M'ndandanda wazopezekamo[Bisani][Show]
- 1. Fotokozani kusiyana pakati pa kuphunzira pamakina, luntha lochita kupanga, ndi kuphunzira mozama.
- 2. Chonde fotokozani mitundu yosiyanasiyana yophunzirira makina.
- 3. Kodi kukondera ndi kusinthanitsa kusinthanitsa ndi chiyani?
- 4. Makina ophunzirira makina asintha kwambiri pakapita nthawi. Kodi munthu amasankha bwanji algorithm yoyenera kuti agwiritse ntchito popatsidwa seti ya data?
- 5. Kodi covariance ndi kugwirizana zimasiyana bwanji?
- 6. Pophunzira makina, kusonkhanitsa kumatanthauza chiyani?
- 7. Kodi mumakonda makina ophunzirira makina otani?
- 8. Kutsika kwa Linear mu Kuphunzira kwa Makina: Ndi Chiyani?
- 9. Fotokozani kusiyana pakati pa KNN ndi k-njira masango.
- 10. Kodi “kukondera” kumatanthauza chiyani kwa inu?
- 11. Kodi Theorem ya Bayes ndi chiyani kwenikweni?
- 12. Mu Machine Learning Model, 'Training Set' ndi 'test Set' ndi chiyani?
- 13. Kodi Hypothesis in Machine Learning ndi chiyani?
- 14. Kodi kuphunzira makina kumatanthauza chiyani, ndipo kungapewedwe bwanji?
- 15. Kodi magulu a Naive Bayes ndi chiyani kwenikweni?
- 16. Kodi Ntchito za Mtengo ndi Ntchito Zotayika zimatanthauza chiyani?
- 17. Nchiyani chimasiyanitsa chitsanzo choberekera ndi chosankha?
- 18. Fotokozani kusiyana pakati pa zolakwika za Type I ndi Type II.
- 19. Pophunzira makina, njira yophunzirira ya Ensemble ndi chiyani?
- 20. Kodi ma parametric model ndi chiyani kwenikweni? Perekani chitsanzo.
- 21. Fotokozani kusefa kogwirizana. Komanso kusefa otengera zomwe zili?
- 22. Kodi mukutanthauza chiyani kwenikweni ndi mndandanda wa Nthawi?
- 23. Fotokozani kusiyana pakati pa ma algorithms a Gradient Boosting ndi Random Forest.
- 24. Chifukwa chiyani mukufunikira matrix osokonezeka? Ndi chiyani?
- 25. Kodi kusanthula chigawo cha mfundo ndi chiyani kwenikweni?
- 26. Kodi nchifukwa ninji kusinthasintha kwa chigawo kuli kofunika kwambiri kwa PCA (kusanthula chigawo chachikulu)?
- 27. Kodi kukhazikika ndi kukhazikika kumasiyana bwanji?
- 28. Kodi kukhazikika ndi kukhazikika kumasiyana bwanji ndi mnzake?
- 29. Kodi “variance inflation factor” ikutanthauza chiyani kwenikweni?
- 30. Kutengera ndi kukula kwa maphunziro, mumasankha bwanji gulu?
- 31. Ndi ndondomeko yanji yophunzirira makina yomwe imatchedwa "wophunzira waulesi" ndipo chifukwa chiyani?
- 32. Kodi ROC Curve ndi AUC ndi chiyani?
- 33. Kodi ma hyperparameter ndi chiyani? Nchiyani chimawapangitsa iwo kukhala osiyana ndi magawo achitsanzo?
- 34. Kodi F1 Score, kukumbukira, ndi kulondola kumatanthauza chiyani?
- 35. Kodi kutsimikizirika ndi chiyani kwenikweni?
- 36. Tiyerekeze kuti mwapeza kuti chitsanzo chanu chili ndi kusiyana kwakukulu. Ndi algorithm iti, m'malingaliro anu, yomwe ili yoyenera kuthana ndi vutoli?
- 37. Ndi chiyani chomwe chimasiyanitsa Regression Regression ndi Lasso regression?
- 38. Chofunika kwambiri ndi chiyani: ntchito yachitsanzo kapena kulondola kwachitsanzo? Ndi iti ndipo chifukwa chiyani mungakonde?
- 39. Kodi mungasamalire bwanji deta yokhala ndi zosagwirizana?
- 40. Kodi mungasiyanitse bwanji pakati pa boosting ndi thumba?
- 41. Kufotokoza kusiyana pakati pa kuphunzira mongobwerezabwereza ndi mongodula mawu.
- Kutsiliza
Mabizinesi akugwiritsa ntchito ukadaulo wotsogola, monga nzeru zamakono (AI) ndi kuphunzira pamakina, kuti awonjezere kupezeka kwa chidziwitso ndi ntchito kwa anthu.
Ukadaulo uwu ukutengedwa ndi mafakitale osiyanasiyana, kuphatikiza mabanki, azachuma, ogulitsa, opanga, ndi azaumoyo.
Imodzi mwamaudindo omwe amafunidwa kwambiri pamabungwe omwe amagwiritsa ntchito AI ndi asayansi azama data, mainjiniya anzeru zopangira, mainjiniya ophunzirira makina, ndi owunika ma data.
Izi positi adzakutsogolerani zosiyanasiyana makina kuphunzira mafunso oyankhulana, kuyambira pachiyambi mpaka zovuta, kukuthandizani kukonzekera mafunso aliwonse omwe mungafunsidwe mukafuna ntchito yanu yabwino.
