{"id":30936,"date":"2019-10-31T21:38:20","date_gmt":"2019-10-31T18:38:20","guid":{"rendered":"https:\/\/prohoster.info\/blog\/mashinnoe-obuchenie-bez-python-anaconda-i-prochih-presmykayushhihsya\/"},"modified":"2019-10-31T21:38:20","modified_gmt":"2019-10-31T18:38:20","slug":"mashinnoe-obuchenie-bez-python-anaconda-i-prochih-presmykayushhihsya","status":"publish","type":"post","link":"https:\/\/prohoster.info\/ro\/blog\/news\/mashinnoe-obuchenie-bez-python-anaconda-i-prochih-presmykayushhihsya","title":{"rendered":"\u00cenv\u0103\u021barea automat\u0103 f\u0103r\u0103 Python, Anaconda \u0219i alte reptile","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Nu, bine\u00een\u021beles, nu vorbesc serios. Trebuie s\u0103 existe o limit\u0103 p\u00e2n\u0103 la care este posibil s\u0103 simplific\u0103m subiectul. Dar pentru primele etape, pentru a \u00een\u021belege conceptele de baz\u0103 \u0219i pentru a ne familiariza rapid cu tema, poate fi acceptabil. Vom discuta la final cum s\u0103 denumim corect acest material (op\u021biuni: \u201e\u00cenv\u0103\u021barea automat\u0103 pentru \u00eencep\u0103tori\u201d, \u201eAnaliza datelor de la \u00eenceput\u201d, \u201eAlgoritmi pentru cei mai mici\u201d). <\/p>\n

La subiect. Am scris c\u00e2teva programe aplicate \u00een MS Excel pentru a vizualiza \u0219i reprezenta \u00een mod clar procesele care au loc \u00een diferite metode de \u00eenv\u0103\u021bare automat\u0103 atunci c\u00e2nd analiz\u0103m date. Seeing is believing, p\u00e2n\u0103 la urm\u0103, cum spun cei din cultura care a dezvoltat majoritatea acestor metode (apropos, nu toate dintre ele. Puternicul \u201emetod\u0103 a vectorilor de suport\u201d, sau SVM, support vector machine \u2013 este inven\u021bia compatriotului nostru Vladimir Vapnik, Institutul din Moscova. 1963, apropo! Acum, desigur, pred\u0103 \u0219i lucreaz\u0103 \u00een SUA).<\/p>\n

Trei fi\u0219iere pentru revizuire
\n<\/a><\/noindex><\/p>\n

1. Clustering prin metoda k-means<\/h2>\n

\nSarcinile acestui tip se \u00eencadreaz\u0103 \u00een \u201e\u00eenv\u0103\u021barea nesupravegheat\u0103\u201d, atunci c\u00e2nd trebuie s\u0103 \u00eemp\u0103r\u021bim datele originale \u00eentr-un num\u0103r prestabilit de categorii, dar nu avem niciun num\u0103r de \u201er\u0103spunsuri corecte\u201d - acestea trebuie extrase din datele \u00een sine. Clasicul exemplu fundamental de identificare a subspeciilor florilor de iris (Ronald Fisher, 1936!), care este considerat prima piatr\u0103 de temelie \u00een acest domeniu de cunoa\u0219tere - este exact de aceast\u0103 natur\u0103.<\/p>\n

Metoda este destul de simpl\u0103. Avem un set de obiecte, reprezentate sub form\u0103 de vectori (seturi de N numere). \u00cen cazul irisurilor, acestea sunt seturi de 4 numere, care caracterizeaz\u0103 floarea: lungimea \u0219i l\u0103\u021bimea petalelor externe \u0219i interne, respectiv (Irisurile lui Fisher - Wikipedia<\/a><\/noindex>). Ca m\u0103sur\u0103 a distan\u021bei sau a similitudinii \u00eentre obiecte, se alege metrica euclidean\u0103 obi\u0219nuit\u0103.<\/p>\n

Apoi, \u00een mod arbitrar (sau nu, vezi mai departe), se aleg centrele clusterelor, iar distan\u021bele de la fiecare obiect la centrele clusterelor sunt calculate. Fiecare obiect \u00een aceast\u0103 etap\u0103 a itera\u021biei este marcat ca apar\u021bin\u00e2nd celui mai apropiat centru. Apoi, centrul fiec\u0103rui cluster este mutat \u00een media aritmetic\u0103 a coordonatelor membrilor s\u0103i (similar cu fizica, este denumit \u0219i \u201ecentru de mas\u0103\u201d), iar procedura se repet\u0103.<\/p>\n

Procesul converge destul de rapid. \u00cen imaginile bidimensionale, acesta arat\u0103 astfel:<\/p>\n

