Selec\u021bia caracteristicilor \u00een \u00cenv\u0103\u021barea Automat\u0103<\/a><\/noindex>. Sper\u0103m c\u0103 va fi util tuturor celor care sunt interesa\u021bi de subiect.<\/p>\n\u00cen lumea real\u0103, datele nu sunt \u00eentotdeauna at\u00e2t de curate pe c\u00e2t \u00ee\u0219i imagineaz\u0103 adesea clien\u021bii de afaceri. De aceea, analiza inteligent\u0103 a datelor (data mining \u0219i data wrangling) este foarte c\u0103utat\u0103. Aceasta ajut\u0103 la identificarea valorilor lips\u0103 \u0219i a modelelor \u00een date structurate utiliz\u00e2nd interog\u0103ri care nu pot fi determinate de oameni. Pentru a g\u0103si \u0219i utiliza aceste modele pentru a prezice rezultatele pe baza leg\u0103turilor descoperite \u00een date, va fi necesar\u0103 \u00eenv\u0103\u021barea automat\u0103 (Machine Learning).<\/p>\n
Pentru a \u00een\u021belege orice algoritm, este esen\u021bial s\u0103 examin\u0103m toate variabilele din date \u0219i s\u0103 clarific\u0103m ce reprezint\u0103 aceste variabile. Acest lucru este extrem de important, deoarece justificarea rezultatelor se bazeaz\u0103 pe \u00een\u021belegerea datelor. Dac\u0103 datele con\u021bin 5 sau chiar 50 de variabile, le putem studia pe toate. Dar ce se \u00eent\u00e2mpl\u0103 dac\u0103 sunt 200? Atunci pur \u0219i simplu nu va fi suficient timp pentru a analiza fiecare variabil\u0103 \u00een parte. Mai mult, unele algoritmi nu func\u021bioneaz\u0103 pentru datele categorice, iar \u00een acest caz va trebui s\u0103 transform\u0103m toate coloanele categorice \u00een variabile cantitative (care pot p\u0103rea cantitative, dar metricile vor ar\u0103ta c\u0103 sunt categorice) pentru a le ad\u0103uga \u00een model. Astfel, num\u0103rul de variabile cre\u0219te, ajung\u00e2nd la aproximativ 500. Ce trebuie s\u0103 facem acum? S-ar putea crede c\u0103 solu\u021bia ar fi reducerea dimensiunii. Algoritmii de reducere a dimensiunii scad num\u0103rul de parametri, dar afecteaz\u0103 negativ interpretabilitatea. Ce ar fi dac\u0103 exist\u0103 alte tehnici care exclud caracteristicile \u0219i \u00een acela\u0219i timp permit o \u00een\u021belegere \u0219i interpretare u\u0219oar\u0103 a celor r\u0103mase?<\/p>\n
\u00cen func\u021bie de faptul c\u0103 analiza se bazeaz\u0103 pe regresie sau clasificare, algoritmii de selec\u021bie a caracteristicilor pot varia, dar ideea principal\u0103 a implement\u0103rii lor r\u0103m\u00e2ne aceea\u0219i.<\/p>\n
Variabile corelate puternic<\/h3>\n
Variabilele care sunt puternic corelate \u00eentre ele ofer\u0103 acelea\u0219i informa\u021bii modelului, a\u0219adar nu este necesar s\u0103 le folosim pe toate pentru analiz\u0103. De exemplu, dac\u0103 setul de date con\u021bine caracteristicile \u201eTimpul online\u201d \u0219i \u201eTraficul utilizat\u201d, putem presupune c\u0103 acestea vor fi corelate \u00eentr-o oarecare m\u0103sur\u0103, iar chiar \u0219i \u00een cazul unui e\u0219antion de date impar\u021bial, vom observa o corela\u021bie puternic\u0103. \u00cen acest caz, modelul are nevoie doar de una dintre aceste variabile. Dac\u0103 folosim ambele, modelul va fi supraaglomerat (overfit) \u0219i va fi p\u0103rtinitor fa\u021b\u0103 de o caracteristic\u0103 specific\u0103.<\/p>\n
Valorile P<\/h3>\n
\u00cen algoritmi precum regresia liniar\u0103, modelul statistic ini\u021bial este \u00eentotdeauna o idee bun\u0103. Acesta ajut\u0103 la eviden\u021bierea importan\u021bei caracteristicilor prin valorile lor p, care au fost ob\u021binute de acest model. Stabilind un nivel de semnifica\u021bie, verific\u0103m valorile p ob\u021binute, iar dac\u0103 vreuna dintre ele este sub nivelul de semnifica\u021bie stabilit, caracteristica respectiv\u0103 este considerat\u0103 semnificativ\u0103, adic\u0103 modificarea valorii sale va conduce probabil la o modificare a valorii \u021bintei (target).<\/p>\n