1. Fotokozani kusiyana pakati pa kuphunzira pamakina, luntha lochita kupanga, ndi kuphunzira mozama.
Luntha lochita kupanga limagwiritsa ntchito njira zosiyanasiyana zophunzirira makina komanso njira zophunzirira mozama zomwe zimalola makompyuta kuti azigwira ntchito pogwiritsa ntchito luntha lofanana ndi laumunthu ndi malingaliro ndi malamulo.
Kuphunzira pamakina kumagwiritsa ntchito ziwerengero zosiyanasiyana komanso njira zophunzirira mwakuya kuti makina azitha kuphunzira kuchokera pa zomwe adachita kale ndikukhala aluso kwambiri pochita ntchito zina paokha popanda kuyang'aniridwa ndi munthu.
Deep Learning ndi mndandanda wa ma aligorivimu omwe amalola pulogalamuyo kuti iphunzire kuchokera payokha ndikuchita ntchito zosiyanasiyana zamalonda, monga kuzindikira mawu ndi zithunzi.
Machitidwe omwe amawulula ma multilayered awo mawindo a neural kuzinthu zambiri zophunzirira amatha kuphunzira mwakuya.
2. Chonde fotokozani mitundu yosiyanasiyana yophunzirira makina.
Kuphunzira makina kulipo m'mitundu itatu yosiyana mokulira:
- Maphunziro Oyang'aniridwa: Mtundu umapanga zolosera kapena ziweruzo pogwiritsa ntchito deta yolembedwa kapena ya mbiriyakale pamaphunziro a makina oyang'aniridwa. Ma data omwe adayikidwapo kapena olembedwa kuti awonjezere tanthauzo lake amatchedwa deta yolembedwa.
- Maphunziro Osayang'aniridwa: Tilibe chidziwitso cholembedwa chamaphunziro osayang'aniridwa. Mu data yomwe ikubwera, chitsanzo chikhoza kupeza machitidwe, zosamvetsetseka, ndi zogwirizana.
- Maphunziro Olimbikitsa: Chitsanzocho chikhoza phunzirani pogwiritsa ntchito kulimbikitsa kuphunzira ndi mphotho zomwe lidapeza chifukwa cha zomwe adachita kale.
3. Kodi kukondera ndi kusinthanitsa kusinthanitsa ndi chiyani?
Kuwotcha mochulukira kumachitika chifukwa cha kukondera, komwe ndi kuchuluka komwe mtundu umatengera deta. Kukondera kumayambitsidwa ndi malingaliro olakwika kapena osavuta kwambiri mwanu makina ophunzirira algorithm.
Kusiyana kumatanthawuza zolakwa zomwe zimadza chifukwa cha zovuta mu algorithm yanu ya ML, yomwe imapangitsa chidwi pakusintha kwakukulu kwa data yophunzitsira ndi kuchulukitsa.
Kusiyanasiyana ndiko kuchuluka kwa ma model omwe amasiyanasiyana malinga ndi zolowetsa.
Mwa kuyankhula kwina, zitsanzo zoyambira ndizokondera kwambiri koma zokhazikika (kusiyana kochepa). Kuwotcha kwambiri ndi vuto ndi zitsanzo zovuta, ngakhale zimagwira zenizeni zachitsanzo (kukondera kochepa).
Pofuna kupewa kusiyanasiyana kwakukulu komanso kukondera kwakukulu, kusinthanitsa pakati pa kukondera ndi kusiyanasiyana ndikofunikira kuti kuchepetse zolakwika.
4. Makina ophunzirira makina asintha kwambiri pakapita nthawi. Kodi munthu amasankha bwanji algorithm yoyenera kuti agwiritse ntchito popatsidwa seti ya data?
Njira yophunzirira pamakina yomwe iyenera kugwiritsidwa ntchito zimangotengera mtundu wa data yomwe ili mumtundu wina.
Pamene deta ili ndi mzere, kubwereza kwa mzere kumagwiritsidwa ntchito. Njira yosungiramo zinthu ingachite bwino ngati deta ikuwonetsa kuti ilibe mzere. Titha kugwiritsa ntchito mitengo yaziganizo kapena SVM ngati deta ikuyenera kuwunikidwa kapena kutanthauziridwa pazamalonda.
Neural network ikhoza kukhala yothandiza kupeza yankho lolondola ngati dataset ili ndi zithunzi, makanema, ndi zomvera.
Kusankha algorithm pazochitika zinazake kapena kusonkhanitsa deta sikungapangidwe pamlingo umodzi wokha.
Pofuna kupanga njira yabwino kwambiri, choyamba tiyenera kufufuza deta pogwiritsa ntchito kufufuza deta (EDA) ndi kumvetsa cholinga chogwiritsira ntchito deta.
5. Kodi covariance ndi kugwirizana zimasiyana bwanji?
Covariance imayang'ana momwe zosinthika ziwiri zimalumikizirana wina ndi mnzake komanso momwe wina angasinthire potengera kusintha kwa wina.
Ngati zotsatira zake ndi zabwino, zimasonyeza kuti pali kugwirizana kwachindunji pakati pa zosinthazo komanso kuti wina akhoza kukwera kapena kuchepa ndi kuwonjezeka kapena kuchepa kwa kusiyana kwa maziko, poganiza kuti zinthu zina zonse zimakhalabe.
Kulumikizana kumayesa ulalo pakati pa mitundu iwiri yachisawawa ndipo kumakhala ndi mikhalidwe itatu yokha: 1, 0, ndi -1.
