InoteveraGaidhi rinotevera
Tokenization uye Byte Pair Encoding
Tekinoroji
Nhungamiro yehunyanzvi
Target encoding inotsiva chikamu chine pfupiso yenhamba yemhedzisiro inocherechedzwa yechikamu ichocho mune inobvumidzwa data yekudzidziswa.
Teerera kugwaro iriAI-yakarondedzerwa · ~3 min
Inogona kuita zvemukati zvemukati zvine akawanda akasiyana hunhu kuve nyore kuenzanisira, asi zvinoda kunyatsopatsanurwa kwemavara ekudzidzisa kudzivirira kuvuza.
Categorical features inotsanangura mapoka akadai semadhipatimendi, vatengesi kana mhando dzechigadzirwa. Imwe-inopisa encoding inogadzira chiratidzo chakasiyana chechikamu chimwe nechimwe. Nezvikamu zvakawanda, chinangwa chekodhi chinopa chimwe chinomiririra: shandura chikamu chega chega nechidimbu chemhedzisiro, senge kunonoka kwekudzoreredza kana chiyero chakanaka chechikamu chebhinari. Chiratidzo chinobatsira ndichowo chinokonzera njodzi. Fungidzira chikamu chinooneka kamwe chete. Chikwata chayo chisina kunyorova zvinoreva kuti mhedzisiro yemusara. Kana kukosha uku kukava mupiro wekufanotaura mutsara mumwe chete, modhi iri kupihwa mhinduro. Kuita kwakasimba kwekudzidzira mune iyi setup kunogona kunyangarika pane nyowani data. Cross-fitting inobatsira kugadzira yakachengeteka dzidziso inomiririra. Kamura dhata rekudzidzisa kuita mapeta, verenga pfupiso dzechikamu uchishandisa mamwe mapeji, uye encode iyo yakabatwa-yakapetwa neiyo pfupiso. Dzokorora kusvikira mutsara wega wega wekudzidzisa une encoding yakavakwa isina mavara ewayo. Iyo yepasi modhi inodzidza kubva pane izvi zvinomiririra. Panguva yekuongorora kana kufanotaura, shandisa mepu dzakadzidzwa kubva kune inobvumidzwa data yekudzidziswa. Kutsvedza kunoderedza pesvedzero yezvikamu zvine mashoma echerechedzo nekukwevera zvipfupiso zvadzo kune yepasi rose dzidziso zvinoreva. Nzvimbo isingawanzo dhipodhi haifanirwe kugamuchira yakanyanyisa encoding chete nekuda kwekuuya kumwe chete kusina kujairika. Huwandu hwekutsvedza isarudzo yekuenzanisira inoda kusimbiswa. Scikit-learn's TargetEncoder inonyora musiyano wakakosha: fit_transform inoshandisa yemukati muchinjika-yakakodzera, uku ichikodzera uyezve kushandura iyo yakafanana dzidziso data haipe iyo yakaenzana dziviriro. Usafunge kuti raibhurari yese inoshandisa maitiro akafanana. Chengetedza encoder mukati mepiipi yekuongorora, tsanangura mabatirwo anoitwa zvikamu zvisingaonekwe, uye chinja nzira yekupatsanura apo kurongeka kwenguva kana zvinhu zvinodzokororwa zvichiita kuti zvakajairika zvakapetwa zvisakodzeri.
Modhi yekutumira inomiririra dhipatimendi ichishandisa nhoroondo yayo yekunonoka kunonoka, yakaverengerwa kubva kudhata rekudzidziswa. Iwo mamepu akafanana akadzidzwa anobva ashandiswa kumarekodhi ekuongorora pasina kutarisa zvabuda.
Muchirevo chekufungidzira kutsetseka, chikamu chine zvibereko zviviri zve 10 uye 20 zvinosanganiswa ne4 zvakamboonekwa pane imwe nzvimbo yepasi rose ye 6. The smoothed value is (30 plus 24) yakakamurwa nevatanhatu, kana 9.
Kupeta kwekudzidzira kune chikamu chinoonekwa mumutsara mumwe chete. Encoding iyo mutsara kubva kune yayo pachayo yaizoratidza zita rayo, saka timu inovaka inomiririra yekudzidziswa kubva kune mamwe mapeta.
Muongorori anoshandisa scikit-learn's TargetEncoder muPipeline. Ivo vanotarisa maitiro ayo akanyorwa-muchinjika-akakodzera pane kufunga kuti kukwana kunoteverwa neshanduko kwakaenzana nefit_transform.
Categorical encoders icharamba ichibatsira uko data rekushanda rine nhamba huru dzekuchinja zviziviso. Mapoka anogona kuvandudza kuvimbika nekutarisa zvikamu zvitsva, zvisingawanzo zvikwata uye shanduko mumhedzisiro yavo maitiro. Ivo vanofanirwawo kuchengetedza iyo yekudzidzira cutoff uye encoder vhezheni padivi pemhando yega yega yakatumirwa, saka fungidziro inogona kuteverwa kune mepu inoshandiswa. Otomatiki mapaipi anogona kuita kuti kuvhuvhuta kuite nyore, asi yakakosha dhizaini sarudzo inoramba iri yemunhu: tarisa kuti ndedzipi mhedzisiro dzaivepo kana fungidziro ingadai yaitwa, uye ive nechokwadi chese chinotorwa chinoremekedza muganho iwoyo.
Tsamba yemahara
Nhatu dzakasimbiswa dzeAI nyaya mangwanani ega ega pakati pevhiki, dzakanyorwa neChirungu chakajeka. Yemahara zvachose, hapana ads.
Imwe email zuva rega rega revhiki. Bvisa kunyoresa kamwe chete. Hatimbotengese kana kugovera kero yako.
Zviedze iwe pachako
Mhinduro yepakarepo pamhinduro yega yega, uye chitupa chinogovaniswa chine ID inobatika kana wapedza kosi.
Tsigira dzidzo yemahara yeAI. AI Understanding ndeye 501(c)(3) isingabatsiri - hapana ads, hapana paywall, zvachose. Ita chipo
Target encoding inotsiva chikamu chine pfupiso yenhamba yemhedzisiro inocherechedzwa yechikamu ichocho mune inobvumidzwa data yekudzidziswa. Inogona kuita zvemukati zvemukati zvine akawanda akasiyana hunhu kuve nyore kuenzanisira, asi zvinoda kunyatsopatsanurwa kwemavara ekudzidzisa kudzivirira kuvuza.
Nekutarisa kumwe chete, chikamu chinoreva kuenzana chayakananga, ichiburitsa mhinduro mune inopinza.
Mamwe mapeta anopa nhamba dzechikamu kuitira kuti mavara akabatwa-kunze asatsanangure kumiririra kwayo.
Nhamba inosvika makumi matatu kuwedzera ina katanhatu, kana 54. Denominator imbiri kuwedzera ina, kana matanhatu. Mhedzisiro ipfumbamwe.
Mienzaniso midiki inogona kuburitsa kusagadzikana kwakanyanyisa, uye kunyorovesa kushatirisa kwavo uchishandisa ruzivo kubva kune yakakura kudzidziswa huwandu.
Iyo yakanyorwa fit_transform maitiro inosanganisira kuchinjika-kukodzera; kukwana kunoteverwa neshanduko pamitsara imwe chete haina kuenzana.
Ramba uchidzidza
Mamwe madhairekitori akasarudzirwa nyaya iyi
InoteveraGaidhi rinotevera
Tokenization uye Byte Pair Encoding
Tekinoroji