Funkciju atlase ma\u0161\u012bnm\u0101c\u012bb\u0101<\/a><\/noindex>. Ceram, ka tas b\u016bs noder\u012bgi visiem, kas nav vienaldz\u012bgi pret \u0161o t\u0113mu.<\/p>\nRe\u0101laj\u0101 pasaul\u0113 dati ne vienm\u0113r ir tik t\u012bri, k\u0101 da\u017ereiz dom\u0101 biznesa pas\u016bt\u012bt\u0101ji. Tie\u0161i t\u0101p\u0113c ir piepras\u012bts inteli\u0123entais datu anal\u012bze (data mining un data wrangling). Tas pal\u012bdz identific\u0113t tr\u016bksto\u0161\u0101s v\u0113rt\u012bbas un raksturus struktur\u0113tajos datos, izmantojot piepras\u012bjumus, kurus cilv\u0113ks nevar noteikt. Lai atrastu un izmantotu \u0161os raksturus rezult\u0101tu prognoz\u0113\u0161anai, izmantojot atkl\u0101t\u0101s saites datos, noder\u0113s ma\u0161\u012bnm\u0101c\u012bba (Machine Learning).<\/p>\n
Lai izprastu jebkuru algoritmu, ir j\u0101apskata visas main\u012bg\u0101s v\u0113rt\u012bbas datos un j\u0101noskaidro, ko \u0161ie main\u012bgie p\u0101rst\u0101v. Tas ir \u0101rk\u0101rt\u012bgi svar\u012bgi, jo rezult\u0101tu pamatojums balst\u0101s uz datu izpratni. Ja datos ir 5 vai pat 50 main\u012bgo, tos visus var izp\u0113t\u012bt. Bet ko dar\u012bt, ja tie ir 200? Tad laika nevar\u0113s pietikt, lai izp\u0113t\u012btu katru atsevi\u0161\u0137o main\u012bgo. Turkl\u0101t da\u017ei algoritmi nedarbojas ar kategoriskajiem datiem, un tad b\u016bs j\u0101konvert\u0113 visi kategoriskie stabi kvantitat\u012bvajos main\u012bgajos (tie var izskat\u012bties kvantitat\u012bvi, bet metrikas par\u0101d\u012bs, ka tie ir kategoriski), lai tos var\u0113tu iek\u013caut model\u012b. T\u0101pat main\u012bgo skaits palielin\u0101s, un to k\u013c\u016bst apm\u0113ram 500. Ko tagad dar\u012bt? Var\u0113tu dom\u0101t, ka risin\u0101jums b\u016bs dimensiju samazin\u0101\u0161ana. Dimensiju samazin\u0101\u0161anas algoritmi samazina parametru skaitu, bet negat\u012bvi ietekm\u0113 interpret\u0113jamo rezult\u0101tu. Ko dar\u012bt, ja past\u0101v citas tehnikas, kas izsl\u0113dz iez\u012bmes un taj\u0101 pa\u0161\u0101 laik\u0101 \u013cauj viegli saprast un interpret\u0113t atliku\u0161os?<\/p>\n
Atkar\u012bb\u0101 no t\u0101, vai anal\u012bze balst\u0101s uz regresiju vai klasifik\u0101ciju, iez\u012bmes atlases algoritmi var at\u0161\u0137irties, ta\u010du to ievie\u0161anas galven\u0101 ideja paliek viena un t\u0101 pati.<\/p>\n
Stipri korel\u0113ti main\u012bgie<\/h3>\n
V\u0101ji savstarp\u0113ji korel\u0113tie main\u012bgie sniedz modelim to pa\u0161u inform\u0101ciju, t\u0101p\u0113c anal\u012bzei nav nepiecie\u0161ams izmantot visus. Piem\u0113ram, ja datu kop\u0101 ir iek\u013cauti r\u0101d\u012bt\u0101ji \"Laiks tie\u0161saist\u0113\" un \"Izmantot\u0101 datu pl\u016bsma\", var pie\u0146emt, ka tie zin\u0101m\u0101 m\u0113r\u0101 b\u016bs korel\u0113ti, un m\u0113s nov\u0113rosim sp\u0113c\u012bgu korel\u0101ciju, pat ja izv\u0113l\u0113simies neobjekt\u012bvu datu paraugu. \u0160\u0101d\u0101 gad\u012bjum\u0101 model\u012b nepiecie\u0161ams tikai viens no \u0161iem main\u012bgajiem. Ja izmantosim abus, modelis k\u013c\u016bs p\u0101rm\u0101c\u012bts (overfit) un iepriek\u0161\u0113jais attiec\u012bb\u0101 uz k\u0101du konkr\u0113tu raksturojumu.<\/p>\n