1. Distribu\u021bia ini\u021bial\u0103 aleatorie a punctelor pe plan \u0219i num\u0103rul de clustere<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\n2. Stabilirea centrelor de cluster \u0219i atribuite punctelor la clusterele lor<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\n3. Mutarea coordonatelor centrelor de cluster, recalcularea apartenen\u021bei punctelor, p\u00e2n\u0103 c\u00e2nd centrele se stabilizeaz\u0103. Se observ\u0103 traiectoria de mi\u0219care a centrului clusterului c\u0103tre pozi\u021bia final\u0103.<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\n\u00cen orice moment, se pot stabili noi centre de cluster (f\u0103r\u0103 a genera o nou\u0103 distribu\u021bie a punctelor!) \u0219i se poate observa c\u0103 procesul de divizare nu este \u00eentotdeauna univoc. Matematic, aceasta \u00eenseamn\u0103 c\u0103, pentru func\u021bia optimizat\u0103 (suma p\u0103tratelor distan\u021belor de la puncte la centrele clusterelor lor), g\u0103sim un minim local, nu global. Aceast\u0103 problem\u0103 poate fi rezolvat\u0103 fie prin alegerea determinist\u0103 a centrelor ini\u021biale ale clusterelor, fie prin explorarea centrelor posibile (uneori este avantajos s\u0103 le plas\u0103m exact \u00een una dintre puncte, atunci avem m\u0103car garan\u021bia c\u0103 nu vom ob\u021bine clustere goale). \u00cen orice caz, un set finit are \u00eentotdeauna o margine inferioar\u0103 precis\u0103. <\/p>\n

Pute\u021bi experimenta cu acest fi\u0219ier la acest link<\/a><\/noindex> (nu uita\u021bi s\u0103 activa\u021bi suportul pentru macrocomenzi. Fi\u0219ierele au fost verificate pentru virusuri)<\/p>\n

Descrierea metodei pe Wikipedia \u2014 Metoda k-medii<\/a><\/noindex><\/p>\n

2. Aproximarea cu polinoame \u0219i divizarea datelor. Supra\u00eenv\u0103\u021barea<\/h2>\n

\nCercet\u0103torul \u0219i popularizatorul \u0219tiin\u021bei datelor K.V. Voron\u021bov vorbe\u0219te pe scurt despre metodele de \u00eenv\u0103\u021bare automat\u0103 ca despre \u201e\u0219tiin\u021ba de a trasa curbe prin puncte\u201d. \u00cen acest exemplu, vom g\u0103si regularit\u0103\u021bi \u00een date folosind metoda minimelor p\u0103trate. <\/p>\n

Este ilustrat\u0103 tehnica de \u00eemp\u0103r\u021bire a datelor originale \u00een \u201edate de \u00eenv\u0103\u021bare\u201d \u0219i \u201edate de control\u201d, precum \u0219i fenomenul de supra\u00eenv\u0103\u021bare sau \u201efine-tuning\u201d pe date. La o aproximare corect\u0103, vom avea o anumit\u0103 eroare pe datele de \u00eenv\u0103\u021bare \u0219i o eroare mai mare pe datele de control. La o aproximare gre\u0219it\u0103 \u2013 o ajustare exact\u0103 la datele de \u00eenv\u0103\u021bare \u0219i o eroare uria\u0219\u0103 pe cele de control.<\/p>\n

(Este un fapt cunoscut c\u0103 prin N puncte se poate trasa o singur\u0103 curb\u0103 de grad N-1, iar aceast\u0103 metod\u0103 \u00een general nu d\u0103 rezultatul dorit. Polinomul de interpolare Lagrange pe Wikipedia<\/a><\/noindex>)<\/p>\n

1. Stabilim distribu\u021bia ini\u021bial\u0103<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\n2. \u00cemp\u0103r\u021bim punctele \u00een \u201edate de \u00eenv\u0103\u021bare\u201d \u0219i \u201edate de control\u201d \u00een propor\u021bie de 70 la 30.<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\n3. Trasez\u0103m o curb\u0103 de aproximare prin punctele de \u00eenv\u0103\u021bare, observ\u0103m eroarea pe care o d\u0103 la datele de testare<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\n4. Trasez\u0103m o curb\u0103 exact\u0103 prin punctele de \u00eenv\u0103\u021bare \u0219i vedem o eroare uria\u0219\u0103 la datele de testare (\u0219i zero la cele de \u00eenv\u0103\u021bare, dar ce folos?).<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\nEste ar\u0103tat, desigur, cea mai simpl\u0103 variant\u0103 cu o singur\u0103 \u00eemp\u0103r\u021bire \u00een submul\u021bimi 'de \u00eenv\u0103\u021bare' \u0219i 'de control', \u00een cazul general aceast\u0103 opera\u021biune se realizeaz\u0103 de mai multe ori pentru o ajustare optim\u0103 a coeficientilor.<\/p>\n

