Iindlela ezahlukeneyo zemodeli
Siza kuhlangabezana nengxaki ngelixa sisebenzisa imodeli yoqeqesho kunye nedatha yovavanyo. Umkhethe, umahluko, kunye nempazamo engenakunqandeka ziintlobo ezinokwenzeka zale mpazamo.
Ngoku, silubiza olu lungelelwaniso phakathi komkhethe kunye nokwahluka kwimodeli yorhwebo-yantlukwano-yantlukwano, kwaye kufuneka ihlale ikhona. Olu tshintsho lwenziwa ngokusetyenziswa kwemfundo edibeneyo.
Nangona kukho iindlela ezahlukeneyo zokuhlanganisa ezikhoyo, kukho iindlela ezimbini eziqhelekileyo zokudibanisa iimodeli ezininzi:
- Indlela yemveli ebizwa ngokuba yi-bagging isebenzisa isethi yoqeqesho ukuvelisa iiseti zoqeqesho ezongezelelweyo.
- Ukonyusa, ubuchule obuntsonkothileyo: Kufana nokufaka ibhegi, ukonyusa kusetyenziswa ukufumana eyona fomula yobunzima iseti yoqeqesho.
20. Ziziphi iimodeli zeparametric? Nika umzekelo.
Kukho inani elincinci leeparameters kwiimodeli zeparametric. Ukuqikelela idatha, konke okufuneka ukwazi ziiparamitha zemodeli.
Le ilandelayo yimizekelo eqhelekileyo: uhlehleso lolungiselelo, uhlehlo lomgca, kunye nee-SVM zomgca. Iimodeli ezingezizo iiparametric ziyaguquguquka kuba zinokuqulatha inani elingenamkhawulo leeparamitha.
Iiparamitha zemodeli kunye nobume bedatha ephawulweyo ziyafuneka kuqikelelo lwedatha. Nantsi imizekelo eqhelekileyo: imifuziselo yesihloko, imithi yesigqibo, kunye nabamelwane abasondeleyo k-.
21. Chaza ukuhluza ngentsebenziswano. Kanye nokucoca okusekwe kumxholo?
Indlela ezanyiweyo kwaye eyinyani yokudala iingcebiso ezilungiselelwe umxholo kukuhluza ngentsebenziswano.
Uhlobo lwenkqubo yeengcebiso ebizwa ngokuba kuhluzo olusebenzisanayo luxela kwangaphambili imathiriyeli entsha ngokulungelelanisa izinto ezikhethwa ngumsebenzisi kunye nezinto ezinomdla ekwabelwana ngazo.
Izinto ezikhethwa ngumsebenzisi kuphela kwento eqwalaselwa ziinkqubo ezisekelwe kumxholo. Ngokubhekiselele kukhetho lwangaphambili lomsebenzisi, iingcebiso ezintsha zinikezelwa kwizinto ezinxulumene nazo.
22. Uthetha ukuthini kanye kanye ngongcelele lweXesha?
Uluhlu lwexesha yingqokelela yamanani ngokwenyuka ngolandelelwano. Ngethuba lexesha elimiselweyo, libeka esweni ukuhamba kwamanqaku edatha akhethiweyo kwaye ngamaxesha athile ibamba amanqaku edatha.
Akukho galelo lincinci okanye liphezulu lexesha kuthotho lwexesha.
Uluhlu lwexesha lusetyenziswa rhoqo ngabahlalutyi ukuhlalutya idatha ngokuhambelana neemfuno zabo ezizodwa.
23. Chaza iiyantlukwano phakathi kwe-Gradient Boosting kunye ne-Random Forest algorithms.
Ihlathi elingaqhelekanga:
- Inani elikhulu lemithi yesigqibo idityaniswe kunye ekugqibeleni kwaye iyaziwa njengamahlathi angaqhelekanga.
- Ngelixa i-gradient boost ivelisa umthi ngamnye ngaphandle kweminye, ihlathi elingenamkhethe lakha umthi ngamnye ngexesha.
- Iiklasi ezininzi ukubona into isebenza kakuhle kunye namahlathi random.
