Zviri Mukati[Viga][Ratidza]
- 1. Tsanangura mutsauko uripo pakati pekudzidza nemuchina, njere dzekugadzira, uye kudzidza kwakadzama.
- 2. Ndapota tsanangurai marudzi akasiyana-siyana ekudzidza kwemichina.
- 3. Chii chinonzi bias maringe nekusiyana kwekutengeserana?
- 4. Michina yekudzidza algorithms yakashanduka zvakanyanya nekufamba kwenguva. Munhu anosarudza sei algorithm chaiyo yekushandisa yakapihwa data set?
- 5. Kovariance uye kuwirirana zvinosiyana sei?
- 6. Mukudzidza muchina, kuungana kunorevei?
- 7. Chii chaunofarira muchina kudzidza algorithm?
- 8. Linear Regression mu Machine Kudzidza: Chii?
- 9. Tsanangura mutsauko uripo pakati peKNN nek-zvinoreva kubatanidza.
- 10. “Kurerekera pakusarudza” kunorevei kwauri?
- 11. Chii chaizvo chinonzi Bayes' Theorem?
- 12. Mumuchina Kudzidza Model, chii chinonzi 'training Set' uye 'test Set'?
- 13. Chii chinonzi Hypothesis muMuchina Kudzidza?
- 14. Chii chinorehwa nemichina yekuwedzera kukwirisa, uye inogona kudzivirirwa sei?
- 15. Chii chaizvo chinonzi Naive Bayes classifiers?
- 16. Chii chinonzi Cost Functions uye Loss Functions zvinorevei?
- 17. Chii chinosiyanisa generative modhi kubva kune inosarura?
- 18. Tsanangura kusiyana pakati pezvikanganiso zveType I neType II.
- 19. Mukudzidza muchina, ndeipi nzira yekudzidza yeEnsemble?
- 20. Chii chaizvo chinonzi parametric modhi? Ipa muenzaniso.
- 21. Rondedzera kusefa kwemubatanidzwa. Pamwe ne content-based kusefa?
- 22. Unorevei chaizvo neTime nhevedzano?
- 23. Tsanangura kusiyana pakati peGradient Boosting neRandom Forest algorithms.
- 24. Sei uchida kuvhiringidzika matrix? Chii?
- 25. Chii chaizvo chiri kuongorora chikamu chemusimboti?
- 26. Sei kutenderera kwechikamu kwakakosha kuPCA (principal component analysis)?
- 27. Kugarisa nguva nenguva kunosiyana sei?
- 28. Kugadzirisa uye kumisa zvakasiyana sei kubva kune imwe?
- 29. Ko “variance inflation factor” inorevei chaizvo?
- 30. Zvichienderana nehukuru hweseti yekudzidziswa, iwe unotora sei classifier?
- 31. Ndeipi algorithm mukudzidza kwemichina inonzi "mudzidzi ane usimbe" uye nei?
- 32. Chii chinonzi ROC Curve neAUC?
- 33. Chii chinonzi hyperparameters? Chii chinovaita vakasiyana kubva pamuenzaniso paramita?
- 34. Chii chinonzi F1 Score, rangarira, uye nemazvo zvinorevei?
- 35. Chii chaizvo chinonzi cross-validation?
- 36. Ngatiti waona kuti modhi yako ine musiyano wakakosha. Ndeipi algorithm, mumaonero ako, inonyanya kukodzera kubata iyi mamiriro?
- 37. Chii chinosiyanisa Ridge regression kubva Lasso regression?
- 38. Ndeipi inonyanya kukosha: kushanda kwemuenzaniso kana kunyatsoita muenzaniso? Ndeipi uye nei uchiifarira?
- 39. Ungagadzirisa sei dataset ine kusaenzana?
- 40. Ungasiyanisa sei pakati pekusimudza uye bhegi?
- 41. Tsanangura mutsauko uripo pakati pekudzidza kwekutepfenyura uye nekuderedza.
- mhedziso
Mabhizinesi ari kushandisa tekinoroji yekucheka-cheka, senge artificial intelligence (AI) uye kudzidza muchina, kuwedzera kuwanikwa kweruzivo uye masevhisi kune vanhu.
Tekinoroji idzi dziri kutambirwa nemaindasitiri akasiyana, anosanganisira emabhanga, emari, zvitoro, kugadzira, uye hutano.
Imwe yeanonyanya kutsvakwa-mushure mekuita kwesangano basa uchishandisa AI ndeye data masayendisiti, ekugadzira njere mainjiniya, mainjiniya ekudzidza muchina, uye vanoongorora data.
Ichi chinyorwa chinokutungamirira iwe kuburikidza nemhando dzakasiyana-siyana machine learning mibvunzo yekubvunzurudza, kubva kune yekutanga kusvika yakaoma, kuti ikubatsire kugadzirira chero mibvunzo yaunga bvunzwa paunenge uchitsvaga basa rako rakakodzera.
1. Tsanangura mutsauko uripo pakati pekudzidza nemuchina, njere dzekugadzira, uye kudzidza kwakadzama.
Artificial intelligence inoshandisa akasiyana ekudzidza muchina uye nzira dzekudzidza dzakadzama dzinobvumira masisitimu emakombuta kuita mabasa achishandisa njere dzakaita sedzemunhu ane pfungwa nemitemo.