Fi\u0219ierul este disponibil aici, verificat cu un antivirus.<\/a><\/noindex> Activa\u021bi macrocomenzile pentru o func\u021bionare corect\u0103<\/p>\n

3. C\u0103derea gradientului \u0219i dinamica schimb\u0103rii erorii<\/h2>\n

\nAici va fi cazul \u00een 4 dimensiuni \u0219i regresia liniar\u0103. Coeficien\u021bii regresiei liniar\u0103 vor fi determina\u021bi pas cu pas prin metoda c\u0103derii gradientului, \u00eencep\u00e2nd cu to\u021bi coeficien\u021bii \u2013 zero. Pe un grafic separat se poate observa dinamica reducerii erorii pe m\u0103sur\u0103 ce coeficien\u021bii sunt ajusta\u021bi tot mai exact. Exist\u0103 posibilitatea de a vizualiza toate cele patru proiec\u021bii \u00een 2 dimensiuni.<\/p>\n

Dac\u0103 set\u0103m un pas prea mare pentru c\u0103derea gradientului, se observ\u0103 c\u0103 de fiecare dat\u0103 vom s\u0103ri peste minim \u0219i vom ajunge la rezultat \u00een mai multe pa\u0219i, de\u0219i, \u00een final, tot vom ajunge (dac\u0103 nu ne vom ridica pasul prea mult \u2013 atunci algoritmul va \u201es\u0103ri\u201d necontrolat). \u0218i graficul dependen\u021bei erorii de pasul itera\u021biei va fi neregulat, nu lin.<\/p>\n

1. Gener\u0103m date, set\u0103m pasul c\u0103derii gradientului<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\n2. Cu un set corect al pasului c\u0103derii gradientului, ajungem lin \u0219i destul de rapid la minim<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\n3. Cu un set incorect al pasului c\u0103derii gradientului s\u0103rim peste maxim, graficul erorii \u2013 neregulat, convergen\u021ba necesit\u0103 mai mul\u021bi pa\u0219i<\/p>\n

\"\u00cenv\u0103\u021barea
\n\u0219i<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\n4. Cu un set complet gre\u0219it al pasului c\u0103derii gradientului ne \u00eendep\u0103rt\u0103m de minim<\/p>\n

\"\u00cenv\u0103\u021barea
\n
\n(Pentru a reproduce procesul cu valorile pasului c\u0103derii gradientului ar\u0103tate \u00een imagini, bifa\u021bi 'datele de referin\u021b\u0103').<\/p>\n

Fi\u0219ierul \u2013 la acest link, trebuie s\u0103 activa\u021bi macrocomenzile, nu sunt virusuri.<\/a><\/noindex><\/p>\n

Cum consider\u0103 comunitatea respectabil\u0103, este acceptabil\u0103 o astfel de simplificare \u0219i metoda de prezentare a materialului? Ar trebui s\u0103 traduc articolul \u00een englez\u0103? <\/b>
\n
Sursa: habr.com<\/a><\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"