Ukunyusa iGradient:
- Ngelixa amahlathi eRandom ejoyina imithi yesigqibo ekupheleni kwenkqubo, oomatshini bokunyusa iGradient badibanisa kwasekuqaleni.
- Ukuba iiparamitha zihlengahlengiswa ngokufanelekileyo, ukunyuswa kwegradient kugqwesa amahlathi angenamkhethe ngokweziphumo, kodwa ayilokhetho lobulumko ukuba isethi yedatha inobuninzi bezinto ezingaphandle, i-anomalies, okanye ingxolo kuba inokubangela ukuba imodeli ibe ngaphezulu.
- Xa kukho idatha engalinganiyo, njengoko kukho kuvavanyo lwexesha lokwenyani lomngcipheko, ukunyuswa kwegradient kusebenza kakuhle.
24. Kutheni ufuna i-matrix yokudideka? Yintoni?
Itheyibhile eyaziwa ngokuba yi-confusion matrix, ngamanye amaxesha eyaziwa ngokuba yimpazamo ye-matrix, isetyenziswa ngokubanzi ukubonisa indlela imodeli yohlelo, okanye umdidi, osebenza ngayo kwiseti yedatha yovavanyo apho amaxabiso okwenene aziwa ngayo.
Isenza ukuba sibone indlela imodeli okanye i-algorithm esebenza ngayo. Kwenza kube lula kuthi ukubona ukungaqondani phakathi kwezifundo ezahlukeneyo.
Isebenza njengendlela yokuvavanya indlela imodeli okanye i-algorithm eyenziwe ngayo.
Uqikelelo lwemodeli yohlelo ludityaniswe lube lubhidano lwematriksi. Amanani okubalwa kwileyibhile yeklasi nganye asetyenziselwe ukwahlula inani lilonke loqikelelo oluchanekileyo nolungachanekanga.
Inika iinkcukacha kwiimpazamo ezenziwe ngumhleli kunye neendidi ezahlukeneyo zeempazamo ezibangelwa ngabahluli.
25. Yintoni kanye kanye uhlalutyo lwecandelo lomgaqo?
Ngokunciphisa inani leenguqu ezidityanisiweyo enye kwenye, injongo kukunciphisa ubungakanani bokuqokelelwa kwedatha. Kodwa kubalulekile ukugcina ukuhlukahluka kangangoko kunokwenzeka.
Izinto eziguquguqukayo zitshintshwa zibe yisethi entsha ngokupheleleyo yeenguqu ezibizwa ngokuba ziiprincipal components.
Ezi PC ziyi-orthogonal kuba ziyi-covariance matrix's eigenvectors.
26. Kutheni ukujikeleziswa kwecandelo kubaluleke kakhulu kwi-PCA (uhlalutyo lwecandelo eliyintloko)?
Ukujikeleza kubalulekile kwi-PCA kuba ikhulisa ukwahlula phakathi kweyantlukwano efunyenwe licandelo ngalinye, ukwenza ukutolika kwecandelo kube lula.
Sifuna amacandelo awandisiweyo ukuvakalisa ukwahluka kwecandelo ukuba amacandelo awajikeleziswanga.
27. Uhlengahlengiso kunye nohlengahlengiso lwahluka njani enye kwenye?
Uhlengahlengiso:
Idatha iyatshintshwa ngexesha lesiqhelo. Kuya kufuneka uqheleke idatha ukuba inezikali ezihluke kakhulu, ngakumbi ukusuka ezantsi ukuya phezulu. Lungisa ikholamu nganye ukuze izibalo ezisisiseko zihambelane.
Ukuqinisekisa ukuba akukho kulahleka kokuchaneka, oku kunokuba luncedo. Ukufumanisa umqondiso ngelixa ungayihoyi ingxolo enye yeenjongo zoqeqesho lomzekelo.
Kukho ithuba lokugqithisa ukuba imodeli inikwe ulawulo olupheleleyo ukunciphisa impazamo.
Uhlengahlengiso:
Ngokuhlengahlengiswa, umsebenzi woqikelelo uyalungiswa. Oku kuxhomekeke kulawulo oluthile ngokuhlengahlengiswa, okuthanda imisebenzi elula yokufaka kunentsonkothileyo.
