{"id":12684,"date":"2022-04-26T12:10:01","date_gmt":"2022-04-26T12:10:01","guid":{"rendered":"https:\/\/stacczero.local\/?p=12684"},"modified":"2025-01-16T19:19:58","modified_gmt":"2025-01-16T19:19:58","slug":"otsustuspuul-pohinevad-masinoppemeetodid","status":"publish","type":"post","link":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/","title":{"rendered":"Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid"},"content":{"rendered":"\n<p>Masin\u00f5ppemeetodi valiku juures m\u00e4ngivad rolli mitmed aspektid, n\u00e4iteks see, millised on andmed, millel meetodit rakendada plaanitakse, aga ka see, milliseid ressursse on v\u00f5imalik mudeli treenimiseks kasutada ning millised on n\u00f5uded treenimiskiirusele, treenitud mudeli suurusele v\u00f5i ennustuskiirusele. Lisaks v\u00f5ib olla oluline, et masin\u00f5ppemudelit saaks lihtsasti t\u00f5lgendada ehk mudeli tehtud otsused oleksid selgitatavad.<\/p>\n\n\n\n<p><strong>Otsustuspuul<\/strong> (ingl <em>decision tree<\/em>) p\u00f5hinevaid masin\u00f5ppemeetodeid kasutatakse nii klassifitseerimise (diskreetsete kategooriate ennustamise) kui ka regressiooni (pidevate v\u00e4\u00e4rtuste ennustamise) jaoks. Seda t\u00fc\u00fcpi meetodeid rakendatakse paljudes valdkondades,<strong> <\/strong>n\u00e4iteks <strong>linnaelanike eelistatud liikumisviisi ennustamiseks<\/strong> <span class=\"footnote_referrer\"><a role=\"button\" tabindex=\"0\" onclick=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_1');\" onkeypress=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_1');\" ><sup id=\"footnote_plugin_tooltip_12684_1_1\" class=\"footnote_plugin_tooltip_text\">[1]<\/sup><\/a><span id=\"footnote_plugin_tooltip_text_12684_1_1\" class=\"footnote_tooltip\"><a href=\"https:\/\/www.mdpi.com\/1424-8220\/15\/7\/15974\/htm\"><em>\u201cSmart City Mobility Application\u2014Gradient Boosting Trees for Mobility Prediction and Analysis Based on Crowdsourced Data\u201d<\/em> <\/a>(Semanjski 2015) <\/span><\/span><script type=\"text\/javascript\"> jQuery('#footnote_plugin_tooltip_12684_1_1').tooltip({ tip: '#footnote_plugin_tooltip_text_12684_1_1', tipClass: 'footnote_tooltip', effect: 'fade', predelay: 0, fadeInSpeed: 200, delay: 400, fadeOutSpeed: 200, position: 'top center', relative: true, offset: [-7, 0], });<\/script> v\u00f5i <strong>molekulaarsete deskriptorite valimiseks ravimite v\u00e4ljat\u00f6\u00f6tamisel<\/strong> <span class=\"footnote_referrer\"><a role=\"button\" tabindex=\"0\" onclick=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_2');\" onkeypress=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_2');\" ><sup id=\"footnote_plugin_tooltip_12684_1_2\" class=\"footnote_plugin_tooltip_text\">[2]<\/sup><\/a><span id=\"footnote_plugin_tooltip_text_12684_1_2\" class=\"footnote_tooltip\"><a href=\"https:\/\/www.sciencedirect.com\/science\/article\/pii\/S0957417416306819?casa_token=Jc23xnqTRuUAAAAA:CDmO8jlPJEOEi_YgSlpVvn5eYu_vHOfYYROPGZBs6k31Mw6Jz9DCkMEkMqBOJ3ZxIpI951Ks\"><em>\u201cAutomatic Selection of Molecular Descriptors Using Random Forest: Application to Drug Discovery\u201d<\/em><\/a> (Cano 2017) <\/span><\/span><script type=\"text\/javascript\"> jQuery('#footnote_plugin_tooltip_12684_1_2').tooltip({ tip: '#footnote_plugin_tooltip_text_12684_1_2', tipClass: 'footnote_tooltip', effect: 'fade', predelay: 0, fadeInSpeed: 200, delay: 400, fadeOutSpeed: 200, position: 'top center', relative: true, offset: [-7, 0], });<\/script>. Kuidas sellised meetodid t\u00f6\u00f6tavad, mis on nende eelised ja puudused ning kuidas nende vahel valida?<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"kuidas-tootavad-erinevad-puupohised-algoritmid\">Kuidas t\u00f6\u00f6tavad erinevad puup\u00f5hised algoritmid?