Selectare direct\u0103<\/h3>\n
Selectarea direct\u0103 este o tehnic\u0103 care implic\u0103 regresia \u00een etape. Construirea modelului \u00eencepe de la zero, adic\u0103 de la un model gol, iar apoi fiecare itera\u021bie adaug\u0103 o variabil\u0103 care \u00eembun\u0103t\u0103\u021be\u0219te modelul \u00een construc\u021bie. Variabila care este ad\u0103ugat\u0103 \u00een model este determinat\u0103 de semnifica\u021bia sa. Acest lucru poate fi calculat folosind diferite metrici. Cea mai comun\u0103 metod\u0103 este aplicarea valorilor p ob\u021binute din modelul statistic ini\u021bial folosind toate variabilele. Uneori, selectarea direct\u0103 poate duce la supraaglomerarea modelului, deoarece modelul poate con\u021bine variabile puternic corelate care ofer\u0103 acelea\u0219i informa\u021bii, chiar dac\u0103 modelul arat\u0103 o \u00eembun\u0103t\u0103\u021bire.<\/p>\n
Selectare invers\u0103<\/h3>\n
Selec\u021bia invers\u0103 const\u0103 de asemenea \u00een excluderea treptat\u0103 a caracteristicilor, \u00eens\u0103 \u00een sens opus fa\u021b\u0103 de selec\u021bia direct\u0103. \u00cen acest caz, modelul ini\u021bial cuprinde toate variabilele independente. Apoi, variabilele sunt excluse (c\u00e2te una pe itera\u021bie), dac\u0103 nu contribuie la noul model de regresie \u00een fiecare itera\u021bie. Excluderea caracteristicilor se bazeaz\u0103 pe valorile p ale modelului ini\u021bial. Aceast\u0103 metod\u0103 implic\u0103 de asemenea incertitudinea la eliminarea variabilelor foarte corelate.<\/p>\n
Excluderea recursiv\u0103 a caracteristicilor<\/h3>\n
RFE este o tehnic\u0103\/algoritm utilizat pe scar\u0103 larg\u0103 pentru selectarea unui num\u0103r exact de caracteristici semnificative. Uneori, metoda este folosit\u0103 pentru a explica un anumit num\u0103r de \u201ecele mai importante\u201d caracteristici care afecteaz\u0103 rezultatele; iar alteori pentru a reduce un num\u0103r foarte mare de variabile (aproximativ 200-400), p\u0103str\u00e2nd doar acelea care contribuie \u00eentr-un anumit fel la model, iar toate celelalte fiind excluse. RFE folose\u0219te un sistem de ranguri. Caracteristicile din setul de date sunt clasificate. Apoi, aceste ranguri sunt utilizate pentru excluderea recursiv\u0103 a caracteristicilor \u00een func\u021bie de coliniaritatea dintre ele \u0219i semnifica\u021bia acestor caracteristici \u00een model. Pe l\u00e2ng\u0103 clasificarea caracteristicilor, RFE poate ar\u0103ta dac\u0103 aceste caracteristici sunt importante sau nu chiar \u0219i pentru un num\u0103r dat de caracteristici (deoarece este foarte probabil ca num\u0103rul selectat de caracteristici s\u0103 nu fie optim, iar num\u0103rul optim de caracteristici poate fi fie mai mare, fie mai mic dec\u00e2t cel selectat).<\/p>\n
Diagram\u0103 de importan\u021b\u0103 a caracteristicilor<\/h3>\n
C\u00e2nd se discut\u0103 despre interpretabilitatea algoritmilor de \u00eenv\u0103\u021bare automatizat\u0103, se discut\u0103 de obicei despre regresiile liniare (care permit analizarea semnifica\u021biei caracteristicilor folosind valorile p) \u0219i arborii de decizie (care arat\u0103 literalmente importan\u021ba caracteristicilor sub form\u0103 de arbore \u0219i ierarhia lor). Pe de alt\u0103 parte, \u00een algoritmi precum Random Forest, LightGBM \u0219i XG Boost, este adesea utilizat\u0103 o diagram\u0103 de importan\u021b\u0103 a caracteristicilor, adic\u0103 se construie\u0219te o diagram\u0103 a variabilelor \u0219i \u201ecantitatea de importan\u021b\u0103\u201d a acestora. Aceasta este deosebit de util\u0103 atunci c\u00e2nd trebuie s\u0103 furniza\u021bi o justificare structurat\u0103 a importan\u021bei caracteristicilor din punctul de vedere al impactului lor asupra afacerii.<\/p>\n