6. Pophunzira makina, kusonkhanitsa kumatanthauza chiyani?
Njira zophunzirira mosayang'aniridwa zomwe zimasonkhanitsira deta pamodzi zimatchedwa clustering. Ndi kusonkhanitsa mfundo za deta, njira yophatikizira ingagwiritsidwe ntchito.
Mutha kuyika ma data onse molingana ndi ntchito zawo pogwiritsa ntchito njirayi.
Mawonekedwe ndi mikhalidwe ya ma data omwe amagwera m'gulu lomwelo ndi ofanana, pomwe zomwe za data zomwe zimagwera m'magulu osiyanasiyana ndizosiyana.
Njirayi ingagwiritsidwe ntchito pofufuza ziwerengero.
7. Kodi mumakonda makina ophunzirira makina otani?
Muli ndi mwayi wowonetsa zomwe mumakonda komanso maluso apadera mu funso ili, komanso chidziwitso chanu chambiri zamakina ophunzirira makina.
Nawa ma aligorivimu angapo ophunzirira makina oti muganizirepo:
- Kukongoletsa kwamzera
- Kutsika kwa mayendedwe
- Naive Bayes
- Mitengo yamisankho
- K amatanthauza
- Random Forest algorithm
- Woyandikana nawo wa K (KNN)
8. Kutsika kwa Linear mu Kuphunzira kwa Makina: Ndi Chiyani?
Algorithm yoyang'aniridwa yophunzirira makina ndikubwerera m'mizere.
Amagwiritsidwa ntchito pofufuza zolosera kuti adziwe kugwirizana kwa mzere pakati pa mitundu yodalira ndi yodziimira.
Linear regression equation ili motere:
Y = A + BX
kumene:
- Cholowa kapena chosiyana chodziyimira pawokha chimatchedwa X.
- Mtundu wodalira kapena wotuluka ndi Y.
- Coefficient ya X ndi b, ndipo kulowera kwake ndi a.
9. Fotokozani kusiyana pakati pa KNN ndi k-njira masango.
Kusiyanitsa kwakukulu ndikuti KNN (njira yogawa, kuphunzira koyang'aniridwa) imafunikira mfundo zolembedwa pomwe k-njira sizitero (clustering algorithm, kuphunzira kosayang'aniridwa).
Mutha kuyika deta yolembedwa pamalo osalembedwa pogwiritsa ntchito a K-Nearest Neighbors. Kuphatikizika kwa njira za K kumagwiritsa ntchito mtunda wapakati pakati pa mfundo kuti aphunzire kupanga magulu osalembedwa.
10. Kodi “kukondera” kumatanthauza chiyani kwa inu?
Kukondera mu gawo lazoyeserera ndi chifukwa cha kusalondola kwa ziwerengero.
Gulu limodzi lachitsanzo limasankhidwa mobwerezabwereza kuposa magulu ena omwe akuyesa chifukwa cha zolakwika.
Ngati chisankhocho sichikuvomerezedwa, chikhoza kubweretsa malingaliro olakwika.
11. Kodi Theorem ya Bayes ndi chiyani kwenikweni?
Tikadziwa zotheka zina, titha kudziwa zotheka pogwiritsa ntchito Bayes' Theorem. Zimapereka mwayi wakumbuyo wa zomwe zachitika potengera zomwe zidachitika kale, mwa kuyankhula kwina.
Njira yabwino yowerengera kuthekera kokhazikika ikuperekedwa ndi chiphunzitso ichi.
Pamene mukupanga zovuta zowonetseratu zamagulu ndikugwirizanitsa chitsanzo ku maphunziro dataset mu kuphunzira makina, theorem ya Bayes imagwiritsidwa ntchito (ie Naive Bayes, Bayes Optimal Classifier).
12. Mu Machine Learning Model, 'Training Set' ndi 'test Set' ndi chiyani?
Seti ya maphunziro:
- Maphunzirowa ali ndi zochitika zomwe zimatumizidwa ku chitsanzo kuti zifufuzidwe ndi kuphunzira.
- Izi ndi zomwe zalembedwa zomwe zidzagwiritsidwe ntchito pophunzitsa chitsanzo.
- Kawirikawiri, 70% ya deta yonse imagwiritsidwa ntchito ngati dataset yophunzitsira.
Mayeso Seti:
- Seti yoyeserera imagwiritsidwa ntchito kuyesa kulondola kwa m'badwo wa fanizo lachitsanzo.
- Timayesa popanda deta yolembedwa ndiyeno timagwiritsa ntchito zilembo kutsimikizira zotsatira.
- 30% yotsalayo imagwiritsidwa ntchito ngati dataset yoyeserera.
13. Kodi Hypothesis in Machine Learning ndi chiyani?
Kuphunzira Kwamakina kumathandizira kugwiritsa ntchito ma dataset omwe alipo kale kuti amvetsetse bwino ntchito yomwe yaperekedwa yomwe imalumikiza zomwe zatuluka. Izi zimadziwika kuti ntchito pafupifupi.
Pachifukwa ichi, kuyerekezera kuyenera kugwiritsidwa ntchito pa ntchito yosadziwika bwino kuti asamutsire zonse zomwe zingatheke potengera zomwe zaperekedwa m'njira yabwino kwambiri.
Pophunzira pamakina, lingaliro ndi chitsanzo chomwe chimathandizira kuyerekezera ntchito yomwe mukufuna ndikumaliza kupanga mapu oyenerera.
Kusankhidwa ndi mapangidwe a ma algorithms amalola kutanthauzira kwa malo omwe angakhalepo omwe angathe kuimiridwa ndi chitsanzo.