<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"otsustuspuu\">Otsustuspuu<\/h3>\n\n\n\n<p>Lihtsaim puup\u00f5hine masin\u00f5ppemeetod on \u00fcksik <strong>otsustuspuu<\/strong>, mis j\u00f5uab ennustuseni sisuliselt sisendi kohta jah-ei k\u00fcsimusi k\u00fcsides. Mudeli ehitamiseks otsib algoritm igal sammul k\u00fcsimust, mis annab otsuse tegemiseks k\u00f5ige enam infot.<\/p>\n\n\n\n<p>Otsustuspuu treenimine l\u00f5peb kas siis, kui mudel on suutnud perfektselt \u00e4ra jaotada kogu treeningandmed, v\u00f5i siis, kui mudel on j\u00f5udnud talle seatud peatumistingimusteni. N\u00e4iteks saab piirata puu s\u00fcgavust ehk seda, kui mitme k\u00fcsimusega peab vastuseni j\u00f5udma. Mida s\u00fcgavam on otsustuspuu, seda komplekssem on mudel ja seda paremini j\u00e4ljendab mudel treeningandmeid, ent liiga t\u00e4pne j\u00e4ljendamine ehk <strong>\u00fclesobitamine<\/strong> (ingl <em>overfitting<\/em>) viib kehvemate ennustusteni uutel andmetel.<\/p>\n\n\n\n<p><img fetchpriority=\"high\" decoding=\"async\" src=\"https:\/\/lh4.googleusercontent.com\/Bq6TdvS1Eh5ASkRyh1qz4M7iYzGfYSJNHkmQiz64Tt9ks8Wgc224qwHD-yzza1HppIauyPPCzeTKg_hDBm9IKZ3sx5WZ6g9GStAvZA2tpq3pHldfTxRuOj9iEv1ef21OnUz7bQbS\" width=\"624\" height=\"663\"><br><strong>Joonis 1<\/strong>. Otsustuspuu<\/p>\n\n\n\n<p>Joonisel 1 on toodud n\u00e4ide otsustuspuust, mis saab sisendiks ilmaolusid kirjeldavate tunnuste v\u00e4\u00e4rtused. Graafil \u00fclalt alla liikudes on selge, millistel juhtudel j\u00f5uab mudel \u00fche v\u00f5i teise ennustuseni (kas m\u00e4ngida tennist v\u00f5i mitte). Selline justkui tagurpidi p\u00f6\u00f6ratud puu, mille juur on \u00fcleval ja lehed all, on levinud viis otsustuspuude kujutamiseks. Otsustuspuude kontekstis t\u00e4hendab juur esimest otsustuspunkti, millest erinevad teekonnad lehtedeni edasi hargnema hakkavad, ning lehed on graafi l\u00f5pps\u00f5lmed, mis m\u00e4\u00e4ravad mudeli ennustuse. Klassifitseerimis\u00fclesande puhul saab ennustuseks see klass, mille treeningn\u00e4iteid kuulub antud lehte rohkem.<\/p>\n\n\n\n<figure class=\"wp-block-table is-style-stripes\"><table><thead><tr><th><\/th><th> Pilvisus <\/th><th>Temperatuur<\/th><th>Niiskus<\/th><th>Tuulisus<\/th><th>Kas m\u00e4ngida tennist?<\/th><\/tr><\/thead><tbody><tr><td><strong>0<\/strong><\/td><td>P\u00e4ikseline <\/td><td>Palav<\/td><td>K\u00f5rge<\/td><td>Tuulevaikne<\/td><td>Ei<\/td><\/tr><tr><td><strong>1<\/strong><\/td><td>P\u00e4ikseline  <\/td><td>Palav <\/td><td>K\u00f5rge <\/td><td>Tuuline<\/td><td>Ei<\/td><\/tr><tr><td><strong>2<\/strong><\/td><td>Pilves<\/td><td>Palav <\/td><td>K\u00f5rge <\/td><td> Tuulevaikne <\/td><td>Jah<\/td><\/tr><tr><td><strong>3<\/strong><\/td><td>Vihmane<\/td><td>Soe<\/td><td>K\u00f5rge <\/td><td> Tuulevaikne <\/td><td>Jah <\/td><\/tr><tr><td><strong>4<\/strong><\/td><td> Vihmane <\/td><td>Jahe<\/td><td>Keskmine<\/td><td> Tuulevaikne <\/td><td>Jah <\/td><\/tr><tr><td><strong>5<\/strong><\/td><td> Vihmane <\/td><td> Jahe <\/td><td>Keskmine <\/td><td> Tuuline <\/td><td>Ei<\/td><\/tr><tr><td><strong>6<\/strong><\/td><td> Pilves <\/td><td> Jahe <\/td><td>Keskmine <\/td><td> Tuuline <\/td><td>Jah <\/td><\/tr><tr><td><strong>7<\/strong><\/td><td> P\u00e4ikseline  <\/td><td> Soe <\/td><td>K\u00f5rge<\/td><td> Tuulevaikne <\/td><td>Ei<\/td><\/tr><tr><td><strong>8<\/strong><\/td><td> P\u00e4ikseline  <\/td><td> Jahe <\/td><td>Keskmine <\/td><td> Tuulevaikne <\/td><td>Jah <\/td><\/tr><tr><td><strong>9<\/strong><\/td><td> Vihmane <\/td><td> Soe <\/td><td>Keskmine <\/td><td> Tuulevaikne <\/td><td>Jah <\/td><\/tr><tr><td><strong>10<\/strong><\/td><td> P\u00e4ikseline  <\/td><td> Soe <\/td><td>Keskmine <\/td><td> Tuuline <\/td><td>Jah <\/td><\/tr><tr><td><strong>11<\/strong><\/td><td> Pilves <\/td><td> Soe <\/td><td>K\u00f5rge<\/td><td> Tuuline <\/td><td>Jah <\/td><\/tr><tr><td><strong>12<\/strong><\/td><td> Pilves <\/td><td> Palav <\/td><td>Keskmine <\/td><td> Tuulevaikne <\/td><td>Jah <\/td><\/tr><tr><td><strong>13<\/strong><\/td><td> Vihmane <\/td><td> Soe <\/td><td>K\u00f5rge<\/td><td> Tuuline <\/td><td>Ei<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p><strong>Joons 2.