Kudzidza kwemichina kunoshandisa nhamba dzakasiyana-siyana uye Deep Kudzidza nzira dzekugonesa michina kudzidza kubva mukuita kwayo kwekutanga uye kuve nehunyanzvi hwekuita mamwe mabasa ega pasina kutariswa nevanhu.
Kudzidza Kwakadzika muunganidzwa wealgorithms inobvumira iyo software kudzidza kubva kwayo uye kuita akasiyana ekutengesa mabasa, akadai sezwi uye kucherechedzwa kwemifananidzo.
Masisitimu anofumura ma multilayered neural networks kusvika kune yakawanda data yekudzidza vanokwanisa kuita zvakadzama kudzidza.
2. Ndapota tsanangurai marudzi akasiyana-siyana ekudzidza kwemichina.
Kudzidza kwemichina kunowanikwa mumhando nhatu dzakasiyana zvakanyanya:
- Kudzidzira Kunotariswa: Modhi inogadzira fungidziro kana mitongo ichishandisa dhata rakanyorwa kana renhoroondo mumuchina wakatariswa kudzidza. Maseti edata akamakwa kana kunyorwa kuti awedzere chirevo chawo anonzi data rakanyorwa.
- Kudzidza Kusina Kutariswa: Hatina dhata rakanzi rekudzidza kusina anotariswa. Mune data rinouya, modhi inogona kuwana mapatani, zvisinganzwisisike, uye zvinowirirana.
- Kusimbisa Kudzidza: Iyo modhi inogona dzidza nekushandisa kusimbisa kudzidza uye mibairo yayakawana pamaitiro ayo ekutanga.
3. Chii chinonzi bias maringe nekusiyana kwekutengeserana?
Overfitting mhedzisiro yekurerekera, inova iyo dhigirii iyo modhi inokodzera iyo data. Bias inokonzerwa nefungidziro isiriyo kana yakapusa mune yako muchina kudzidza algorithm.
Kusiyana kunoreva kukanganisa kunokonzerwa nekuoma mune yako ML algorithm, iyo inoburitsa kunzwa kune yakakura madhigirii ekusiyana mukudzidziswa data uye kuwandisa.
Kusiyana ndiko kuti yakawanda sei modhi inosiyana zvichienderana nekuiswa.
Mune mamwe mazwi, mhando dzekutanga dzakarerekera zvakanyanya asi dzakagadzikana (yakaderera musiyano). Kuwedzeredza idambudziko nemhando dzakaoma, kunyangwe zvakadaro dzinotora chokwadi chemuenzaniso (yakaderera kusarura).
Kuti udzivise mutsauko wepamusoro uye kurerekera kwepamusoro, kutengeserana pakati pekurerekera nekusiyana kunodiwa pakuderedza kukanganisa kwakanyanya.
4. Michina yekudzidza algorithms yakashanduka zvakanyanya nekufamba kwenguva. Munhu anosarudza sei algorithm chaiyo yekushandisa yakapihwa data set?
Iyo nzira yekudzidza yemuchina inofanirwa kushandiswa chete zvinoenderana nerudzi rwe data mune chaiyo dataset.
Kana data iri mutsara, kudzoreredzwa kwemutsara kunoshandiswa. Iyo yekubhegi nzira yaizoita zvirinani kana data yakaratidza isiri-mutsara. Tinogona kushandisa miti yesarudzo kana SVM kana iyo data ichifanira kuongororwa kana kududzirwa nekuda kwekutengesa.
Neural network inogona kubatsira kuwana mhinduro chaiyo kana dataset ichisanganisira mafoto, mavhidhiyo, uye odhiyo.
Sarudzo yealgorithm yeimwe mamiriro kana kuunganidzwa kwedata haigone kuitwa pachiyero chimwe chete.
Nechinangwa chekugadzira yakanakisa nzira, tinofanira kutanga taongorora data tichishandisa exploratory data analysis (EDA) tonzwisisa chinangwa chekushandisa dataset.
5. Kovariance uye kuwirirana zvinosiyana sei?
Covariance inoongorora kuti mabhii maviri akabatana sei kune mumwe nemumwe uye kuti munhu angachinja sei mukupindura shanduko mune imwe.
Kana chigumisiro chiri chakanaka, chinoratidza kuti pane hukama hwakananga pakati pezvipembenene uye kuti imwe inogona kusimuka kana kuderera nekuwedzera kana kuderera kwechigadziko chepasi, tichifunga kuti mamwe mamiriro ose anoramba aripo.
Correlation inoyera chinongedzo pakati pezviviri zvisina kurongeka uye ine matatu chete akasiyana maitiro: 1, 0, uye -1.
6. Mukudzidza muchina, kuungana kunorevei?
Nzira dzekudzidza dzisina anotariswa idzo dzinoisa data mumapoka dzinonzi clustering. Nekuunganidzwa kwe data point, iyo clustering tekinoroji inogona kushandiswa.
Iwe unogona kuunganidza ese e data mapoinzi zvinoenderana nemabasa avo uchishandisa zano iri.