P-v\u0113rt\u012bbas<\/h3>\n
T\u0101dos algoritmos k\u0101 line\u0101r\u0101 regresija s\u0101kotn\u0113j\u0101 statistisk\u0101 mode\u013ca izveide vienm\u0113r ir laba ideja. T\u0101 pal\u012bdz par\u0101d\u012bt r\u0101d\u012bt\u0101ju noz\u012bm\u012bgumu, balstoties uz to p-v\u0113rt\u012bb\u0101m, kuras ieg\u016btas \u0161aj\u0101 model\u012b. Iest\u0101dot noz\u012bm\u012bguma l\u012bmeni, m\u0113s p\u0101rbaud\u0101m ieg\u016bt\u0101s p-v\u0113rt\u012bbas, un, ja k\u0101da v\u0113rt\u012bba izr\u0101d\u0101s zem\u0101ka par noteikto noz\u012bm\u012bguma l\u012bmeni, \u0161is r\u0101d\u012bt\u0101js tiek uzskat\u012bts par noz\u012bm\u012bgu, tas ir, izmai\u0146as t\u0101 v\u0113rt\u012bb\u0101, visticam\u0101k, rad\u012bs izmai\u0146as m\u0113r\u0137a v\u0113rt\u012bb\u0101.<\/p>\n
Taisn\u0101 atlase<\/h3>\n
Taisn\u0101 atlase ir tehnika, kas ietver pak\u0101peniskas regresijas pielietojumu. Mode\u013ca izveidi s\u0101k ar piln\u012bgu nulli, proti, tuk\u0161u modeli, bet p\u0113c tam katra iter\u0101cija pievieno main\u012bgo, kas uzlabo veidojamo modeli. T\u0101s main\u012bg\u0101 pievieno\u0161ana modelim tiek noteikta p\u0113c t\u0101s noz\u012bm\u012bguma. To var apr\u0113\u0137in\u0101t, izmantojot da\u017e\u0101das metriskas. Visizplat\u012bt\u0101kais veids ir p-v\u0113rt\u012bbu pielieto\u0161ana, kas ieg\u016btas s\u0101kotn\u0113j\u0101 statistiskaj\u0101 model\u012b ar visiem main\u012bgajiem. Da\u017ereiz taisn\u0101 atlase var novest pie mode\u013ca p\u0101rm\u0101c\u012bbas, jo model\u012b var b\u016bt sp\u0113c\u012bgi korel\u0113ti main\u012bgie, pat ja tie sniedz modeli vienu un to pa\u0161u inform\u0101ciju (ta\u010du modelis \u0161aj\u0101 gad\u012bjum\u0101 r\u0101da uzlabojumus).<\/p>\n
Aizmugur\u0113j\u0101 atlase<\/h3>\n
Atgriezenisk\u0101 atlase ir sol\u012bga izsl\u0113g\u0161ana, ta\u010du pret\u0113j\u0101 virzien\u0101 sal\u012bdzin\u0101jum\u0101 ar tie\u0161o. \u0160aj\u0101 gad\u012bjum\u0101 s\u0101kotn\u0113j\u0101 mode\u013ca ietver visas neatkar\u012bg\u0101s main\u012bg\u0101s. P\u0113c tam main\u012bgie tiek izsl\u0113gti (pa vienam katr\u0101 iter\u0101cij\u0101), ja tie neieg\u016bst v\u0113rt\u012bbu jaunajai regresijas modelim katr\u0101 iter\u0101cij\u0101. Izsl\u0113g\u0161anas pamat\u0101 ir p-v\u0113rt\u012bbas s\u0101kotn\u0113j\u0101 model\u012b. \u0160aj\u0101 metod\u0113 ar\u012b past\u0101v nenoteikt\u012bba, izsl\u0113dzot stipri korel\u0113tus main\u012bgos.<\/p>\n
Rekurs\u012bv\u0101 main\u012bgo izsl\u0113g\u0161ana<\/h3>\n