<\/strong> Andmestik<\/p>\n\n\n\n<p>N\u00e4itemudeli treenimiseks kasutatud v\u00e4ike n\u00e4idisandmestik on toodud joonisel 2. Tunnused on siin pilvisus, temperatuur, niiskus ja tuulisus. Ennustatav muutuja on kategooriline \u201cKas m\u00e4ngida tennist?\u201d, mist\u00f5ttu on tegemist klassifitseerimis\u00fclesandega.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"parima-kusimuse-leidmine\">Parima k\u00fcsimuse leidmine<\/h3>\n\n\n\n<p>Algoritm otsib igal sammul k\u00fcsimust (tunnust ja v\u00e4\u00e4rtust), mis k\u00f5ige paremini eraldaks erinevatesse klassidesse kuuluvad treeningn\u00e4ited \u00fcksteisest. Selleks proovitakse l\u00e4bi palju erinevaid v\u00f5imalikke k\u00fcsimusi ja valitakse nende hulgast parim, seejuures on valiku tegemiseks mitu v\u00f5imalust. N\u00e4iteks kasutatakse k\u00fcsimuse headuse hindamiseks <a href=\"https:\/\/victorzhou.com\/blog\/gini-impurity\/\"><strong>Gini ebapuhtust<\/strong><\/a> (ingl <em>Gini impurity<\/em>). See on funktsioon, mis n\u00e4itab, kui t\u00f5en\u00e4oline on, et andmepunkt saab vale prognoosi, kui prognoosida suvaliselt klasside jaotuste p\u00f5hjal \u2013 mida madalam v\u00e4\u00e4rtus, seda parem k\u00fcsimus. Gini ebapuhtus on minimaalne siis, kui k\u00fcsimus jaotab erinevad klassid \u00fcksteisest t\u00e4ielikult ja selle tulemusena tekivad n-\u00f6 <strong>puhtad<\/strong> <strong>lehed <\/strong>(ingl <em>pure nodes<\/em>), millega puu antud haru l\u00f5peb.<\/p>\n\n\n\n<p>F\u00fc\u00fcsikatunnist v\u00f5ib olla tuttav <strong>entroopia<\/strong> m\u00f5iste, mis m\u00f5\u00f5dab s\u00fcsteemi korratust v\u00f5i korrastamatust. Sarnaselt on entroopia m\u00f5iste kasutusel <a href=\"https:\/\/machinelearningmastery.com\/what-is-information-entropy\/\">informatsiooniteoorias<\/a>, kus see m\u00f5\u00f5dab andmete ebam\u00e4\u00e4rasust. Otsustuspuu treenimise k\u00e4igus kasutatakse ka entroopiat headuse kriteeriumina. Sellisel juhul p\u00fc\u00fcab algoritm leida sellise k\u00fcsimuse, millele vastamine v\u00e4hendaks entroopiat ehk teisis\u00f5nu annaks uut informatsiooni v\u00f5imalikult palju.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"otsustuspuud-teisteks-ulesanneteks\">Otsustuspuud teisteks \u00fclesanneteks<\/h3>\n\n\n\n<p>Eeltoodud n\u00e4ites oli otsustuspuu \u00fclesandeks kahe klassiga ehk binaarne klassifitseerimine, kuid otsustuspuid on v\u00f5imalik sama lihtsasti kasutada ka enamate klasside puhul. Lisaks on selle meetodiga v\u00f5imalik lahendada regressiooni\u00fclesandeid, millisel juhul nimetatakse treenitud mudelit ka <strong>regressioonipuuks<\/strong>.&nbsp;<\/p>\n\n\n\n<p>Kuna regressiooni puhul ennustatakse pidevaid v\u00e4\u00e4rtusi, tuleb kasutada teistsuguseid kriteeriume parima k\u00fcsimuse leidmiseks. \u00dcks valik on n\u00e4iteks <a href=\"https:\/\/scikit-learn.org\/stable\/modules\/model_evaluation.html#mean-squared-error\"><em>Mean Squared Error<\/em> ehk MSE<\/a>, mille eesm\u00e4rk on minimeerida erinevust ennustatud ja tegelike v\u00e4\u00e4rtuste vahel. Selleks leitakse, millise k\u00fcsimuse puhul jagunevad treeningn\u00e4ited nii, et erinevus \u00fchte gruppi j\u00e4\u00e4vate n\u00e4idete keskmise v\u00e4\u00e4rtuse ja nende tegelike v\u00e4\u00e4rtuste vahel oleks minimaalne.