Pamalingaliro amodzi, zilembo zocheperako h (h) zimagwiritsidwa ntchito, koma likulu h (H) limagwiritsidwa ntchito pamalingaliro onse omwe akufufuzidwa. Tiwunikanso mwachidule zolemba izi:
- Lingaliro (h) ndi mtundu wina womwe umathandizira kupanga mapu a zolowa mpaka zotuluka, zomwe pambuyo pake zitha kugwiritsidwa ntchito pakuwunika ndi kulosera.
- Hypothesis set (H) ndi malo osakira amalingaliro omwe angagwiritsidwe ntchito kupanga mapu pazotuluka. Kukonzekera kwa nkhani, ma model, ndi makonzedwe a zitsanzo ndi zitsanzo zochepa chabe za malire amtundu uliwonse.
14. Kodi kuphunzira makina kumatanthauza chiyani, ndipo kungapewedwe bwanji?
Pamene makina akuyesera kuphunzira kuchokera ku deta yosakwanira, kuwonjezereka kumachitika.
Zotsatira zake, kuchuluka kwa data kumalumikizidwa mosagwirizana ndi kuchuluka kwa data. Njira yotsimikizika yolumikizira imalola kuti kuchulukitsitsa kupewedwe pama dataset ang'onoang'ono. Dongosolo la data lagawidwa magawo awiri munjira iyi.
Dongosolo loyesera ndi maphunziro lidzakhala ndi magawo awiriwa. Deta yophunzitsira imagwiritsidwa ntchito kupanga chitsanzo, pamene deta yoyesera imagwiritsidwa ntchito kuyesa chitsanzo pogwiritsa ntchito zolowetsa zosiyanasiyana.
Umu ndi momwe mungapewere kuchulukitsa.
15. Kodi magulu a Naive Bayes ndi chiyani kwenikweni?
Njira zosiyanasiyana zamagulu zimapanga magulu a Naive Bayes. Ma algorithms omwe amadziwika kuti ogawa magulu onse amagwira ntchito pamalingaliro ofanana.
Lingaliro lopangidwa ndi osazindikira a Bayes classifiers ndikuti kupezeka kapena kusapezeka kwa chinthu chimodzi sikukhudzana ndi kupezeka kapena kusapezeka kwa chinthu china.
Mwa kuyankhula kwina, izi ndi zomwe timatcha "zopanda pake" chifukwa zimapangitsa kulingalira kuti mawonekedwe amtundu uliwonse ndi ofunika kwambiri komanso odziimira.
Kugawika kumachitika pogwiritsa ntchito ma naive Bayes classifiers. Ndiosavuta kugwiritsa ntchito ndikupanga zotsatira zabwino kuposa zolosera zovuta kwambiri pamene maziko odziyimira pawokha ndi owona.
Posanthula zolemba, kusefa sipamu, ndi machitidwe opangira, amagwiritsidwa ntchito.
16. Kodi Ntchito za Mtengo ndi Ntchito Zotayika zimatanthauza chiyani?
Mawu akuti "kutaya ntchito" amatanthauza ndondomeko ya kutayika kwa kompyuta pamene chidutswa chimodzi chokha cha deta chikuganiziridwa.
Mosiyana, timagwiritsa ntchito mtengo kuti tidziwe kuchuluka kwa zolakwika pazambiri zambiri. Palibe kusiyana kwakukulu komwe kulipo.
Mwa kuyankhula kwina, pamene ntchito zamtengo wapatali zimaphatikiza kusiyana kwa deta yonse ya maphunziro, ntchito zotayika zimapangidwira kuti zizindikire kusiyana pakati pa zikhalidwe zenizeni ndi zonenedweratu za mbiri imodzi.
17. Nchiyani chimasiyanitsa chitsanzo choberekera ndi chosankha?
Chitsanzo cha tsankho chimaphunzira kusiyana pakati pa magulu angapo a deta. Chitsanzo chopanga chimatengera mitundu yosiyanasiyana ya data.
Pa zovuta zamagulu, zitsanzo za tsankho nthawi zambiri zimakhala bwino kuposa zitsanzo zina.
18. Fotokozani kusiyana pakati pa zolakwika za Type I ndi Type II.
Zonama zabodza zimagwera m'gulu la zolakwika za Type I, pomwe zolakwika zabodza zimapita pansi pa zolakwika za Type II (kunena kuti palibe chomwe chachitika pomwe zidachitikadi).
19. Pophunzira makina, njira yophunzirira ya Ensemble ndi chiyani?
Njira yotchedwa ensemble learning imasakaniza mitundu yambiri yophunzirira makina kuti apange mitundu yamphamvu kwambiri.
Chitsanzo chikhoza kukhala chosiyana pazifukwa zosiyanasiyana. Zifukwa zingapo ndi:
- Anthu Osiyanasiyana
- Ma Hypotheses osiyanasiyana
- Njira zowonetsera
Tidzakumana ndi vuto pogwiritsa ntchito maphunziro ndi kuyesa kwa chitsanzo. Kukondera, kusiyana, ndi zolakwika zosasinthika ndi mitundu yotheka ya cholakwika ichi.
Tsopano, timatcha kulinganiza kumeneku pakati pa kukondera ndi kusiyanasiyana kwachitsanzo kusinthanitsa kwa tsankho, ndipo kuyenera kukhalapo nthawi zonse. Kusinthanitsa uku kumachitika pogwiritsa ntchito maphunziro a ensemble.
Ngakhale pali njira zingapo zophatikizira zomwe zilipo, pali njira ziwiri zofananira zophatikiza mitundu yambiri:
- Njira yachibadwidwe yotchedwa bagging imagwiritsa ntchito maphunzirowa kuti apange maphunziro owonjezera.
