InoteveraGaidhi rinotevera
Vhidhiyo-ku-Vhidhiyo Style Shanduko
AI Yekutarisa
Nhungamiro yehunyanzvi
Logi, Bhokisi-Cox, uye Yeo-Johnson shanduko inogadzirisa kugovera kwenhamba, kazhinji ichideredza kurudyi skew uye kudzikamisa mutsauko usati waita modhi.
Teerera kugwaro iriAI-yakarondedzerwa · ~3 min
Ivo vanoshandura chiyero uye dudziro yezvakakosha, saka sarudzo yeshanduko, yakaiswa paramita, uye inverse fungidziro inofanirwa kubatwa nemazvo.
Shanduko inoshandura chikero chenhamba cheshanduko kuti mapatani ave nyore kuti modhi imiririre. A logarithm inokura zvishoma nezvishoma kune hombe dzakanaka hukuru, saka inogona kuderedza kurudyi skew uye kushandura hukama hwekuwedzera kuita hwekuwedzera. Shanduko inorevawo kuti misiyano pachiyero chakashandurwa haichaenderane nemisiyano yakaenzana pachiyero chepakutanga. Iyo yakajairwa logarithm inoda zvinhu zvakanaka. A log1p shanduko inoshandisa log(1 plus x) uye inogara zero asi ichiri kuda x yakakura pane isina kunaka. Kuwedzera kutenderedza nguva dzose kubvumidza zvakaipa zvinoshandura dudziro uye zvinofanirwa kururamiswa. Bhokisi-Cox inotarisa mhuri yekushandurwa kwesimba uye inofungidzira parameter, asi iyo yakajairwa fomu inoda zvakanyanya kunaka hunhu. Yeo-Johnson imwe mhuri inoshandura simba inokwanisa kubata zero uye maitiro asina kunaka. Idzi nzira dzinogona kubatsira neakatsvetwa maficha kana zvinangwa, asi hazvisi zvepasirese preprocessing zvinodiwa. Miti-yakavakirwa modhi inogona kunge isinganyanyi kunetseka kune monotonic rescaling pane mamodheru anoenderana nehukama hukama kana chinhambwe, kunyangwe shanduko yekugovera inogona kukanganisa mamwe mafambiro ebasa. Shanduko haibvisi zvinobuda kunze, kumisa zvakajairika, kugadzirisa zvakarerekera sampling, kana kuvimbisa zvirinani kufanotaura. Enzanisa maitiro emuenzaniso uye zvikanganiso pane yakabatwa-kunze data. Nezve shanduko dzakanangwa, enderana neshanduko paramita pane yekudzidziswa data uye shandura zvinangwa zvekudzidziswa zvinoenderana. Panguva yekufanotaura, shandisa shanduko yakatsauka kuti udzokere kumayuniti ekutanga. Nekuti nonlinear inverses haachengetedze maavhareji, kungoburitsa avhareji yakafanotaurwa kukosha kwegi kunogona kuburitsa fungidziro yakarerekera yepakutanga-chiyero chechimiro chinoreva. Iko kururamisa kwakakodzera kunoenderana nemafungiro ekufungidzira uye chinangwa chekuongorora. Shandisa pombi kudzivirira kukodzera shanduko paramita uchishandisa yekusimbisa kana bvunzo mienzaniso. Rekodha shanduko uye yakarongedzwa paramende kuitira kuti inferensi ishandise iwo mepu yakafanana. Taura mametrics muchiyero chinoenderana nebasa rinoshanda, uye ongorora zvakasara uye subgroup maitiro pane kutonga chete yakashandurwa kugovera.
Modhi yemakoshero akanaka emari anoshandisa shanduko yelogi kuderedza kurudyi skew, yobva yashuma fungidziro kumashure pachikero chepakutanga neshanduko inotenderedza.
Pombi inoshandisa Bhokisi-Cox kune chiyero chakasimba uye inokodzera parameter yayo uchishandisa data rekudzidzisa chete.
Dhatabheti rinosanganisira zero uye zvisina kunaka hunhu, saka muongorori anofunga Yeo-Johnson kana imwe nzira inoruramiswa pane kushandisa plain logarithm.
Chikwata chinoenzanisa maitiro ekusarira uye kuita-kunze kwekuita shanduko isati yatanga uye yapera pane kutora imwe inotaridzika seGaussian inovimbisa modhi iri nani.
Shanduko dzinoramba dzichibatsira sezvishandiso zvakajeka zvekufananidza fungidziro dzemuenzaniso uye kumiririra kuwanda kwechimiro. Mapaipi ega ega anogona kukwana paramita zvakachengeteka uye kuchengetedza mepu yakatenderedza, nepo mhando dzemazuva ano dzinogona kuderedza kudiwa kwekugadzira nemaoko kugovera mune mamwe mabasa. Hapana maitiro anobvisa kukosha kwekuenzanisa mhedzisiro pachiyero icho vashandisi vane hanya nacho. Sezvo mapaipi ekuendesa ari kushanduka, zvikwata zvinofanirwa kuchengetedza metadata yeshanduko uye kutarisa mitsara yekuisa yehukoshi hwakaiswa mepu isina kumira zvakanaka. Rekodha ma paramita akaiswa neiyo modhi artifact.
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
Logi, Bhokisi-Cox, uye Yeo-Johnson shanduko inogadzirisa kugovera kwenhamba, kazhinji ichideredza kurudyi skew uye kudzikamisa mutsauko usati waita modhi. Ivo vanoshandura chiyero uye dudziro yezvakakosha, saka sarudzo yeshanduko, yakaiswa paramita, uye inverse fungidziro inofanirwa kubatwa nemazvo.
Iyo logarithm inokura zvishoma nezvishoma kune akakura ekuisa uye inogona kuderedza kurudyi skew.
Yeo-Johnson inotsigira zvakanaka uye zvakaipa zvinopinda nemapazi akasiyana.
Kushandisa data rakabatirwa kufungidzira preprocessing paramita kunoburitsa ruzivo mukuongorora.
Shanduko inotenderedza inodzosa fungidziro kumayunitsi ekutanga ezvawakananga.
Pamabasa asina kutevedzana, kupesana kwetarisiro kunowanzosiyana netarisiro pachiyero chekutanga.
Ramba uchidzidza
Mamwe madhairekitori akasarudzirwa nyaya iyi
InoteveraGaidhi rinotevera
Vhidhiyo-ku-Vhidhiyo Style Shanduko
AI Yekutarisa