\u041d\u0435\u0442, \u043d\u0443 \u044f, \u043a\u043e\u043d\u0435\u0447\u043d\u043e, \u043d\u0435 \u0432\u0441\u0435\u0440\u044c\u0435\u0437. \u0414\u043e\u043b\u0436\u0435\u043d \u0436\u0435 \u0431\u044b\u0442\u044c \u043f\u0440\u0435\u0434\u0435\u043b, \u0434\u043e \u043a\u0430\u043a\u043e\u0439 \u0441\u0442\u0435\u043f\u0435\u043d\u0438 \u0432\u043e\u0437\u043c\u043e\u0436\u043d\u043e \u0443\u043f\u0440\u043e\u0449\u0430\u0442\u044c \u043f\u0440\u0435\u0434\u043c\u0435\u0442. \u041d\u043e \u0434\u043b\u044f \u043f\u0435\u0440\u0432\u044b\u0445 \u044d\u0442\u0430\u043f\u043e\u0432, \u043f\u043e\u043d\u0438\u043c\u0430\u043d\u0438\u044f \u0431\u0430\u0437\u043e\u0432\u044b\u0445 \u043a\u043e\u043d\u0446\u0435\u043f\u0446\u0438\u0439 \u0438 \u0431\u044b\u0441\u0442\u0440\u043e\u0433\u043e \u00ab\u0432\u044a\u0435\u0437\u0436\u0430\u043d\u0438\u044f\u00bb \u0432 \u0442\u0435\u043c\u0443, \u043c\u043e\u0436\u0435\u0442 \u0431\u044b\u0442\u044c, \u0438 \u0434\u043e\u043f\u0443\u0441\u0442\u0438\u043c\u043e. \u0410 \u043a\u0430\u043a \u043f\u0440\u0430\u0432\u0438\u043b\u044c\u043d\u043e \u043f\u043e\u0438\u043c\u0435\u043d\u043e\u0432\u0430\u0442\u044c \u0434\u0430\u043d\u043d\u044b\u0439 \u043c\u0430\u0442\u0435\u0440\u0438\u0430\u043b (\u0432\u0430\u0440\u0438\u0430\u043d\u0442\u044b: \u00ab\u041c\u0430\u0448\u0438\u043d\u043d\u043e\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0435 \u0434\u043b\u044f \u0447\u0430\u0439\u043d\u0438\u043a\u043e\u0432\u00bb, \u00ab\u0410\u043d\u0430\u043b\u0438\u0437 \u0434\u0430\u043d\u043d\u044b\u0445 \u0441 \u043f\u0435\u043b\u0435\u043d\u043e\u043a\u00bb, \u00ab\u0410\u043b\u0433\u043e\u0440\u0438\u0442\u043c\u044b \u0434\u043b\u044f \u0441\u0430\u043c\u044b\u0445 \u043c\u0430\u043b\u0435\u043d\u044c\u043a\u0438\u0445\u00bb), \u043e\u0431\u0441\u0443\u0434\u0438\u043c \u0432 \u043a\u043e\u043d\u0446\u0435. \u041a […]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":22912,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[702],"tags":[],"class_list":["post-30936","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news"],"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\udd47\u00cenv\u0103\u021barea automat\u0103 f\u0103r\u0103 Python, Anaconda \u0219i altele asemenea | ProHoster","description":"Nu, bine\u00een\u021beles c\u0103 nu vorbesc serios. Trebuie s\u0103 existe o limit\u0103, p\u00e2n\u0103 la ce punct este posibil s\u0103 simplifici subiectul. Dar pentru primele etape, \u00een\u021belegerea conceptelor de baz\u0103 \u0219i o intrare rapid\u0103 \u00een tem\u0103, poate c\u0103 este totu\u0219i acceptabil.","canonical_url":"https:\/\/prohoster.info\/ro\/blog\/news\/mashinnoe-obuchenie-bez-python-anaconda-i-prochih-presmykayushhihsya","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\u041c\u0430\u0448\u0438\u043d\u043d\u043e\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0435 \u0431\u0435\u0437 Python, Anaconda \u0438 \u043f\u0440\u043e\u0447\u0438\u0445 \u043f\u0440\u0435\u0441\u043c\u044b\u043a\u0430\u044e\u0449\u0438\u0445\u0441\u044f | ProHoster","og:description":"\u041d\u0435\u0442, \u043d\u0443 \u044f, \u043a\u043e\u043d\u0435\u0447\u043d\u043e, \u043d\u0435 \u0432\u0441\u0435\u0440\u044c\u0435\u0437. \u0414\u043e\u043b\u0436\u0435\u043d \u0436\u0435 \u0431\u044b\u0442\u044c \u043f\u0440\u0435\u0434\u0435\u043b, \u0434\u043e \u043a\u0430\u043a\u043e\u0439 \u0441\u0442\u0435\u043f\u0435\u043d\u0438 \u0432\u043e\u0437\u043c\u043e\u0436\u043d\u043e \u0443\u043f\u0440\u043e\u0449\u0430\u0442\u044c \u043f\u0440\u0435\u0434\u043c\u0435\u0442. \u041d\u043e \u0434\u043b\u044f \u043f\u0435\u0440\u0432\u044b\u0445 \u044d\u0442\u0430\u043f\u043e\u0432, \u043f\u043e\u043d\u0438\u043c\u0430\u043d\u0438\u044f \u0431\u0430\u0437\u043e\u0432\u044b\u0445 \u043a\u043e\u043d\u0446\u0435\u043f\u0446\u0438\u0439 \u0438 \u0431\u044b\u0441\u0442\u0440\u043e\u0433\u043e \u00ab\u0432\u044a\u0435\u0437\u0436\u0430\u043d\u0438\u044f\u00bb \u0432 \u0442\u0435\u043c\u0443, \u043c\u043e\u0436\u0435\u0442 \u0431\u044b\u0442\u044c, \u0438 \u0434\u043e\u043f\u0443\u0441\u0442\u0438\u043c\u043e.","og:url":"https:\/\/prohoster.info\/ro\/blog\/news\/mashinnoe-obuchenie-bez-python-anaconda-i-prochih-presmykayushhihsya","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":"2019-10-31T18:38:20+00:00","article:modified_time":"2019-10-31T18:38:20+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"30936","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":"2026-01-21 03:45:19","breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2021-03-01 03:27:05","updated":"2026-01-21 03:45:19","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\/30936","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=30936"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/30936\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/media\/22912"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/media?parent=30936"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/categories?post=30936"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/tags?post=30936"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}