28. Ukwenziwa kwesiqhelo kunye nokubekwa emgangathweni kwahluke njani kwenye?
Ezona zindlela zimbini zisetyenziswa kakhulu zokulinganisa amanqaku kukuqheleka kunye nokubekwa emgangathweni.
Uhlengahlengiso:
- Ukubuyisela kwakhona idatha ukuze ihambelane noluhlu lwe- [0,1] luyaziwa njengesiqhelo.
- Xa zonke iiparameters kufuneka zibe nesikali esifanayo esilungileyo, ukuqheleka kuyanceda, kodwa ii-outliers zeseti yedatha zilahlekile.
Uhlengahlengiso:
- Idatha ihlaziywa ukuba ibe nentsingiselo ye-0 kunye nokuphambuka okusemgangathweni kwe-1 njengenxalenye yenkqubo yokulinganisa (i-Unit variance)
29. Ithetha ukuthini kanye kanye “i-variance factor inflation”?
Umyinge wokumahluko wemodeli ukuya kumahluko wemodeli kunye nenye kuphela eyahlukileyo ezimeleyo eyaziwa ngokuba yi-variation inflation factor (VIF).
I-VIF iqikelela ubungakanani be-multicollinearity ekhoyo kwiseti yeenguqu ezininzi zokuhlehla.
Ukwahluka kwemodeli (VIF) iModeli kunye noMahluko omnye oZimeleyo
30. Ngokusekelwe kubungakanani beseti yoqeqesho, ungamkhetha njani umdidi?
I-bias ephezulu, imodeli ephantsi yomahluko yenza ngcono kwiseti yoqeqesho olufutshane njengoko ukufakwa ngokugqithisileyo kunqabile. I-Naive Bayes ngomnye umzekelo.
Ukuze ubonise intsebenziswano enzima ngakumbi kwiseti yoqeqesho olukhulu, imodeli ene-bias ephantsi kunye nokuhluka okuphezulu kuyakhethwa. Uhlengahlengiso lolungiselelo ngumzekelo omhle.
31. Yiyiphi i-algorithm ekufundeni komatshini ebizwa ngokuba "ngumfundi onobuvila" kwaye ngoba?
Umfundi olivila, i-KNN yi-algorithm yokufunda ngomatshini. Ngenxa yokuba i-K-NN ibala ngokuguquguqukayo umgama ngexesha ngalinye inqwenela ukuwuhlela endaweni yokufunda nawaphi na amaxabiso afundiweyo ngomatshini okanye izinto eziguquguqukayo ukusuka kwidatha yoqeqesho, ibamba ngentloko iseti yedatha yoqeqesho.
Oku kwenza u-K-NN abe ngumfundi olivila.
32. Yintoni i-ROC Curve kunye ne-AUC?
Ukusebenza kwemodeli yokuhlelwa kuyo yonke imigangatho imelwe ngegraphical yi-ROC curve. Inezinga eliqinisekileyo eliqinisekileyo kunye nemilinganiselo yobuxoki.
Ukubeka nje, indawo ephantsi kwe-ROC curve yaziwa ngokuba yi-AUC (Indawo ephantsi kwe-ROC Curve). Indawo ye-ROC yejika-mbini ukusuka (0,0) ukuya kwi-AUC ilinganiswa (1,1). Ukuvavanya imodeli yokuhlela yokubini, isetyenziswa njengeenkcukacha-manani zokusebenza.
33. Yintoni i-hyperparameters? Yintoni eyenza ukuba bahluke kwiiparamitha zemodeli?
Ukuguquguquka kwangaphakathi kwemodeli kwaziwa njengeparamitha yemodeli. Ukusebenzisa idatha yoqeqesho, ixabiso lepharamitha liqikelelwe.
Ayaziwa kwimodeli, i-hyperparameter yinto eguquguqukayo. Ixabiso alinakuqinisekiswa kwidatha, ngoko ke basetyenziswa rhoqo ukubala iiparamitha zemodeli.