Iwo maficha uye hunhu hwemapoinzi edata anowira muchikamu chimwe chete akafanana, nepo iwo emapoinzi edata anowira mumapoka akasiyana akasiyana.
Iyi nzira inogona kushandiswa kuongorora data yenhamba.
7. Chii chaunofarira muchina kudzidza algorithm?
Iwe une mukana wekuratidza zvaunofarira uye matarenda akasarudzika mumubvunzo uyu, pamwe neruzivo rwako rwakakwana rwemakina mazhinji ekudzidza muchina.
Heano mashoma akajairika muchina kudzidza algorithms yekufunga nezvayo:
- Kudzoreredza kwemutsara
- Kugadziriswa kwemagetsi
- Naïve Bayes
- Zvisarudzo miti
- K zvinoreva
- Random sango algorithm
- K-muvakidzani wepedyo (KNN)
8. Linear Regression mu Machine Kudzidza: Chii?
Iyo inotariswa yemuchina yekudzidza algorithm ndeye mutsara regression.
Inoshandiswa mukuongorora kwekufungidzira kuti ione kuwirirana kwemutsara pakati pezvinoenderana uye zvakazvimiririra zvakasiyana.
Linear regression's equation yakaita seinotevera:
Y = A + BX
kupi:
- Iyo yekupinda kana yakazvimirira vhezheni inonzi X.
- Iyo inotsamira kana kubuda inosiyana ndeye Y.
- X's coefficient is b, uye kupindira kwayo ndia.
9. Tsanangura mutsauko uripo pakati peKNN nek-zvinoreva kubatanidza.
Musiyano wekutanga ndewekuti KNN (nzira yekuisa mumapoka, kudzidza kunotariswa) inoda mapoinzi akanyorwa nepo k-zvinoreva zvisingaite (clustering algorithm, kudzidza kusingatarisirwe).
Unogona kuisa data rakanyorwa munzvimbo isina kunyorwa uchishandisa K-Nearest Neighbours. K-kureva kubatanidza kunoshandisa avhareji chinhambwe pakati pemapoinzi kudzidza kuunganidza mapoinzi asina kunyorwa.
10. “Kurerekera pakusarudza” kunorevei kwauri?
Kurerekera muchikamu chesampling yechiyedzo kunokonzerwa nekusarongeka kwehuwandu.
Rimwe boka remuenzaniso rinosarudzwa kakawanda kupfuura mamwe mapoka mukuedza nekuda kwekusarurama.
Kana sarudzo ikasabvumwa, inogona kuguma nemhedzisiro isiriyo.
11. Chii chaizvo chinonzi Bayes' Theorem?
Kana tave kuziva zvimwe zvingangoitika, tinokwanisa kuona mukana wekushandisa Bayes 'Theorem. Inopa mukana wepashure wechiitiko chinoenderana neruzivo rwekare, nemamwe mazwi.
Nzira inonzwika yekufungidzira zvingangoitika inopiwa nedzidziso iyi.
Paunenge uchigadzira kupatsanura kufanotaura matambudziko ekuenzanisa uye kuenzanisa modhi kune yekudzidziswa dataset mukudzidza muchina, Bayes' theorem inoshandiswa (kureva Naive Bayes, Bayes Optimal Classifier).
12. Mumuchina Kudzidza Model, chii chinonzi 'training Set' uye 'test Set'?
Seti yekudzidzira:
- Iyo yekudzidziswa seti ine zviitiko zvinotumirwa kune modhi yekuongorora uye kudzidza.
- Iyi ndiyo data yakanyorwa iyo ichashandiswa kudzidzisa modhi.
- Kazhinji, 70% yedata rese rinoshandiswa sedhata rekudzidzisa.
Test Set:
- Iyo test set inoshandiswa kuongorora iyo modhi ye hypothesis chizvarwa chechokwadi.
- Isu tinoyedza tisina dhata rakanyorwa uye toshandisa mavara kuratidza zvabuda.
- Iyo 30% yasara inoshandiswa seyese dataset.
13. Chii chinonzi Hypothesis muMuchina Kudzidza?
Kudzidza kweMichina kunogonesa kushandiswa kwemaseti aripo kuti anzwisise zviri nani basa rakapihwa rinobatanidza kuisa kune zvinobuda. Izvi zvinozivikanwa sekushanda approximation.
Muchiitiko ichi, fungidziro inofanirwa kushandiswa kune isingazivikanwe chinangwa chebasa kuendesa zvese zvinofungidzirwa zvinoonekwa zvichibva pane yakapihwa mamiriro nenzira yakanakisa inogoneka.
Mukudzidza kwemuchina, fungidziro imodhi inobatsira mukufungidzira basa rinotarirwa uye kuzadzisa kwakakodzera kupinza-kune-kubuda mepu.
Kusarudzwa uye dhizaini yealgorithms inobvumira tsananguro yenzvimbo yezvingangofungidzirwa zvinogona kumiririrwa nemuenzaniso.
Pafungidziro imwe chete, mavara madiki h (h) anoshandiswa, asi capital h (H) inoshandiswa pafungidziro yese nzvimbo iri kutsvagwa. Tichaongorora zvinyorwa izvi zvishoma:
- A hypothesis (h) imwe modhi inofambisa mepu yekupinda kune zvinobuda, iyo inogona kuzoshandiswa pakuongorora uye kufanotaura.