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"eelised-ja-puudused\">Eelised ja puudused<\/h3>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-plus-sign.png\" alt=\"\" class=\"wp-image-12726\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Lihtsasti t\u00f5lgendatav<br><\/strong>Otsustuspuu suureks eeliseks t\u00e4psuse poolest v\u00f5imekamate konkurentide ees on selle t\u00f5lgendatavus: puumudeli saab \u00fcks-\u00fchele visualiseerida graafina, millelt on lihtne v\u00e4lja lugeda, kuidas mudel otsuseni j\u00f5udis \u2013 v\u00e4hemalt seni, kuni otsustuspuu on hoomatava s\u00fcgavusega.<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-plus-sign.png\" alt=\"\" class=\"wp-image-12726\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Ei vaja palju t\u00f6\u00f6d andmete ettevalmistamiseks<br><\/strong><meta charset=\"utf-8\">Otsustuspuu suudab toime tulla erinevate andmet\u00fc\u00fcpidega, n\u00e4iteks nii arvuliste kui kategooriliste v\u00e4\u00e4rtustega (see v\u00f5ib siiski s\u00f5ltuda implementatsioonist), ning ei n\u00f5ua erinevalt m\u00f5nest teisest meetodist andmete eelnevat normaliseerimist.<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-minus-sign.png\" alt=\"\" class=\"wp-image-12737\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Oht treeningandmetel \u00fclesobitada<\/strong><br><meta charset=\"utf-8\">Nagu eespool mainitud, on otsustuspuul oht treeningandmed detailideni \u201ep\u00e4he \u00f5ppida\u201d, mist\u00f5ttu v\u00e4heneb mudeli \u00fcldistusv\u00f5ime<strong> <\/strong>ehk oskus teha kvaliteetseid ennustusi uutele andmetele.<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-minus-sign.png\" alt=\"\" class=\"wp-image-12737\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Astmelised, mitte sujuvad ennustused<\/strong><br><meta charset=\"utf-8\">Kuna otsustuspuu hargnemiskohtades s\u00f5ltub haru valimine k\u00fcsimusest (nt \u201cKas temperatuur on &lt;20\u00b0C?\u201d) , v\u00f5ib juhtuda, et tunnuse v\u00e4\u00e4rtuse oluline suurenemine (nt 10 kraadilt 19 kraadile) ei muuda otsustuspuu ennustust, aga seej\u00e4rel v\u00e4ike lisa (nt 19-lt 20-le) p\u00f6\u00f6rab ennustuse vastupidiseks.&nbsp;<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-minus-sign.png\" alt=\"\" class=\"wp-image-12737\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Kergelt m\u00f5jutatav muutustest treeningandmestikus<\/strong><br>Treenitud mudelid v\u00f5ivad olla \u00fcksteisest v\u00e4ga erinevad, isegi kui treeningandmestikud on peaaegu identsed \u2013 v\u00e4ikestel muutustel andmestikus v\u00f5ib olla suur m\u00f5ju l\u00f5ppmudeli struktuurile ja seel\u00e4bi ka ennustustele.&nbsp;<\/p>\n<\/div>\n<\/div>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"otsustusmets-aka-juhuslik-mets\">Otsustusmets aka juhuslik mets<\/h3>\n\n\n\n<p><strong>Otsustusmets<\/strong> (ingl <em>random forest<\/em>) liigitub <strong>ansambel\u00f5ppe <\/strong>(ingl <em>ensemble learning<\/em>) alla, kus l\u00f5ppennustus moodustub kombinatsioonina mitme mudeli ennustusest. Otsustusmetsa (joonis 3) puhul kuulub ansamblisse hulk otsustuspuid, millest iga\u00fcks on treenitud erineval <em>bootstrap<\/em>\u2019itud valimil treeningandmestikust.&nbsp;<\/p>\n\n\n\n<p><em>Bootstrappimine<\/em> t\u00e4hendab, et originaalsest treeningandmestikust suurusega <em>n<\/em> v\u00f5etakse juhuslikult <em>n<\/em> treeningn\u00e4idet, seejuures p\u00e4rast iga v\u00f5tmist asetatakse valitud n\u00e4ide ka tagasi \u2013 seega v\u00f5ib \u00fcks ja sama n\u00e4ide valimisse j\u00f5uda mitu korda, samal ajal m\u00f5ni teine sealt \u00fcldse puududa.