34. Lithetha ukuthini inqaku le-F1, khumbula, kunye nokuchaneka?
Ukubhideka koMlinganiselo yimetric esetyenziselwe ukulinganisa impumelelo yemodeli yohlelo. La mabinzana alandelayo anokusetyenziswa ukuchaza ngcono i-metric yokubhideka:
I-TP: IiNgcono eziyinyaniso - La ngamaxabiso alungileyo ebelindelwe ngokufanelekileyo. Icebisa ukuba amaxabiso eklasi eqikelelweyo kunye neklasi eyiyo zombini zilungile.
TN: Inyaniso engalunganga- La ngamaxabiso abi axelwe kwangaphambili ngokuchanekileyo. Iphakamisa ukuba zombini ixabiso leklasi yokwenyani kunye neklasi elilindelekileyo libi.
Ezi xabiso-izinto ezintle ezingeyonyani kunye nezinto ezingalunganga-ziyenzeka xa iklasi yakho yokwenyani ihluke kwiklasi elindelweyo.
Ngoku,
Umlinganiselo wezinga eliqinisekileyo eliqinisekileyo (TP) kuyo yonke imigqaliselo eyenziwe kwiklasi yokwenyani ibizwa ngokuba yi-recall, ekwabizwa ngokuba bubuntununtunu.
Ukukhumbula kwakhona yi-TP/(TP+FN).
Ukuchaneka ngumlinganiselo wexabiso eliqinisekileyo eliqikelelwayo, elithelekisa inani lezinto ezintle imodeli iqikelela ngokwenene ukuba zingaphi iipositi ezichanekileyo eziziqikelele ngokuchanekileyo.
Ukuchaneka yi-TP/(TP + FP)
Eyona metric yentsebenzo ilula ukuqondwa kukuchaneka, okulinani nje loqwalaselo oluqikelelwe kwangaphambili kulo lonke uqwalaselo.
Ukuchaneka kuyalingana no (TP+TN)/(TP+FP+FN+TN).
Ukuchaneka kunye nokuKhumbula zilinganiswe kwaye zilinganiswe ukubonelela ngeNqaku leF1. Ngenxa yoko, eli nqaku liqwalasela zombini iimpawu zobuxoki kunye nezichasi zobuxoki.
I-F1 ihlala inexabiso ngakumbi kunokuchaneka, ngakumbi ukuba unolwabiwo lodidi olungalinganiyo, nokuba ngokuqondayo akukho lula ukuyiqonda njengokuchaneka.
Ukuchaneka okugqwesileyo kuphunyezwa xa ixabiso lezinto ezingalunganga kunye nezichasi zobuxoki zifaniswa. Kukhethwa ukubandakanya zombini i-Precision kunye ne-Recall ukuba iindleko ezinxulumene ne-positives yobuxoki kunye ne-negatives yobuxoki yahluke kakhulu.
35. Yintoni kanye kanye ukuqinisekiswa okunqamlezileyo?
Indlela yokulinganisa kwakhona ngokweenkcukacha-manani ebizwa ngokuba yi-cross-validation ekufundeni koomatshini isebenzisa iiseti zedatha ezininzi ukuqeqesha kunye nokuvavanya i-algorithm yokufunda komatshini kwinani lemijikelo.
Ibhetshi entsha yedatha engasetyenziswanga ukuqeqesha imodeli ivavanywa kusetyenziswa ukuqinisekiswa okunqamlezayo ukubona ukuba imodeli iqikelela kangakanani. Ukugqithiswa kwedatha kuthintelwe ngokuqinisekiswa okunqamlezayo.
I-K-Songa Eyona ndlela isetyenziswa rhoqo yokulinganisa kwakhona yahlula isethi yedatha iyonke kwiiseti ze-K zobukhulu obulinganayo. Ibizwa ngokuba yi-cross-validation.
36. Masithi ufumanise ukuba imodeli yakho inomahluko obalulekileyo. Yeyiphi i-algorithm, ngokoluvo lwakho, ifaneleke kakhulu ukusingatha le meko?
Ukulawula ukuguquguquka okuphezulu
Kufuneka sisebenzise ubuchule bokufaka ingxowa kwiingxaki ezinoguquko olukhulu.