RFE ir pla\u0161i izmantota tehnika\/algoritms prec\u012bza noz\u012bm\u012bgu main\u012bgo skaita izv\u0113l\u0113. Da\u017ereiz metodi izmanto, lai izskaidrotu k\u0101du skaitu \"visnoz\u012bm\u012bg\u0101kajiem\" main\u012bgajiem, kas ietekm\u0113 rezult\u0101tus; citreiz, lai samazin\u0101tu \u013coti lielu main\u012bgo skaitu (ap 200-400), atst\u0101jot tikai tos, kas iegulda kaut k\u0101du ieguld\u012bjumu model\u012b, bet visus p\u0101r\u0113jos izsl\u0113dz. RFE izmanto rangu sist\u0113mu. Main\u012bgajiem datu kop\u0101 tiek pie\u0161\u0137irti rangi. P\u0113c tam \u0161ie rangi tiek izmantoti rekurs\u012bvai main\u012bgo izsl\u0113g\u0161anai atkar\u012bb\u0101 no to kolinearit\u0101tes un noz\u012bm\u012bbas \u0161ajos main\u012bgajos model\u012b. Papildus main\u012bgo rang\u0113\u0161anai, RFE var par\u0101d\u012bt, cik noz\u012bm\u012bgi \u0161ie main\u012bgie ir vai nav pat doto main\u012bgo skaitam (jo ir \u013coti iesp\u0113jams, ka izv\u0113l\u0113tais main\u012bgo skaits var b\u016bt neoptim\u0101ls, un optim\u0101lais skaits var b\u016bt gan liel\u0101ks, gan maz\u0101ks par izv\u0113l\u0113to).<\/p>\n
Main\u012bgo noz\u012bm\u012bguma diagramma<\/h3>\n
Run\u0101jot par ma\u0161\u012bnm\u0101c\u012b\u0161an\u0101s algoritmu interpret\u0113jam\u012bbu, parasti tiek apspriesta line\u0101r\u0101 regresija (kas \u013cauj analiz\u0113t main\u012bgo noz\u012bm\u012bbu, izmantojot p-v\u0113rt\u012bbas) un l\u0113mumu koki (kas burtiski par\u0101da main\u012bgo noz\u012bm\u012bbu koka form\u0101, k\u0101 ar\u012b vi\u0146u hierarhiju). No otras puses, t\u0101dos algoritmos k\u0101 Random Forest, LightGBM un XG Boost, bie\u017ei tiek izmantota main\u012bgo noz\u012bm\u012bguma diagramma, t.i., tiek izveidota diagramma ar main\u012bgajiem un \"to noz\u012bm\u012bguma skaitu\". Tas ir \u012bpa\u0161i noder\u012bgi, kad nepiecie\u0161ams sniegt struktur\u0113tu pamatojumu main\u012bgo noz\u012bm\u012bbai attiec\u012bb\u0101 uz vi\u0146u ietekmi uz biznesu.<\/p>\n
Regul\u0101raiz\u0101cija<\/h3>\n
Regul\u0113\u0161ana tiek veikta, lai kontrol\u0113tu l\u012bdzsvaru starp noslieci (bias) un novirzi (variance). Nosliece par\u0101da, cik sava mode\u013ca ir p\u0101rslodzes gad\u012bjums (overfit) apm\u0101c\u012bbu datu kop\u0101. Novirze r\u0101da, cik at\u0161\u0137ir\u012bgi bija prognozes starp apm\u0101c\u012bbu un testa datu kop\u0101m. Ide\u0101l\u0101 gad\u012bjum\u0101 gan nosliecei, gan novirzei j\u0101b\u016bt zem\u0101m. \u0160eit n\u0101k pal\u012bg\u0101 regul\u0113\u0161ana! Ir divas pamattehnikas:<\/p>\n
L1 regul\u0113\u0161ana - Lasso: Lasso sods mode\u013ca svaru koeficientiem, lai main\u012btu to noz\u012bm\u012bbu model\u012b un var pat tos nullo (t.i., izsl\u0113gt \u0161os main\u012bgos no gal\u012bg\u0101 mode\u013ca). Parasti Lasso tiek izmantots, ja datu kop\u0101 ir liels skaits main\u012bgo un ir nepiecie\u0161ams izsl\u0113gt da\u017eus no tiem, lai lab\u0101k izprastu, k\u0101 svar\u012bg\u0101s iez\u012bmes ietekm\u0113 modeli (t.i., t\u0101s iez\u012bmes, kuras izv\u0113l\u0113j\u0101s Lasso un kuru noz\u012bm\u012bba tika noteikta).<\/p>\n
L2 regul\u0113\u0161ana - Ridge metode: Ridge uzdevums ir saglab\u0101t visus main\u012bgos un taj\u0101 pa\u0161\u0101 laik\u0101 pie\u0161\u0137irt tiem noz\u012bm\u012bbu, pamatojoties uz ieguld\u012bjumu mode\u013ca efektivit\u0101t\u0113. Ridge b\u016bs laba izv\u0113le, ja datu kop\u0101 ir neliels skaits main\u012bgo un visi ir nepiecie\u0161ami secin\u0101jumu un ieg\u016bto rezult\u0101tu interpret\u0101cijai.<\/p>\n