- Kukweza, njira yotsogola kwambiri: Mofanana ndi thumba, kukweza kumagwiritsidwa ntchito kupeza njira yolemetsa yophunzitsira.
20. Kodi ma parametric model ndi chiyani kwenikweni? Perekani chitsanzo.
Pali magawo ochepa mumitundu yama parametric. Kuti muwonetsetse deta, zomwe muyenera kudziwa ndi magawo amtunduwo.
Zotsatirazi ndi zitsanzo zodziwika bwino: kusinthika kwazinthu, kutsika kwa mzere, ndi ma SVM amzere. Mitundu yopanda parametric imatha kusinthika chifukwa imatha kukhala ndi magawo ambiri opanda malire.
Magawo achitsanzo ndi momwe zinthu ziliri ndizofunikira pakulosera kwa data. Nazi zitsanzo zodziwika bwino: zitsanzo za mutu, mitengo yachisankho, ndi oyandikana nawo k-pafupi.
21. Fotokozani kusefa kogwirizana. Komanso kusefa otengera zomwe zili?
Njira yoyeserera ndi yowona yopangira malingaliro ogwirizana ndi kusefa kwapagulu.
Njira yolimbikitsira yotchedwa collaborative filtering imaneneratu zatsopano polinganiza zokonda za ogwiritsa ntchito ndi zomwe amagawana.
Zokonda za ogwiritsa ndiye chinthu chokhacho chomwe machitidwe olimbikitsa opangira zinthu amaganizira. Potengera zomwe wogwiritsa adasankha kale, malingaliro atsopano amaperekedwa kuchokera kuzinthu zogwirizana.
22. Kodi mukutanthauza chiyani kwenikweni ndi mndandanda wa Nthawi?
Mndandanda wa nthawi ndi mndandanda wa manambala mokwerera. Pa nthawi yokonzedweratu, imayang'anitsitsa kayendetsedwe ka deta yosankhidwa ndipo nthawi ndi nthawi imajambula mfundo za deta.
Palibe kuyika kwanthawi kochepa kapena kopitilira muyeso kwa mndandanda wanthawi.
Mndandanda wa nthawi nthawi zambiri umagwiritsidwa ntchito ndi akatswiri kusanthula deta mogwirizana ndi zofunikira zawo zapadera.
23. Fotokozani kusiyana pakati pa ma algorithms a Gradient Boosting ndi Random Forest.
Forest Forest:
- Mitengo yambiri yosankha imasonkhanitsidwa pamodzi pamapeto ndipo imadziwika kuti nkhalango zachisawawa.
- Ngakhale kuti kukwera kwamitengo kumapanga mtengo uliwonse mopanda mitengo ina, nkhalango mwachisawawa imamanga mtengo uliwonse pa nthawi.
- Multiclass kuzindikira kwa chinthu zimagwira ntchito bwino ndi nkhalango zosasintha.
Kuchulukitsa kwa Gradient:
- Pomwe nkhalango Zachisawawa zimalumikizana ndi mitengo yachigamulo kumapeto kwa ndondomekoyi, Gradient Boosting Machines amaziphatikiza kuyambira pachiyambi.
- Ngati magawo asinthidwa moyenera, kukwera kwa gradient kumapambana nkhalango zosasinthika malinga ndi zotsatira, koma sichosankha mwanzeru ngati deta ili ndi zotuluka zambiri, zosokoneza, kapena phokoso chifukwa zitha kupangitsa kuti mtunduwo ukhale wokwanira.
- Pakakhala deta yosawerengeka, monga momwe zimakhalira nthawi yeniyeni yowunika zoopsa, kukweza kwa gradient kumachita bwino.
24. Chifukwa chiyani mukufunikira matrix osokonezeka? Ndi chiyani?
Gome lomwe limadziwika kuti confusion matrix, lomwe nthawi zina limadziwika kuti matrix olakwika, limagwiritsidwa ntchito kwambiri kuwonetsa momwe mtundu wamagulu, kapena gulu, umagwirira ntchito pagulu la data yoyeserera yomwe milingo yeniyeni imadziwika.
Zimatithandiza kuwona momwe chitsanzo kapena algorithm imagwirira ntchito. Zimapangitsa kukhala kosavuta kwa ife kuwona kusamvana pakati pa maphunziro osiyanasiyana.
Imagwira ntchito ngati njira yowunikira momwe chitsanzo kapena algorithm imagwirira ntchito.
Zolosera zamtundu wamagulu amapangidwa kukhala matrix osokoneza. Mawerengedwe a ma label a kalasi iliyonse adagwiritsidwa ntchito kuti afotokoze kuchuluka kwa maulosi olondola ndi olakwika.
Limapereka tsatanetsatane wa zolakwika zopangidwa ndi owerengera komanso mitundu yosiyanasiyana ya zolakwika zomwe zimayambitsidwa ndi ogawa.
25. Kodi kusanthula chigawo cha mfundo ndi chiyani kwenikweni?
Pochepetsa kuchuluka kwa zosinthika zomwe zimalumikizana wina ndi mnzake, cholinga chake ndikuchepetsa kukula kwa zosonkhanitsira deta. Koma m'pofunika kusunga zosiyanasiyana mmene ndingathere.
Zosinthazi zimasinthidwa kukhala gulu latsopano lamitundu yosiyanasiyana yotchedwa principal components.
Ma PC awa ndi orthogonal popeza ndi ma eigenvector a covariance matrix.