Iisampulu ephindaphindiweyo yedatha engakhethiyo iya kusetyenziswa yi-algorithm ye-bagging ukwahlula idatha kumaqela amancinci. Emva kokuba idatha ihlulwe, sinokusebenzisa idatha engacwangciswanga kunye nenkqubo ethile yoqeqesho ukuvelisa imithetho.
Emva koko, ukuvota kungasetyenziselwa ukudibanisa ukuqikelelwa kwemodeli.
37. Yintoni eyahlula ukuhlehla kwe-Ridge kwi-Lasso regression?
Iindlela ezimbini ezisetyenziswa ngokubanzi yiLasso (ekwabizwa ngokuba yi-L1) kunye ne-Ridge (ngamanye amaxesha ibizwa ngokuba yi-L2) ukuhlehla. Zisetyenziselwa ukuthintela ukugqithiswa kwedatha.
Ukuze ufumane esona sisombululo singcono kwaye unciphise ubunzima, obu buchule busetyenziselwa ukohlwaya i-coefficients. Ngokuhlawulisa inani elipheleleyo lee-coefficients ezipheleleyo, i-Lasso regression isebenza.
Umsebenzi wesohlwayo kwi-Ridge okanye i-L2 yokuhlehla ivela kwi-sum of square of coefficients.
38. Yiyiphi eyona nto ibalulekileyo: ukusebenza kwemodeli okanye ukuchaneka kwemodeli? Yeyiphi kwaye kutheni ungayithanda?
Lo ngumbuzo okhohlisayo, ngoko ke umntu kufuneka aqale aqonde ukuba yintoni iModeli yokuSebenza. Ukuba ukusebenza kuchazwa njengesantya, ngoko kuxhomekeke kuhlobo lwesicelo; Nasiphi na isicelo esibandakanya imeko yexesha lokwenyani iya kufuna isantya esiphezulu njengenxalenye ebalulekileyo.
Umzekelo, ezona ziPhumo zibalaseleyo zoPhando ziya kuba zincinci ukuba iziphumo zoMbuzo zithatha ixesha elide ukufika.
Ukuba iNtsebenzo isetyenziswe njengesizathu sokuba kutheni ukuchaneka kunye nokukhumbula kufuneka kubekwe phambili ngaphezu kokuchaneka, ngoko amanqaku eF1 aya kuba luncedo ngakumbi kunokuchaneka ekuboniseni imeko yeshishini kuyo nayiphi na isethi yedatha engalinganiyo.
39. Ungalulawula njani uluhlu lwedatha olunokungalingani?
Iseti yedatha engalungelelananga ingazuza kubuchule bokusampula. Ukuthathwa kwesampulu kunokwenziwa nokuba kungaphantsi okanye kwisampulu engaphezulu.
Ngaphantsi kweSampuli isivumela ukuba sinciphise ubungakanani beklasi yesininzi ukuze sihambelane neklasi encinci, encedisa ekwandeni kwesantya malunga nokugcinwa kunye nokuphunyezwa kwexesha elisebenzayo kodwa kunokubangela ukulahlekelwa kwedatha ebalulekileyo.
Ukuze kulungiswe umba welahleko yolwazi ebangelwe kukuthatha iisampulu ngokugqithisileyo, senza isampula yodidi lwabaNcinci; nangona kunjalo, oku kusibangela ukuba singene kwimiba egqwesileyo.
Izicwangciso ezongezelelweyo ziquka:
- Iqela-Esekwe Phezu kweSampulu- Ubuncinci kunye neemeko zeklasi zesininzi zixhomekeke kubuchule bokudibanisa i-K kule meko. Oku kwenziwa ukufumana amaqela esethi yedatha. Emva koko, iqoqo ngalinye lixutywe ngaphezulu kwesampulu ukuze zonke iiklasi zibe nobukhulu obufanayo kwaye onke amaqoqo ngaphakathi kweklasi abe nenani elilinganayo lemizekelo.