<\/p>\n\n\n\n<p>Lisaks on iga otsustuspuu treenimisel piiratud arv juhuslikult valitud tunnuseid, mille alusel teha puus j\u00e4rgmine jaotus ehk k\u00fcsida j\u00e4rgmine k\u00fcsimus. M\u00f5lema mainitud erip\u00e4ra eesm\u00e4rk on lisada mudelitesse juhuslikkust, mille abil suureneb erinevus ansamblisse kuuluvate \u00fcksikute otsustuspuude vahel. Kui l\u00f5ppennustuse saamiseks nende erinevate mudelite ennustused kombineeritakse, saavutatakse parem \u00fcldistusv\u00f5ime kui \u00fcksikut otsustuspuud kasutades. Teisis\u00f5nu \u2013 v\u00e4heneb risk \u00fclesobitamiseks treeningandmestikul.<\/p>\n\n\n\n<p>Klassifitseerimis\u00fclesande puhul valitakse l\u00f5ppennustuseks see, mis on k\u00f5igi ansamblisse kuuluvate mudelite ennustuste seas \u00fclekaalus, regressiooni\u00fclesande puhul saab l\u00f5ppennustuseks k\u00f5igi mudelite ennustuste keskmine v\u00e4\u00e4rtus.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/lh6.googleusercontent.com\/SjFitBHl-3xxmDH3n3OTcbcd2ZFJuialU47HHhi8503qqi-ZLb5zaqBo6UHLaZlQGkRtCW110AQ0Chdzz0p7vh9FG3YUXbBGhOBoPkloH3khE6PVyDNYS5Fen0m9dSW15WJzNLT3\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>Joonis 3.<\/strong> Otsustusmets<br>Allikas: <a href=\"https:\/\/blog.toadworld.com\/2018\/08\/31\/random-forest-machine-learning-in-r-python-and-sql-part-1\">Random Forest Machine Learning in R, Python and SQL &#8211; Part 1<\/a><\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"eelised-ja-puudused\">Eelised ja puudused<\/h3>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-plus-sign.png\" alt=\"\" class=\"wp-image-12726\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Ei vaja palju t\u00f6\u00f6d andmete ettevalmistamiseks<\/strong><br><meta charset=\"utf-8\">Kuna otsustusmets koosneb otsustuspuudest, kehtib sama eelis v\u00e4hese n\u00f5udlikkuse kohta andmete ettevalmistamise suhtes ka siin.&nbsp;<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-plus-sign.png\" alt=\"\" class=\"wp-image-12726\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong><meta charset=\"utf-8\">Parem \u00fcldistusv\u00f5ime kui \u00fcksikul otsustuspuul<\/strong><br>M\u00e4rkimisv\u00e4\u00e4rne erinevus on eelmainitud parem \u00fcldistusv\u00f5ime uutel andmetel, mis saavutatakse hulga mudelite ansamblisse koondamise ja nende ennustuste kombineerimise abil.<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-plus-sign.png\" alt=\"\" class=\"wp-image-12726\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong><meta charset=\"utf-8\">Parem t\u00f6\u00f6kindlus suure arvu tunnuste korral<\/strong><br><meta charset=\"utf-8\">Otsustusmets tuleb paremini toime andmestikega, kus on suhteliselt<strong> <\/strong>palju tunnuseid \u2013 \u00fcksik otsustuspuu on sellisel juhul eriti aldis \u00fclesobitamisele.<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-minus-sign.png\" alt=\"\" class=\"wp-image-12737\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Raskemini t\u00f5lgendatav kui otsustuspuu<\/strong><br><meta charset=\"utf-8\">Kuna l\u00f5ppennustuses osalevad k\u00f5ik ansambli mudelid (neid v\u00f5ib olla sadades v\u00f5i enamgi), ei ole iga mudeli graafina kujutamine ja sealt ennustusk\u00e4igu v\u00e4lja lugemine enam nii lihtne. Anal\u00fc\u00fcsides iga \u00fcksiku mudeli struktuuri, on v\u00f5imalik teha j\u00e4reldusi selle kohta, millised tunnused treeningandmestikus m\u00f5jutasid ansamblit kokkuv\u00f5ttes enim, kuid \u00fcksikule uuele n\u00e4itele tehtud ennustust on siiski keeruline lahti seletada. Selliste \u201cmusta kasti\u201d t\u00fc\u00fcpi (ingl <em>black box<\/em>) mudelite ennustuste selgitamine on omaette uurimissuund <span class=\"footnote_referrer\"><a