26. Kodi nchifukwa ninji kusinthasintha kwa chigawo kuli kofunika kwambiri kwa PCA (kusanthula chigawo chachikulu)?
Kusinthasintha ndikofunikira mu PCA chifukwa kumakulitsa kulekanitsa pakati pazosiyanasiyana zomwe zimapezedwa ndi gawo lililonse, kupangitsa kutanthauzira kwagawo kukhala kosavuta.
Timafunikira zigawo zowonjezera kuti tiwonetse kusiyana kwa zigawo ngati zigawozo sizikuzungulira.
27. Kodi kukhazikika ndi kukhazikika kumasiyana bwanji?
Kukhazikika:
Deta imasinthidwa panthawi ya normalization. Muyenera kusintha deta ngati ili ndi masikelo omwe ndi osiyana kwambiri, makamaka kuyambira otsika mpaka apamwamba. Sinthani ndime iliyonse kuti ziwerengero zofunika zonse zigwirizane.
Kuonetsetsa kuti palibe kutayika kolondola, izi zingakhale zothandiza. Kuzindikira chizindikiro ndikunyalanyaza phokoso ndi chimodzi mwa zolinga za maphunziro a chitsanzo.
Pali mwayi wowonjezera ngati chitsanzocho chikupatsidwa mphamvu zonse kuti muchepetse zolakwika.
Kukhazikika:
Pokhazikika, ntchito yolosera imasinthidwa. Izi zitha kulamulidwa ndi kukhazikika, komwe kumathandizira magwiridwe antchito osavuta kuposa ovuta.
28. Kodi kukhazikika ndi kukhazikika kumasiyana bwanji ndi mnzake?
Njira ziwiri zomwe zimagwiritsidwa ntchito kwambiri pakukulitsa mawonekedwe ndizokhazikika komanso zokhazikika.
Kukhazikika:
- Kuyikanso deta kuti igwirizane ndi [0,1] kumadziwika kuti normalization.
- Pamene magawo onse ayenera kukhala ndi sikelo yofananira, kukhazikika kumakhala kothandiza, koma zotuluka za seti ya data zimatayika.
Kukhazikika:
- Deta imasinthidwa kuti ikhale ndi tanthauzo la 0 ndi kupatuka kokhazikika kwa 1 monga gawo la njira yokhazikika (Kusiyanasiyana kwa Unit)
29. Kodi “variance inflation factor” ikutanthauza chiyani kwenikweni?
Chiŵerengero cha kusiyana kwachitsanzo ndi kusiyana kwa chitsanzo chokhala ndi mtundu umodzi wokha wodziyimira pawokha amadziwika kuti variation inflation factor (VIF).
VIF imayesa kuchuluka kwa ma multicollinearity omwe amapezeka mumagulu angapo osinthika.
Kusiyanasiyana kwa Model (VIF) Yokhala ndi Kusiyana Kumodzi Kodziyimira Pamodzi
30. Kutengera ndi kukula kwa maphunziro, mumasankha bwanji gulu?
Kukondera kwakukulu, mtundu wocheperako umachita bwino pakuphunzitsidwa kwakanthawi kochepa chifukwa kuchulukitsitsa ndikochepa. Naive Bayes ndi chitsanzo chimodzi.
Pofuna kuwonetsa kuyanjana kovutirapo pamaphunziro akulu, chitsanzo chokhala ndi tsankho lochepa komanso kusiyana kwakukulu ndikwabwino. Logistic regression ndi chitsanzo chabwino.
31. Ndi ndondomeko yanji yophunzirira makina yomwe imatchedwa "wophunzira waulesi" ndipo chifukwa chiyani?
Wophunzira waulesi, KNN ndi njira yophunzirira makina. Chifukwa K-NN imawerengera mozama mtunda nthawi iliyonse yomwe ikufuna kuyika m'malo mophunzira zomwe amaphunzira pamakina kapena zosintha kuchokera pazomwe amaphunzitsidwa, imaloweza pamtima zowerengera.
Izi zimapangitsa K-NN kukhala wophunzira waulesi.
32. Kodi ROC Curve ndi AUC ndi chiyani?
Kuchita kwa mtundu wamagulu pazigawo zonse kumayimiridwa ndi mawonekedwe a ROC. Lili ndi mlingo weniweni komanso mfundo zabodza.
Mwachidule, malo omwe ali pansi pa ROC curve amadziwika kuti AUC (Area Under the ROC Curve). Dera la mbali ziwiri la ROC curve kuchokera ku (0,0) kupita ku AUC limayesedwa (1,1). Poyesa mitundu ya binary, imagwiritsidwa ntchito ngati chiwerengero cha magwiridwe antchito.
33. Kodi ma hyperparameter ndi chiyani? Nchiyani chimawapangitsa iwo kukhala osiyana ndi magawo achitsanzo?
Kusintha kwamkati kwachitsanzo kumadziwika ngati parameter yachitsanzo. Pogwiritsa ntchito deta yophunzitsira, mtengo wa parameter umayerekezedwa.
Zosadziwika kwa chitsanzocho, hyperparameter ndi yosinthika. Mtengowo sungadziwike kuchokera ku data, chifukwa chake amagwiritsidwa ntchito pafupipafupi kuwerengera magawo amitundu.
34. Kodi F1 Score, kukumbukira, ndi kulondola kumatanthauza chiyani?
Chisokonezo Measure ndi metric yomwe imagwiritsidwa ntchito kuti iwunikire mayendedwe amtundu wamagulu. Mawu otsatirawa atha kugwiritsidwa ntchito kufotokozera bwino za chisokonezo:
TP: Zowona Zowona - Izi ndi zabwino zomwe zinkayembekezeredwa bwino. Zikusonyeza kuti zikhalidwe za kalasi yomwe akuyembekezeredwa ndi kalasi yeniyeni zonse ndi zabwino.