- A hypothesis set (H) inzvimbo inotsvakwa yefungidziro inogona kushandiswa kugadzira mepu kune zvinobuda. Mamiriro enyaya, modhi, uye magadzirirwo emhando mimwe mienzaniso mishoma yezvipimo zvegeneric.
14. Chii chinorehwa nemichina yekuwedzera kukwirisa, uye inogona kudzivirirwa sei?
Kana muchina uchiedza kudzidza kubva kune isina kukwana dataset, overfitting inoitika.
Nekuda kweizvozvo, overfitting iri inversely yakabatana nehuwandu hwe data. Iyo yekuchinjisa-yekusimbisa nzira inobvumira overfitting kudziviswa kune madiki dataset. Dataset yakakamurwa kuita zvikamu zviviri nenzira iyi.
Iyo dataset yekuyedza uye yekudzidziswa ichave nezvikamu zviviri izvi. Iyo dataset yekudzidzira inoshandiswa kugadzira modhi, nepo dhatabheti rekuyedza rinoshandiswa kuongorora modhi uchishandisa akasiyana ekuisa.
Iyi ndiyo nzira yekudzivirira overfitting.
15. Chii chaizvo chinonzi Naive Bayes classifiers?
Dzakasiyana siyana nzira dzinogadzira iyo Naive Bayes classifiers. Seti yemaalgorithms anozivikanwa seaya maclassifiers ese anoshanda pane imwechete yakakosha pfungwa.
Iko fungidziro yakaitwa nevasina ruzivo veBayes classifiers ndeyekuti kuvepo kwechimwe chinhu kana kusavapo hakuna chekuita pakuvapo kana kusavapo kwechimwe chimiro.
Mune mamwe mazwi, izvi ndizvo zvatinodaidza se "naive" sezvo zvichiita fungidziro yekuti dhatabheti yega yega yakakosha uye yakazvimirira.
Classification inoitwa uchishandisa naive Bayes classifiers. Izvo zviri nyore kushandisa uye kuburitsa zvirinani mhedzisiro pane zvakanyanya kuomarara zvinofanotaura kana nzvimbo yekuzvimiririra ichokwadi.
Mukuongorora zvinyorwa, kusefa spam, uye masisitimu ekurudziro, anoshandiswa.
16. Chii chinonzi Cost Functions uye Loss Functions zvinorevei?
Mutsara wekuti "kurasikirwa nebasa" unoreva maitiro ekurasikirwa kwekombuta kana chidimbu chimwe chete che data chichiverengerwa.
Zvakasiyana, isu tinoshandisa mutengo wekuita kuona huwandu hwemhosho yedata rakawanda. Hapana musiyano unokosha uripo.
Mune mamwe mazwi, nepo mutengo unoshanda uchibatanidza mutsauko weiyo yese dataset yekudzidziswa, kurasikirwa mabasa akagadzirirwa kutora mutsauko pakati peiyo chaiyo uye yakafanotaurwa kukosha kune imwechete rekodhi.
17. Chii chinosiyanisa generative modhi kubva kune inosarura?
Modhi yerusarura inodzidza misiyano pakati peakawanda data data. A generative modhi inotora pamhando dzakasiyana dze data.
Pamatambudziko ekuronga, mhando dzerusarura dzinowanzopfuura mamwe mamodheru.
18. Tsanangura kusiyana pakati pezvikanganiso zveType I neType II.
Nhema dzenhema dzinowira pasi pechikamu cheType I zvikanganiso, nepo zvisiri izvo zvinoenda pasi peType II zvikanganiso (zvichinzi hapana chakaitika kana chanyatsoita).
19. Mukudzidza muchina, ndeipi nzira yekudzidza yeEnsemble?
Maitiro anonzi ensemble kudzidza anosanganisa akawanda emuchina ekudzidza modhi kuti abudise mamwe ane simba modhi.
Muenzaniso unogona kusiyanisa nekuda kwezvikonzero zvakasiyana. Zvikonzero zvakawanda ndezvi:
- Vanhu Vakasiyana-siyana
- Various Hypotheses
- Nzira dzakasiyana-siyana dzekuenzanisira
Isu tichasangana nenyaya tichishandisa iyo modhi yekudzidzira uye yekuyedza data. Rusarura, musiyano, uye kukanganisa kusingadzoreki ndiwo marudzi anogoneka echikanganiso ichi.
Zvino, isu tinodaidza iyi chiyero pakati pekurerekera uye musiyano mumuenzaniso kutengeserana-kusiyana-siyana, uye inofanira kugara iripo. Kuchinjana uku kunoitwa kuburikidza nekushandisa ensemble kudzidza.
Kunyangwe paine akasiyana ensemble maitiro aripo, kune maviri akajairwa mazano ekubatanidza akawanda mamodheru:
- Nzira yekuzvarwa inonzi bagging inoshandisa seti yekudzidziswa kugadzira mamwe maseti ekudzidziswa.