- I-SMOTE: iSynthetic Minority Over-sampling Technique-Isilayidi sedatha evela kwiklasi yabancinci isetyenziswa njengomzekelo, emva koko iimeko ezongezelelweyo ezenziweyo ezithelekisekayo ziveliswa kwaye zongezwa kwidathasethi yokuqala. Le ndlela isebenza kakuhle ngamanqaku edatha yamanani.
40. Ungahlula njani phakathi kwe-boosting kunye nebhegi?
IiTekhnoloji ze-Ensemble zineenguqulelo ezaziwa ngokuba yibhegi kunye nokunyusa.
Ibhegi-
Kwii-algorithms ezinoguquko oluphezulu, ibhegi yindlela esetyenziselwa ukuthoba umahluko. Olunye usapho olunjalo lwabafundi olutyekele ekubeni nomkhethe lusapho lomthi wesigqibo.
Uhlobo lwedatha ethi imithi yezigqibo iqeqeshwe kuyo inempembelelo ebalulekileyo ekusebenzeni kwayo. Ngenxa yoku, nangona ukulungiswa kuphezulu kakhulu, ukuhlanganiswa kweziphumo ngamanye amaxesha kunzima kakhulu ukufumana kuzo.
Ukuba idatha yoqeqesho lwemithi yesigqibo itshintshiwe, iziphumo ziyahluka kakhulu.
Ngenxa yoko, i-bagging isetyenziswa, apho imithi emininzi yesigqibo yenziwe, nganye iqeqeshwa ngokusebenzisa isampuli yedatha yokuqala, kwaye umphumo wokugqibela ngumyinge wazo zonke ezi modeli zahlukeneyo.
Ukunyusa:
I-Boosting bubuchule bokwenza uqikelelo nge-n-ebuthathaka yesixokelelwano sohlelo apho umdidiyeli ngamnye obuthathaka enzela iintsilelo zabo bahlulahlula abanamandla. Sibhekisa kumdidi osebenza kakubi kwiseti yedatha enikiweyo “njengomdidi obuthathaka.”
Ukunyusa ngokucacileyo yinkqubo kune-algorithm. Uhlengahlengiso lolungiselelo kunye nemithi yezigqibo ezingenzulu yimizekelo eqhelekileyo yabahleli ababuthathaka.
I-Adaboost, i-Gradient Boosting, kunye ne-XGBoost zezona zi-algorithms zokunyusa zaziwayo, nangona kunjalo, zininzi ngakumbi.
41. Cacisa umahluko phakathi kwemfundo yokufundisa kunye neyokuxhuzulelwa imali.
Xa ufunda ngomzekelo kwiseti yemizekelo eqatshelweyo, imodeli isebenzisa ukufunda ngenkuthalo ukufikelela kwisigqibo esibanzi. Kwelinye icala, ngokufunda ngokunciphisa, imodeli isebenzisa isiphumo ngaphambi kokuba yenze eyayo.
Ukufunda ngenkuthalo yinkqubo yokwenza izigqibo ngokuqatshelweyo.
Imfundo ethotyiweyo yinkqubo yokuyila imigqaliselo esekelwe kwingqikelelo.
isiphelo
Halala! Le yimibuzo ephezulu engama-40 nangaphezulu yodliwano-ndlebe yokufunda ngomatshini ozaziyo ngoku iimpendulo zayo. Isayensi yedatha kunye kukubhadla okungeyonyani imisebenzi iya kuqhubeka ifunwa njengoko iteknoloji ihambela phambili.
Abaviwa abahlaziya ulwazi lwabo kwezi teknoloji zokusika kwaye baphucule isethi yabo yezakhono banokufumana iindidi ezininzi zamathuba omsebenzi kunye nomvuzo okhuphisanayo.
Ungaqhubeka nokuphendula udliwano-ndlebe ngoku ukuba unokuqonda okuqinileyo malunga nendlela yokuphendula eminye imibuzo ebuzwa ngokubanzi ngomatshini wodliwanondlebe wodliwano-ndlebe.
Ngokuxhomekeka kusukelo lwakho, thabatha eli nyathelo lilandelayo. Lungiselela udliwano-ndlebe ngokundwendwela iHashdork's Uluhlu lodliwano-ndlebe.
Shiya iMpendulo