TN: Zoyipa Zowona- Izi ndizovuta zomwe zidanenedweratu molondola. Zimasonyeza kuti zonse zamtengo wapatali za kalasi yeniyeni ndi kalasi yoyembekezeredwa ndizolakwika.
Makhalidwe awa - zabwino zabodza ndi zolakwika zabodza - zimachitika pamene kalasi yanu yeniyeni ikusiyana ndi kalasi yomwe ikuyembekezeredwa.
Tsopano,
Chiŵerengero cha mulingo wowona wabwino (TP) pazowonera zonse zomwe zidachitika m'kalasi lenileni zimatchedwa kukumbukira, komwe kumadziwikanso kuti sensitivity.
Kukumbukira ndi TP/(TP+FN).
Kulondola ndi muyeso wa mtengo wabwino wolosera, womwe umafananiza chiwerengero cha zabwino zomwe chitsanzocho chimaneneratu kuti ndi zingati zolondola zomwe zimaneneratu molondola.
Precision ndi TP/(TP + FP)
Chiyerekezo chosavuta kumvetsetsa ndicholondola, chomwe ndi gawo chabe la zowoneratu zonenedweratu pazowonera zonse.
Kulondola ndikufanana ndi (TP+TN)/(TP+FP+FN+TN).
Precision and Recall amalemedwa ndikuyesedwa kuti apereke F1 Score. Zotsatira zake, mphambu iyi imaganizira zabwino zonse zabodza komanso zolakwika zabodza.
F1 nthawi zambiri imakhala yofunikira kwambiri kuposa kulondola, makamaka ngati muli ndi magawo osagwirizana, ngakhale mwachidziwitso sizosavuta kumvetsetsa ngati kulondola.
Kulondola kwabwino kwambiri kumatheka pamene mtengo wazinthu zabodza ndi zolakwika zabodza zikufanana. Ndikwabwino kuphatikizira zonse ziwiri Precision ndi Recall ngati ndalama zomwe zimagwirizanitsidwa ndi zabwino zabodza ndi zolakwika zabodza zimasiyana kwambiri.
35. Kodi kutsimikizirika ndi chiyani kwenikweni?
Njira yowerengeranso ziwerengero yotchedwa kutsimikizira pakuphunzira pamakina imagwiritsa ntchito magulu angapo a data kuti aphunzitse ndikuwunika njira yophunzirira pamakina osiyanasiyana.
Gulu latsopano la deta lomwe silinagwiritsidwe ntchito pophunzitsa chitsanzo limayesedwa pogwiritsa ntchito kutsimikizira kuti muwone momwe chitsanzocho chikuneneratu. Kuchulukitsitsa kwa data kumaletsedwa chifukwa cha kutsimikizika kopitilira muyeso.
K-Fold Njira yofananira yomwe imagwiritsidwa ntchito kwambiri imagawaniza seti yonse ya data kukhala ma K a makulidwe ofanana. Kumatchedwa cross-validation.
36. Tiyerekeze kuti mwapeza kuti chitsanzo chanu chili ndi kusiyana kwakukulu. Ndi algorithm iti, m'malingaliro anu, yomwe ili yoyenera kuthana ndi vutoli?
Kusamalira kusinthasintha kwakukulu
Tiyenera kugwiritsa ntchito njira yonyamula katundu pamavuto omwe ali ndi kusiyana kwakukulu.
Zitsanzo zobwerezabwereza za data mwachisawawa zitha kugwiritsidwa ntchito ndi bagging algorithm kugawa deta m'magulu ang'onoang'ono. Deta ikagawika, titha kugwiritsa ntchito deta mwachisawawa komanso njira yophunzitsira kuti tipange malamulo.
Pambuyo pake, kuvota kungagwiritsidwe ntchito kuphatikiza zolosera zachitsanzo.
37. Ndi chiyani chomwe chimasiyanitsa Regression Regression ndi Lasso regression?
Njira ziwiri zomwe zimagwiritsidwa ntchito kwambiri ndi Lasso (yomwe imatchedwanso L1) ndi Ridge (yomwe nthawi zina imatchedwa L2) regression. Amagwiritsidwa ntchito poletsa kuchuluka kwa data.
Kuti mupeze yankho labwino kwambiri ndikuchepetsa zovuta, njirazi zimagwiritsidwa ntchito kulanga ma coefficients. Mwa kulanga chiwopsezo chonse cha ma coefficients, regression ya Lasso imagwira ntchito.
Ntchito ya chilango mu Ridge kapena L2 regression imachokera ku kuchuluka kwa ma coefficients.
38. Chofunika kwambiri ndi chiyani: ntchito yachitsanzo kapena kulondola kwachitsanzo? Ndi iti ndipo chifukwa chiyani mungakonde?
Ili ndi funso lachinyengo, motero munthu ayenera kumvetsetsa kuti Model Performance ndi chiyani. Ngati ntchito imatanthauzidwa ngati liwiro, ndiye kuti imadalira mtundu wa ntchito; ntchito iliyonse yokhudzana ndi nthawi yeniyeni ingafune kuthamanga kwambiri ngati gawo lofunikira.
Mwachitsanzo, Zotsatira zabwino kwambiri zakusaka sizikhala zofunikira ngati zotsatira za Mafunso zitenga nthawi yayitali kuti zifike.