- Kusimudzira, imwe nzira yakaomesesa: Zvakafanana nekubhegi, kukwidziridza kunoshandiswa kutsvaga yakanakira uremu fomula yekuseti yekudzidzisa.
20. Chii chaizvo chinonzi parametric modhi? Ipa muenzaniso.
Kune huwandu hushoma hwema parameter mune parametric modhi. Kufanotaura data, zvese zvaunoda kuziva ndiwo maparamita emuenzaniso.
Iyi inotevera mienzaniso yakajairika: logistic regression, linear regression, uye linear SVMs. Non-parametric modhi inoshanduka sezvo ichigona kuve nenhamba isina muganho yemaparamita.
Iyo modhi maparamendi uye chimiro che data rakacherechedzwa zvinodiwa pakufanotaura kwedata. Heino mimwe mienzaniso yakajairika: misoro mienzaniso, miti yesarudzo, uye k-vavakidzani vepedyo.
21. Rondedzera kusefa kwemubatanidzwa. Pamwe ne content-based kusefa?
Nzira yakaedzwa-uye-yechokwadi yekugadzira mazano anoenderana nekusefa kwekubatana.
Imwe nzira yekurudziro inodaidzwa kuti collaborative filtering inofanotaura zvinhu zvitsva nekuenzanisa zvido zvevashandisi nezvido zvakagovaniswa.
Sarudzo dzemushandisi ndicho chinhu chega chinotariswa nemukati-based recommender system. Tichifunga nezvesarudzo dzekare dzemushandisi, kurudziro nyowani dzinopihwa kubva kune zvine chekuita nazvo.
22. Unorevei chaizvo neTime nhevedzano?
A time series muunganidzwa wenhamba muhurongwa hwekukwira. Pamusoro penguva yakatarwa, inotarisisa mafambiro eakasarudzwa data mapoinzi uye nguva nenguva inotora data data.
Iko hakuna hushoma kana huwandu hwenguva yekuisa yenguva yakatevedzana.
Nguva dzakatevedzana dzinowanzo shandiswa nevaongorori kuongorora data zvichienderana nezvavanoda zvakasiyana.
23. Tsanangura kusiyana pakati peGradient Boosting neRandom Forest algorithms.
Random Sango:
- Nhamba huru yemiti yesarudzo inosanganiswa pamwe chete kumagumo uye inozivikanwa semasango asina kurongeka.
- Nepo kuwedzera kwe gradient kuchigadzira muti wega wega wakazvimiririra kubva kune mumwe, sango risingaite rinovaka muti wega wega panguva.
- Multiclass kuona kwechinhu inoshanda zvakanaka nemasango asina kurongeka.
Kuwedzera Gradient:
- Nepo Masango Masango achijoinha miti yesarudzo pakupera kwemaitiro, Gradient Boosting Machines inovasanganisa kubva pakutanga.
- Kana ma paramita akagadziridzwa nenzira kwayo, gradient inokwidziridza inodarika masango asina kurongeka maererano nemhedzisiro, asi haisi sarudzo yakangwara kana iyo data seti iine zvakawanda zvekubuda, anomalies, kana ruzha sezvo zvichigona kuita kuti modhi ive yakawandisa.
- Kana paine data isina kuenzana, sezvazviri mune chaiyo-nguva yekuongorora njodzi, gradient inosimudzira inoita nemazvo.
24. Sei uchida kuvhiringidzika matrix? Chii?
Tafura inozivikanwa seconfusion matrix, dzimwe nguva inozivikanwa seye kukanganisa matrix, inoshandiswa zvakanyanya kuratidza kuti mhando yemhando, kana classifier, inoita sei paseti yedata rebvunzo rinozivikanwa hunhu chaihwo.
Inotigonesa kuona kuti modhi kana algorithm inoita sei. Zvinoita kuti zvive nyore kwatiri kuona kusanzwisisana pakati pezvidzidzo zvakasiyana.
Inoshanda senzira yekuongorora kuti modhi kana algorithm inoitwa sei.
Mafungiro emhando yemhando yemhando anounganidzwa kuita nyonganiso matrix. Nhamba yekuverenga yekirasi yega yega yakashandiswa kuparura huwandu hwehuwandu hwekufanotaura kwakarurama uye kusina kururama.
Inopa ruzivo pamusoro pekukanganisa kwakaitwa nemugadziri pamwe nemhando dzakasiyana dzekukanganisa kunokonzerwa ne classifiers.
25. Chii chaizvo chiri kuongorora chikamu chemusimboti?
Nekudzikisa huwandu hwemhando dzakasiyana-siyana dzakabatanidzwa kune imwe neimwe, chinangwa ndechekuderedza chiyero chekuunganidza data. Asi zvakakosha kuchengetedza kusiyana zvakanyanya sezvinobvira.
Izvo zvakasiyana-siyana zvinoshandurwa kuita seti nyowani yezvakasiyana zvinonzi principal components.
Aya maPC ane orthogonal sezvo ari covariance matrix's eigenvectors.
26. Sei kutenderera kwechikamu kwakakosha kuPCA (principal component analysis)?
Kutenderera kwakakosha muPCA nekuti inogonesa kupatsanurwa pakati pemisiyano inowanikwa nechikamu chimwe nechimwe, zvichiita kuti chikamu chekududzira chive nyore.