role=\"button\" tabindex=\"0\" onclick=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_3');\" onkeypress=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_3');\" ><sup id=\"footnote_plugin_tooltip_12684_1_3\" class=\"footnote_plugin_tooltip_text\">[3]<\/sup><\/a><span id=\"footnote_plugin_tooltip_text_12684_1_3\" class=\"footnote_tooltip\"><a href=\"https:\/\/arxiv.org\/abs\/1705.08504\">\u201c<em>Interpreting Blackbox Models via Model Extraction<\/em>\u201d <\/a>(Bastani 2017) <\/span><\/span><script type=\"text\/javascript\"> jQuery('#footnote_plugin_tooltip_12684_1_3').tooltip({ tip: '#footnote_plugin_tooltip_text_12684_1_3', tipClass: 'footnote_tooltip', effect: 'fade', predelay: 0, fadeInSpeed: 200, delay: 400, fadeOutSpeed: 200, position: 'top center', relative: true, offset: [-7, 0], });<\/script> <span class=\"footnote_referrer\"><a role=\"button\" tabindex=\"0\" onclick=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_4');\" onkeypress=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_4');\" ><sup id=\"footnote_plugin_tooltip_12684_1_4\" class=\"footnote_plugin_tooltip_text\">[4]<\/sup><\/a><span id=\"footnote_plugin_tooltip_text_12684_1_4\" class=\"footnote_tooltip\"><a href=\"https:\/\/arxiv.org\/pdf\/1312.1121.pdf\"><em>\u201cInterpreting Random Forest Classification Models Using a Feature Contribution Method\u201d<\/em> <\/a>(Palczewska 2014) <\/span><\/span><script type=\"text\/javascript\"> jQuery('#footnote_plugin_tooltip_12684_1_4').tooltip({ tip: '#footnote_plugin_tooltip_text_12684_1_4', tipClass: 'footnote_tooltip', effect: 'fade', predelay: 0, fadeInSpeed: 200, delay: 400, fadeOutSpeed: 200, position: 'top center', relative: true, offset: [-7, 0], });<\/script>.<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-minus-sign.png\" alt=\"\" class=\"wp-image-12737\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong><meta charset=\"utf-8\">Ressursin\u00f5udlikum<\/strong><br><meta charset=\"utf-8\">Ansambel\u00f5ppemudelitega kaasneb ka k\u00fcsimus arvutuslike ressursside kohta. N\u00e4iteks saja mudeliga otsustusmets vajab rohkem salvestusruumi kui \u00fcksik otsustuspuu. Samuti muutub treenimine ja ennustamine aeglasemaks. Otsustusmetsa kasuks r\u00e4\u00e4gib siiski see, et ansamblisse kuuluvaid mudeleid saab treenida ja hiljem ka rakendada \u00fcksteisest s\u00f5ltumatult, mist\u00f5ttu on v\u00f5imalik neid protsesse paralleliseerida.<\/p>\n<\/div>\n<\/div>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"gradient-boosting\"><em>Gradient boosting<\/em><\/h3>\n\n\n\n<p><strong><em>Gradient boosting<\/em><\/strong><em> <\/em>on teine ansambel\u00f5ppe alla kuuluv masin\u00f5ppemeetod, mis koosneb sageli samuti otsustuspuudest. Erinevalt otsustusmetsa meetodist toimub siin mudelite treenimine <strong>j\u00e4rjestikuselt, mitte paralleelselt<\/strong> \u2013 see t\u00e4hendab, et iga j\u00e4rgnev mudel s\u00f5ltub eelmistest. Ansamblit rakendades moodustub l\u00f5ppennustus k\u00f5igi ansamblisse kuuluvate mudelite v\u00e4ljundite kokkuliitmise teel (joonis 4).<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/lh5.googleusercontent.com\/eGzlA-71zkTSGMd-t9t0cOtV2pdbBL4nLulXSiPJpm0NhGpg-npSripD2j9LXHXVsMbXt1VQt8JaW9ouTOhXtGlUV1zS3uob0s1zzkyYEHv9ZgxTm57oiN4kRqj3oOOqpw7kMXUv\" alt=\"\"\/><\/figure>\n\n\n\n<p><strong>Joonis 4. <\/strong><em>Gradient boosting<\/em><br>Allikas: <a href=\"http:\/\/arogozhnikov.github.io\/images\/gbdt_attractive_picture.png\">http:\/\/arogozhnikov.github.io\/images\/gbdt_attractive_picture.png<\/a><\/p>\n\n\n\n<p><em>Boosting<\/em>-meetodite p\u00f5hiidee seisneb selles, et treenitakse hulk \u201cn\u00f5rku ennustajaid\u201d (ingl <em>weak learners<\/em>), kusjuures iga j\u00e4rgneva mudeli treenimisel p\u00fc\u00fctakse parandada eelnevate mudelite vigu. Eesm\u00e4rk on saavutada n\u00f5rkade ennustajate nutika kombineerimise tulemusena tugev ennustaja. N\u00f5rgaks ennustajaks loetakse masin\u00f5ppemudelit, mille ennustust\u00e4psus on vaid veidi parem juhuslikust ennustamisest. N\u00e4iteks kahe klassiga klassifitseerimise puhul oleks juhusliku ennustaja t\u00e4psus 50%, seega n\u00f5rga ennustaja t\u00e4psus peab olema sellest k\u00f5rgem.