Ngati Magwiridwe agwiritsidwa ntchito ngati kulungamitsa chifukwa chake kulondola ndi kukumbukira kuyenera kuyika patsogolo kulondola, ndiye kuti F1 idzakhala yothandiza kwambiri kuposa kulondola pakuwonetsa vuto labizinesi pazosankha zilizonse zomwe sizili bwino.
39. Kodi mungasamalire bwanji deta yokhala ndi zosagwirizana?
Dongosolo losawerengeka la data litha kupindula ndi njira zoyeserera. Sampling ikhoza kuchitidwa m'njira yapansi kapena yoposa.
Pansi pa Sampling imatilola kuchepetsa kukula kwa gulu lalikulu kuti lifanane ndi gulu laling'ono, lomwe limathandizira kuwonjezereka kwachangu pokhudzana ndi kusungirako ndi kuthamangitsidwa kwa nthawi koma kungayambitsenso kutaya deta yamtengo wapatali.
Pofuna kuthetsa vuto la kutayika kwa chidziwitso chifukwa cha sampuli zambiri, timatsanzira gulu la Ochepa; komabe, izi zimatipangitsa kukumana ndi zovuta zochulukirapo.
Njira zinanso ndi izi:
- Cluster-Based Over Sampling- Ochepa komanso ambiri am'magulu amatsatiridwa payekhapayekha ku njira yophatikizira ya K-njira izi. Izi zimachitika kuti mupeze magulu a dataset. Kenako, gulu lirilonse limachulukirachulukira kotero kuti makalasi onse akhale ndi kukula kofanana ndipo magulu onse mkati mwa kalasi amakhala ndi kuchuluka kofanana.
- SMOTE: Synthetic Minority Over-sampling Technique- Chigawo cha data kuchokera m'gulu la anthu ochepa chimagwiritsidwa ntchito monga chitsanzo, pambuyo pake zochitika zina zowonjezera zomwe zimafanana ndi izo zimapangidwa ndikuwonjezeredwa ku dataset yoyambirira. Njirayi imagwira ntchito bwino ndi ma data a manambala.
40. Kodi mungasiyanitse bwanji pakati pa boosting ndi thumba?
Ensemble Techniques ali ndi mitundu yomwe imadziwika kuti thumba ndi kulimbikitsa.
Bagging-
Kwa ma algorithms okhala ndi kusiyanasiyana kwakukulu, kunyamula ndi njira yomwe imagwiritsidwa ntchito kuti muchepetse kusiyana. Banja limodzi lotere la magulu omwe amakonda kukondera ndi banja lachisankho.
Mtundu wa data yomwe mitengo yachisankho imaphunzitsidwa imakhudza kwambiri momwe amagwirira ntchito. Pachifukwa ichi, ngakhale kukonzedwa bwino kwambiri, zotulukapo nthawi zina zimakhala zovuta kwambiri kuzipeza.
Ngati chidziwitso cha mitengo yachisankho chasinthidwa, zotsatira zake zimasiyana kwambiri.
Zotsatira zake, matumba amagwiritsidwa ntchito, momwe mitengo yambiri yosankha imapangidwira, yomwe iliyonse imaphunzitsidwa pogwiritsa ntchito chitsanzo cha deta yoyambirira, ndipo zotsatira zake zimakhala pafupifupi mitundu yonseyi.
Kulimbikitsa:
Kukweza ndi njira yolosera ndi gulu la n-ofooka momwe gulu lililonse lofooka limapanga zofooka za magulu ake amphamvu. Timatchula za gulu lomwe limagwira ntchito molakwika pa data yomwe yaperekedwa ngati "gulu lofooka."
Kukweza mwachiwonekere ndi njira osati ma algorithm. Kubwerera m'mbuyo ndi mitengo yaziganizo zosazama ndi zitsanzo zodziwika bwino za magulu ofooka.
Adaboost, Gradient Boosting, ndi XGBoost ndi njira ziwiri zolimbikitsira zodziwika bwino, komabe, pali zina zambiri.
41. Kufotokoza kusiyana pakati pa kuphunzira mongobwerezabwereza ndi mongodula mawu.
Pophunzira mwachitsanzo kuchokera ku zitsanzo zomwe zawonedwa, chitsanzo chimagwiritsa ntchito maphunziro ochititsa chidwi kuti afike pamapeto omveka bwino. Kumbali inayi, ndi kuphunzira kwapang'onopang'ono, chitsanzocho chimagwiritsa ntchito zotsatira musanapange zake.
Kuphunzira mochititsa chidwi ndi njira yopezera mfundo kuchokera pazowonera.
Kuphunzira kochepetsetsa ndi njira yopangira zowonera potengera zomwe akuganiza.
Kutsiliza
Zikomo! Awa ndi mafunso 40 apamwamba komanso apamwamba ofunsira kuphunzira pamakina omwe tsopano mukudziwa mayankho ake. Sayansi ya data ndi nzeru zochita kupanga ntchito zidzapitirizabe kufunidwa pamene teknoloji ikupita patsogolo.
Ofuna kupititsa patsogolo chidziwitso chawo cha matekinoloje apamwambawa ndikuwongolera luso lawo atha kupeza mwayi wosiyanasiyana wa ntchito ndi malipiro ampikisano.
Mutha kupitiliza kuyankha zofunsazo tsopano popeza mukumvetsetsa bwino momwe mungayankhire ena mwamafunso omwe amafunsidwa kwambiri pophunzira makina.
Malingana ndi zolinga zanu, chitani zotsatirazi. Konzekerani zoyankhulana poyendera Hashdork's Nkhani Zoyankhulana.
Siyani Mumakonda