Tinoda zvikamu zvakawedzerwa kuratidza mutsauko wechikamu kana zvikamu zvisina kutenderedzwa.
27. Kugarisa nguva nenguva kunosiyana sei?
Normalization:
Data inochinjwa panguva normalization. Iwe unofanirwa kugadzirisa iyo data kana iine zviyero zvakasiyana zvakanyanya, kunyanya kubva pasi kusvika kumusoro. Rongedza koramu yega yega kuitira kuti nhamba dzakakosha dzienderane.
Kuti uone kuti hapana kurasikirwa kwechokwadi, izvi zvinogona kubatsira. Kuona chiratidzo uchiregeredza ruzha ndicho chimwe chezvinangwa zvekudzidzira modhi.
Pane mukana wekuwandisa kana modhi yakapihwa kutonga kwakazara kuderedza kukanganisa.
Regularization:
Mukugadzirisa, basa rekufungidzira rinogadziriswa. Izvi zviri pasi pehumwe kutonga kuburikidza nekugadzirisa, izvo zvinofarira zviri nyore kuita mabasa pane akaomarara.
28. Kugadzirisa uye kumisa zvakasiyana sei kubva kune imwe?
Iwo maviri anonyanya kushandiswa matekiniki ekuyera chimiro ndeye normalization uye standardization.
Normalization:
- Kudzoreredza iyo data kuti ikwane [0,1] renji inozivikanwa seyakajairika.
- Kana ese maparamendi achifanira kunge aine chikero chakafanana, normalization inobatsira, asi iyo data seti yekubuda inorasika.
Regularization:
- Dhata inodzokororwa kuti ive nerevo ye0 uye yakajairwa kutsauka kwe1 sechikamu cheiyo standardization process (Unit musiyano)
29. Ko “variance inflation factor” inorevei chaizvo?
Huyero hwemusiyano wemodhi kune musiyano weiyo modhi ine imwe chete yakazvimirira shanduko inozivikanwa sevariation inflation factor (VIF).
VIF inofungidzira kuwanda kwemulticollinearity iripo mune seti yeanoverengeka regression variables.
Kusiyana kweiyo modhi (VIF) Modhi ine Imwe Yakazvimiririra Yakasiyana Variance
30. Zvichienderana nehukuru hweseti yekudzidziswa, iwe unotora sei classifier?
Kurerekera kwepamusoro, modhi yemhando yakaderera inoita zvirinani kune pfupi kudzidziswa seti sezvo kufutisa kuri kushoma. Naive Bayes mumwe muenzaniso.
Kuti umiririre kupindirana kwakaoma kweseti yakakura yekudzidziswa, modhi ine kurerekera kwakaderera uye mutsauko wepamusoro unodiwa. Logistic regression muenzaniso wakanaka.
31. Ndeipi algorithm mukudzidza kwemichina inonzi "mudzidzi ane usimbe" uye nei?
Mudzidzi ane usimbe, KNN muchina wekudzidza algorithm. Nekuti K-NN ine simba inoverengera chinhambwe nguva yega yega yainoda kurongedza pane kudzidza chero muchina-akadzidzwa makoshero kana akasiyana kubva kudhata rekudzidzisa, inorangarira dhata rekudzidzisa.
Izvi zvinoita kuti K-NN ave mudzidzi ane usimbe.
32. Chii chinonzi ROC Curve neAUC?
Kuita kwemhando yemhando pazvikumbaridzo zvese kunomiririrwa nemifananidzo neROC curve. Iine chiyero chechokwadi chechokwadi uye chenhema chiyero chechiyero chemaitiro.
Zvichitaurwa zviri nyore, nzvimbo iri pasi peROC curve inozivikanwa seAUC (Nzvimbo Pasi peROC Curve). Iyo ROC curve's two-dimensional nharaunda kubva (0,0) kuenda kuAUC inoyerwa (1,1). Pakuongorora mabhinari emhando yemhando, inoshandiswa sehuwandu hwekuita.
33. Chii chinonzi hyperparameters? Chii chinovaita vakasiyana kubva pamuenzaniso paramita?
Kuchinja kwemukati kwemuenzaniso kunozivikanwa semuenzanisi parameter. Uchishandisa ruzivo rwekudzidzisa, kukosha kweparameter inofananidzwa.
Zvisingazivikanwe kune modhi, hyperparameter inoshanduka. Iko kukosha hakugone kutsanangurwa kubva kune data, saka ivo vanowanzo shandiswa kuverenga modhi paramita.
34. Chii chinonzi F1 Score, rangarira, uye nemazvo zvinorevei?
Iyo nyonganiso Measure ndiyo metric inoshandiswa kuyera kushanda kweiyo yemhando modhi. Aya anotevera mitsara anogona kushandiswa kutsanangura zvirinani kuvhiringidzika metric:
TP: Chokwadi Positives - Aya ndiwo maitiro akanaka aitarisirwa nemazvo. Zvinoratidza kuti hunhu hwekirasi yakatarwa uye kirasi chaiyo zvese zvakanaka.