T\u0101 k\u0101 Ridge atst\u0101j visus main\u012bgos, bet Lasso lab\u0101k nosaka to noz\u012bm\u012bbu, tika izstr\u0101d\u0101ts algoritms, kas apvieno lab\u0101k\u0101s abu regul\u0113\u0161anas iez\u012bmes un ir paz\u012bstams k\u0101 Elastic-Net.<\/p>\n
V\u0113l ir daudz veidu, k\u0101 izv\u0113l\u0113ties iez\u012bmes ma\u0161\u012bnm\u0101c\u012bb\u0101, bet galven\u0101 ideja vienm\u0113r paliek t\u0101da pati: par\u0101d\u012bt main\u012bgo noz\u012bm\u012bgumu un p\u0113c tam izsl\u0113gt da\u017eus no tiem, pamatojoties uz ieg\u016bto noz\u012bm\u012bbu. Noz\u012bm\u012bgums ir \u013coti subjekt\u012bvs j\u0113dziens, jo tas nav viens, bet visai liels metrikas un diagrammu kopums, ko var izmantot, lai atrastu galven\u0101s iez\u012bmes.<\/p>\n
Paldies par las\u012b\u0161anu! Lai veicas m\u0101c\u012bb\u0101s!<\/p>\n
Avots: habr.com<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"\u041f\u0440\u0438\u0432\u0435\u0442, \u0425\u0430\u0431\u0440! \u041c\u044b \u0432 \u00ab\u0420\u0435\u043a\u0441\u043e\u0444\u0442\u00bb \u043f\u0435\u0440\u0435\u0432\u0435\u043b\u0438 \u043d\u0430 \u0440\u0443\u0441\u0441\u043a\u0438\u0439 \u044f\u0437\u044b\u043a \u0441\u0442\u0430\u0442\u044c\u044e Feature Selection in Machine Learning. \u041d\u0430\u0434\u0435\u0435\u043c\u0441\u044f, \u0431\u0443\u0434\u0435\u0442 \u043f\u043e\u043b\u0435\u0437\u043d\u043e \u0432\u0441\u0435\u043c, \u043a\u0442\u043e \u043d\u0435\u0440\u0430\u0432\u043d\u043e\u0434\u0443\u0448\u0435\u043d \u043a \u0442\u0435\u043c\u0435. \u0412 \u0440\u0435\u0430\u043b\u044c\u043d\u043e\u043c \u043c\u0438\u0440\u0435 \u0434\u0430\u043d\u043d\u044b\u0435 \u043d\u0435 \u0432\u0441\u0435\u0433\u0434\u0430 \u0442\u0430\u043a\u0438\u0435 \u0447\u0438\u0441\u0442\u044b\u0435, \u043a\u0430\u043a \u043f\u043e\u0440\u043e\u0439 \u0434\u0443\u043c\u0430\u044e\u0442 \u0431\u0438\u0437\u043d\u0435\u0441-\u0437\u0430\u043a\u0430\u0437\u0447\u0438\u043a\u0438. \u0418\u043c\u0435\u043d\u043d\u043e \u043f\u043e\u044d\u0442\u043e\u043c\u0443 \u0432\u043e\u0441\u0442\u0440\u0435\u0431\u043e\u0432\u0430\u043d \u0438\u043d\u0442\u0435\u043b\u043b\u0435\u043a\u0442\u0443\u0430\u043b\u044c\u043d\u044b\u0439 \u0430\u043d\u0430\u043b\u0438\u0437 \u0434\u0430\u043d\u043d\u044b\u0445 (data mining \u0438 data wrangling). \u041e\u043d \u043f\u043e\u043c\u043e\u0433\u0430\u0435\u0442 \u0432\u044b\u044f\u0432\u043b\u044f\u0442\u044c \u043d\u0435\u0434\u043e\u0441\u0442\u0430\u044e\u0449\u0438\u0435 \u0437\u043d\u0430\u0447\u0435\u043d\u0438\u044f \u0438 \u043f\u0430\u0442\u0442\u0435\u0440\u043d\u044b \u0432 \u0441\u0442\u0440\u0443\u043a\u0442\u0443\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u043d\u044b\u0445 […]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[688],"tags":[],"class_list":["post-93155","post","type-post","status-publish","format-standard","hentry","category-administrirovanie"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\n","aioseo_head_json":{"title":"\ud83e\udd47Iez\u012b\u010du izv\u0113le ma\u0161\u012bnm\u0101c\u012bb\u0101 | ProHoster","description":"Sveiki, Habr! M\u0113s \"Rexoft\" esam p\u0101rtulkoju\u0161i krievu valod\u0101 rakstu \"Feature Selection in Machine\".","canonical_url":"https:\/\/prohoster.info\/lv\/blog\/administrirovanie\/otbor-priznakov-v-mashinnom-obuchenii","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"lv_LV","og:site_name":"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b","og:type":"article","og:title":"\ud83e\udd47\u041e\u0442\u0431\u043e\u0440 \u043f\u0440\u0438\u0437\u043d\u0430\u043a\u043e\u0432 \u0432 \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u043c \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0438 | ProHoster","og:description":"\u041f\u0440\u0438\u0432\u0435\u0442, \u0425\u0430\u0431\u0440!\u041c\u044b \u0432 \u00ab\u0420\u0435\u043a\u0441\u043e\u0444\u0442\u00bb \u043f\u0435\u0440\u0435\u0432\u0435\u043b\u0438 \u043d\u0430 \u0440\u0443\u0441\u0441\u043a\u0438\u0439 \u044f\u0437\u044b\u043a \u0441\u0442\u0430\u0442\u044c\u044e Feature Selection in Machine.","og:url":"https:\/\/prohoster.info\/lv\/blog\/administrirovanie\/otbor-priznakov-v-mashinnom-obuchenii","og:image":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:secure_url":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:width":350,"og:image:height":350,"article:published_time":"2020-09-03T11:42:13+00:00","article:modified_time":"2020-09-03T11:42:13+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"93155","title":null,"description":null,"keywords":null,"keyphrases":null,"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":null,"og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"","isEnabled":true},"graphs":[]},"schema_type":null,"schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":null,"robots_max_videopreview":null,"robots_max_imagepreview":"large","priority":null,"frequency":null,"local_seo":null,"seo_analyzer_scan_date":null,"breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2021-02-28 11:54:00","updated":"2022-09-30 17:42:24","focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/lv\/wp-json\/wp\/v2\/posts\/93155","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/lv\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/lv\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/lv\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/lv\/wp-json\/wp\/v2\/comments?post=93155"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/lv\/wp-json\/wp\/v2\/posts\/93155\/revisions"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/lv\/wp-json\/wp\/v2\/media?parent=93155"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/lv\/wp-json\/wp\/v2\/categories?post=93155"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/lv\/wp-json\/wp\/v2\/tags?post=93155"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}