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"gradientlaskumine\">Gradientlaskumine<\/h3>\n\n\n\n<p>Millele viitab s\u00f5na <em>gradient <\/em>meetodi nimes? <strong>Gradientlaskumine<\/strong> (ingl <em>gradient descent<\/em>) on masin\u00f5ppes levinud optimeerimisalgoritm, mille eesm\u00e4rk on minimeerida kahju ehk erinevust ennustatud ja tegelike v\u00e4\u00e4rtuste vahel. Tehisn\u00e4rviv\u00f5rkudes kasutatakse gradientlaskumist mudeli parameetrite uuendamiseks treenimise k\u00e4igus. Selleks leitakse, kuidas iga parameeter m\u00f5jutab kahju v\u00e4\u00e4rtust, arvutades kahju leidmiseks kasutatava funktsiooni ehk kaofunktsiooni osatuletised iga parameetri suhtes. Need osatuletised moodustavadki gradiendi ning n\u00e4itavad, millises suunas tuleks parameetreid muuta, et kaofunktsioon k\u00f5ige j\u00e4rsemalt kasvaks. Kuna eesm\u00e4rk on kaofunktsiooni mitte maksimeerida, vaid minimeerida (sellest ka s\u00f5na \u201claskumine\u201d), siis muudetakse parameetreid gradiendile vastupidises suunas.<\/p>\n\n\n\n<p>\u00dcks varasemaid <em>boosting<\/em>-meetodeid <a href=\"https:\/\/www.sciencedirect.com\/science\/article\/pii\/S002200009791504X\">AdaBoost<\/a> kasutas j\u00e4rjestikuste mudelite treenimiseks treeningn\u00e4idete kaalumise meetodit, mille abil suunata j\u00e4rgneva mudeli treenimisel rohkem t\u00e4helepanu keerulistele juhtudele. <em>Gradient boosting <\/em>meetodis kasutatakse aga kaalumise asemel gradientlaskumist. Sellega leitakse, kuidas senise ansambli ennustused m\u00f5jutavad kaofunktsiooni, ning j\u00e4rgmise mudeli treenimisel p\u00fc\u00fctakse selle ennustusi suunata vastavalt gradiendile.<\/p>\n\n\n\n<h3 class=\"has-brand-purple-color has-text-color wp-block-heading\" id=\"eelised-ja-puudused\">Eelised ja puudused<\/h3>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-plus-sign.png\" alt=\"\" class=\"wp-image-12726\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Sama hea v\u00f5i parem t\u00e4psus kui otsustusmetsal<\/strong><br><meta charset=\"utf-8\"><em>Gradient boosting <\/em>mudelid on v\u00f5rdlustes n\u00e4idanud sama head v\u00f5i paremat t\u00e4psust kui otsustusmetsa mudelid, mis kinnitab <em>boosting<\/em>-meetodi efektiivsust <span class=\"footnote_referrer\"><a role=\"button\" tabindex=\"0\" onclick=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_5');\" onkeypress=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_5');\" ><sup id=\"footnote_plugin_tooltip_12684_1_5\" class=\"footnote_plugin_tooltip_text\">[5]<\/sup><\/a><span id=\"footnote_plugin_tooltip_text_12684_1_5\" class=\"footnote_tooltip\"><a href=\"https:\/\/www.sciencedirect.com\/science\/article\/pii\/S0957417417302397?casa_token=pG6Xp0Gz414AAAAA:f0g0PGZOJgpmvQGrKJWcT78ZZb7VLwwnarGz0GYoYXjQ6blcaFGQR4G3rp2NGzb_8TIwQSYP\"><em>\u201cAn Up-to-Date Comparison of State-of-the-Art Classification Algorithms\u201d<\/em><\/a><em> <\/em>(Zhang 2017) <\/span><\/span><script type=\"text\/javascript\"> jQuery('#footnote_plugin_tooltip_12684_1_5').tooltip({ tip: '#footnote_plugin_tooltip_text_12684_1_5', tipClass: 'footnote_tooltip', effect: 'fade', predelay: 0, fadeInSpeed: 200, delay: 400, fadeOutSpeed: 