Regularizare<\/h3>\n
Regularea se face pentru a controla echilibrul \u00eentre p\u0103rtinire (bias) \u0219i varia\u021bie (variance). P\u0103rtinirea arat\u0103 c\u00e2t de mult s-a suprasolicitat (overfit) modelul pe setul de date de antrenare. Varia\u021bia arat\u0103 c\u00e2t de diferite au fost predic\u021biile \u00eentre seturile de date de antrenare \u0219i de testare. Ideal, at\u00e2t p\u0103rtinirea, c\u00e2t \u0219i varia\u021bia ar trebui s\u0103 fie mici. Aici intervine regularea! Exist\u0103 dou\u0103 tehnici de baz\u0103:<\/p>\n
Regularea L1 \u2014 Lasso: Lasso penalizeaz\u0103 coeficientii de greutate ai modelului pentru a schimba importan\u021ba lor \u00een model \u0219i poate chiar s\u0103-i anuleze (adic\u0103 s\u0103 elimine aceste variabile din modelul final). De obicei, Lasso este utilizat atunci c\u00e2nd setul de date con\u021bine un num\u0103r mare de variabile \u0219i este necesar s\u0103 se excluz\u0103 unele dintre ele pentru a \u00een\u021belege mai bine cum influen\u021beaz\u0103 caracteristicile importante modelul (adic\u0103 acele caracteristici care au fost selectate de Lasso \u0219i pentru care s-a stabilit importan\u021ba).<\/p>\n
Regularea L2 \u2014 prin metoda Ridge: Scopul Ridge este de a p\u0103stra toate variabilele \u0219i de a le acorda totodat\u0103 importan\u021b\u0103 bazat\u0103 pe contribu\u021bia lor la eficien\u021ba modelului. Ridge va fi o alegere bun\u0103 dac\u0103 setul de date con\u021bine un num\u0103r mic de variabile \u0219i toate sunt necesare pentru interpretarea concluziilor \u0219i a rezultatelor ob\u021binute.<\/p>\n
Deoarece Ridge p\u0103streaz\u0103 toate variabilele, iar Lasso stabile\u0219te mai bine importan\u021ba lor, a fost dezvoltat un algoritm care combin\u0103 cele mai bune caracteristici ale ambelor regul\u0103ri \u0219i este cunoscut sub numele de Elastic-Net.<\/p>\n
Exist\u0103 \u0219i multe alte modalit\u0103\u021bi de selec\u021bie a caracteristicilor pentru \u00eenv\u0103\u021barea automat\u0103, dar ideea principal\u0103 r\u0103m\u00e2ne mereu aceea\u0219i: a demonstra importan\u021ba variabilelor \u0219i apoi a exclude unele dintre ele pe baza importan\u021bei ob\u021binute. Importan\u021ba este un termen foarte subiectiv, deoarece nu este unul, ci un \u00eentreg set de metrici \u0219i diagrame care pot fi utilizate pentru a g\u0103si caracteristicile cheie.<\/p>\n
Mul\u021bumim pentru lectur\u0103! \u00cenv\u0103\u021bare pl\u0103cut\u0103!<\/p>\n
Sursa: 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\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\udd47Selec\u021bia caracteristicilor \u00een \u00eenv\u0103\u021barea automat\u0103 | ProHoster","description":"Salut, Habr! Noi, la \u201eRexoft\u201d, am tradus \u00een limba rom\u00e2n\u0103 articolul Selec\u021bia Caracteristicilor \u00een Ma\u0219ini.","canonical_url":"https:\/\/prohoster.info\/ro\/blog\/administrirovanie\/otbor-priznakov-v-mashinnom-obuchenii","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"ro_RO","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\/ro\/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\/ro\/wp-json\/wp\/v2\/posts\/93155","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/comments?post=93155"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/93155\/revisions"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/media?parent=93155"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/categories?post=93155"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/tags?post=93155"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}