TN: Chokwadi Negatives- Aya ndiwo maitiro akashata ayo akafanotaurwa nemazvo. Inoratidza kuti zvose zvakakosha zvekirasi chaiyo uye kirasi inotarisirwa hazvina kunaka.
Aya maitiro - emanyepo uye manyepo asina kunaka - anoitika kana kirasi yako chaiyo inosiyana nekirasi inotarisirwa.
zvino,
Chiyero cheyero yechokwadi yakanaka mwero (TP) kune zvese zvakacherechedzwa zvakaitwa mukirasi chaiyo inonzi kurangarira, inozivikanwawo sekunzwa.
Kuyeuka ndeye TP/(TP+FN).
Kururamisa chiyero chehutano hwakanaka hwekufanotaura, iyo inoenzanisa nhamba yezvakanaka iyo muenzaniso inonyatsofanotaura kuti ingani yakarurama inofanotaura.
Precision is TP/(TP + FP)
Iyo iri nyore kuita metric yekunzwisisa ndeyechokwadi, inongori chikamu chezvakafanotaurwa zvinoonekwa kune zvese zvinoonekwa.
Kururama kwakaenzana ne(TP+TN)/(TP+FP+FN+TN).
Precision uye Recall zvakayerwa uye zvakayerwa kupa iyo F1 Score. Nekuda kweizvozvo, chibodzwa ichi chinotarisa zvese zviri zviviri manyepo uye manyepo enhema.
F1 inowanzokosha kupfuura huroyi, kunyanya kana uine kusaenzana kwekirasi kugovera, kunyangwe kana intuitively isiri nyore kunzwisisa sekurongeka.
Kururama kwakanakisisa kunowanikwa apo mari yezvinyorwa zvenhema uye zvisizvo zvenhema zvinofananidzwa. Zviri nani kusanganisa zvese Precision uye Recall kana mitengo inosanganisirwa nenhema yakanaka uye nhema dzisina kunaka dzakasiyana zvakanyanya.
35. Chii chaizvo chinonzi cross-validation?
Nzira yekuongororazve nhamba inonzi kuyambuka-kusimbisa mukudzidza kwemuchina inoshandisa akati wandei dataset kudzidzisa uye kuongorora muchina wekudzidza algorithm pane akati wandei.
Batch nyowani yedata iyo isina kushandiswa kudzidzisa modhi inoedzwa uchishandisa muchinjiko-kusimbisa kuona kuti iyo modhi inofanotaura sei. Data overfitting inodzivirirwa kuburikidza nemuchinjiko-kusimbisa.
K-Peta Iyo inonyanya kushandiswa nzira yekuenzanisa zvakare inotsemura dataset rese kuita K seti dzakaenzana. Inonzi cross-validation.
36. Ngatiti waona kuti modhi yako ine musiyano wakakosha. Ndeipi algorithm, mumaonero ako, inonyanya kukodzera kubata iyi mamiriro?
Kugadzirisa kusiyanisa kwakanyanya
Isu tinofanirwa kushandisa nzira yekubhegi kune matambudziko nekusiyana kwakakura.
Kudzokororwa sampling yedata isina kurongeka yaizoshandiswa neiyo bagging algorithm kugovera iyo data mumapoka madiki. Kana iyo data yave yakakamurwa, isu tinogona kushandisa zvisina tsarukano data uye chaiyo yekudzidzisa maitiro kugadzira mitemo.
Mushure meizvozvo, kuvhota kwaigona kushandiswa kubatanidza kufanotaura kwemuenzaniso.
37. Chii chinosiyanisa Ridge regression kubva Lasso regression?
Nzira mbiri dzinoshandiswa zvakanyanya dzekugara dziri Lasso (inonziwo L1) uye Ridge (dzimwe nguva inonzi L2) regression. Iyo inoshandiswa kudzivirira kuwandisa kwedata.
Kuti uwane mhinduro yakanakisisa uye kuderedza kuoma, maitiro aya anoshandiswa kuranga coefficients. Nekuranga huwandu hwehuwandu hwehuwandu hwema coefficients, iyo Lasso regression inoshanda.
Basa rechirango muRidge kana L2 regression rinotorwa kubva muhuwandu hwemakona emacoefficients.
38. Ndeipi inonyanya kukosha: kushanda kwemuenzaniso kana kunyatsoita muenzaniso? Ndeipi uye nei uchiifarira?
Uyu mubvunzo unonyengera, saka munhu anofanira kutanga anzwisisa kuti Model Performance chii. Kana kushanda kuchitsanangurwa sekukurumidza, saka kunovimba nemhando yekushandisa; chero application inosanganisira chaiyo-nguva mamiriro ingada kumhanya kwakanyanya sechinhu chakakosha.
Semuyenzaniso, iwo akanakisa Kutsvaga Mhedzisiro anozove mashoma kukosha kana Query mhinduro yakatora nguva yakareba kuti isvike.
Kana Performance ikashandiswa sechikonzero chekuti nei kunyatso uye kuyeuka kuchifanira kuiswa pamberi pechokwadi, ipapo F1 mamakisi anozonyanya kubatsira pane kurongeka mukuratidza bhizinesi kesi kune chero data set isina kuenzana.
39. Ungagadzirisa sei dataset ine kusaenzana?
Iyo isina kuenzana dataset inogona kubatsirwa nesampling matekiniki. Sampling inogona kuitwa nenzira yepasi kana yakawandisa.
Pasi peSampling inotibvumira kudzikisa saizi yekirasi yeruzhinji kuti ienderane nekirasi yevashoma, iyo inobatsira mukuwedzera kukurumidza maererano nekuchengetedza uye kumhanya-nguva kuuraya asi zvinogona zvakare kukonzera kurasikirwa kwe data rakakosha.
Kuti tigadzirise nyaya yekurasika kweruzivo kunokonzerwa nekuwedzeredza, isu upsample the Minority class; zvisinei, izvi zvinotiita kuti timhanye mumatambudziko akawandisa.
Mamwe mazano anosanganisira:
- Cluster-Yakavakirwa Pamusoro PeSampling- Iwo mashoma uye mazhinji makirasi makirasi anoiswa ega pasi peK-nzira yekuunganidza nzira mune ino mamiriro. Izvi zvinoitwa kuti uwane dataset clusters. Zvadaro, sumbu rimwe nerimwe rinowedzerwa kuitira kuti makirasi ose ave nehukuru hwakafanana uye mapoka ose mukati mekirasi ane nhamba yakaenzana yezviitiko.
- SMOTE: Synthetic Minority Over-sampling Technique- Chidimbu chedhata kubva kukirasi yevashoma chinoshandiswa semuenzaniso, mushure mezvo mamwe maekiseni ekuwedzera anofananidzwa nawo anogadzirwa uye akawedzerwa kune yekutanga dataset. Iyi nzira inoshanda zvakanaka nenhamba dze data point.
40. Ungasiyanisa sei pakati pekusimudza uye bhegi?
Ensemble Techniques ine shanduro dzinozivikanwa sekubhegi uye kuwedzera.
Bagging-
Kune maalgorithms ane musiyano wepamusoro, bagging inzira inoshandiswa kudzikisa musiyano. Imwe mhuri yakadai yevadzidzisi inogara yakarerekera ndiyo mhuri yemuti wesarudzo.
Rudzi rwe data iyo miti yesarudzo inodzidziswa ine yakakosha pakuita kwavo. Nekuda kweizvi, kunyangwe nekukwenenzverwa kwepamusoro-soro, kuwanda kwemhedzisiro dzimwe nguva kwakaoma zvakanyanya kuwana mazviri.
Kana data yekudzidziswa kwemiti yesarudzo ikashandurwa, mhedzisiro yacho inosiyana zvakanyanya.
Nekuda kweizvozvo, mabhegi anoshandiswa, umo miti yakawanda yesarudzo inogadzirwa, imwe neimwe inodzidziswa uchishandisa sampu ye data rekutanga, uye mhedzisiro ndiyo avhareji yeese akasiyana mamodheru.
Kukurudzira:
Boosting ndiyo dhizaini yekufanotaura nen-isina simba classifier system umo imwe neimwe isina kusimba classifier inogadzira kushomeka kweiyo yakasimba classifiers. Isu tinotaura nezve classifier inoita zvakashata pane yakapihwa data seti "isina simba classifier."
Kusimudzira zviri pachena maitiro kwete algorithm. Logistic regression uye isina kudzika sarudzo miti mienzaniso yakajairika yevasina kusimba classifiers.
Adaboost, Gradient Boosting, uye XGBoost ndiwo maviri anonyanya kufarirwa algorithms, zvisinei, kune akawanda akawanda.
41. Tsanangura mutsauko uripo pakati pekudzidza kwekutepfenyura uye nekuderedza.
Paunenge uchidzidza nemuenzaniso kubva kune yakacherechedzwa mienzaniso, modhi inoshandisa inductive kudzidza kusvika pamhedziso yakajairika. Nekune rimwe divi, nekudzidza kwekubvisa, modhi inoshandisa mhedzisiro isati yagadzira yayo.
Inductive learning inzira yekukwevera mhedziso kubva pane zvakaonekwa.
Deductive learning inzira yekugadzira zvinocherechedzwa zvichibva pane zvinofungidzirwa.
mhedziso
Makorokoto! Iyi ndiyo yepamusoro makumi mana uye pamusoro mibvunzo yekubvunzurudza yekudzidza muchina iyo iwe yave kuziva mhinduro kwairi. Data sainzi uye chakagadzirwa njere mabasa acharamba achidiwa sezvo tekinoroji inofambira mberi.
Vanokwikwidza vanovandudza ruzivo rwavo rweaya ekucheka-kumucheto matekinoroji uye nekuvandudza hunyanzvi hwavo seti vanogona kuwana akasiyana siyana ekuita basa nemubhadharo wemakwikwi.
Iwe unogona kuenderera mberi nekupindura kubvunzurudza ikozvino iwe uine nzwisiso yakasimba yekuti ungapindura sei kune yakanyanyo bvunzwa mibvunzo yekubvunzurudza muchina.
Zvichienderana nezvinangwa zvako, tora danho rinotevera. Gadzirira kubvunzurudzwa nekushanyira Hashdork's Hurukuro Series.
Leave a Reply