200, position: 'top center', relative: true, offset: [-7, 0], });<\/script> <span class=\"footnote_referrer\"><a role=\"button\" tabindex=\"0\" onclick=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_6');\" onkeypress=\"footnote_moveToReference_12684_1('footnote_plugin_reference_12684_1_6');\" ><sup id=\"footnote_plugin_tooltip_12684_1_6\" class=\"footnote_plugin_tooltip_text\">[6]<\/sup><\/a><span id=\"footnote_plugin_tooltip_text_12684_1_6\" class=\"footnote_tooltip\"><a href=\"https:\/\/www.sciencedirect.com\/science\/article\/pii\/S1386505617302368?casa_token=U3zCfbpeoJcAAAAA:URCBRY9aUxhahYz5TXRGDP39hfOuzYKLDobkND6KoM72Ij_krurqguxiKtNWpGy00wfHDCvx\"><em>\u201cPrediction of Lung Cancer Patient Survival via Supervised Machine Learning Classification Techniques\u201d<\/em><\/a> (Lynch 2017) <\/span><\/span><script type=\"text\/javascript\"> jQuery('#footnote_plugin_tooltip_12684_1_6').tooltip({ tip: '#footnote_plugin_tooltip_text_12684_1_6', tipClass: 'footnote_tooltip', effect: 'fade', predelay: 0, fadeInSpeed: 200, delay: 400, fadeOutSpeed: 200, position: 'top center', relative: true, offset: [-7, 0], });<\/script>.<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-plus-sign.png\" alt=\"\" class=\"wp-image-12726\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Kiirem ennustusprotsess kui otsustusmetsal<\/strong><br><meta charset=\"utf-8\">Kuna <em>gradient boosting <\/em>ansambel koosneb enamasti v\u00e4iksematest mudelitest kui otsustusmets, on sellega l\u00f5ppennustuseni j\u00f5udmine kiirem. See v\u00f5imaldab uutele n\u00e4idetele kiiremini reageerida ja ennustusi v\u00e4ljastada. See omadus v\u00f5ib osutuda v\u00e4ga t\u00e4htsaks reaalajas kasutatavate s\u00fcsteemide puhul.<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-minus-sign.png\" alt=\"\" class=\"wp-image-12737\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Raskemini t\u00f5lgendatav kui otsustuspuu<\/strong><br>Sarnaselt otsustusmetsale on ennustused raskemini selgitatavad kui \u00fcksiku otsutuspuu puhul, sest l\u00f5ppennustus moodustub paljude mudelite ennustuste kombineerimise teel.<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-minus-sign.png\" alt=\"\" class=\"wp-image-12737\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Tundlikkus erandlike treeningn\u00e4idete suhtes<\/strong><br><em>Gradient boosting <\/em>meetod on erandlike treeningn\u00e4idete suhtes tundlikum, sest algoritm p\u00fc\u00fcab iga j\u00e4rgneva mudeliga \u00fcha enam raskeid juhtumeid \u00f5igesti ennustada, mis v\u00f5ib viia soovimatute tulemusteni. Samal p\u00f5hjusel m\u00f5jutavad algoritmi negatiivselt valesti m\u00e4rgendatud treeningandmed.<\/p>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:40px\">\n<figure class=\"wp-block-image size-full mt-5\"><img decoding=\"async\" width=\"36\" height=\"36\" src=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/brand-minus-sign.png\" alt=\"\" class=\"wp-image-12737\"\/><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:80%\">\n<p><strong>Ajamahukam treenimine<\/strong><br>Erinevalt otsustusmetsast ei ole <em>gradient boosting <\/em>ansamblisse kuuluvaid mudeleid v\u00f5imalik paralleelselt treenida, sest mudelid on \u00fcksteisest s\u00f5ltuvad.<\/p>\n<\/div>\n<\/div>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"kuidas-puupohiseid-meetodeid-rakendada\">Kuidas puup\u00f5hiseid meetodeid rakendada?<\/h2>\n\n\n\n<p>Nii otsustuspuu, otsustusmetsa kui ka <em>gradient boosting <\/em>algoritme saab \u00fcles seada, kasutades olemasolevaid teeke. N\u00e4iteks <em>gradient boosting <\/em>meetodist on loodud erinevaid variatsioone, millel igal omad eelised ja edasiarendused.<\/p>\n\n\n\n<ul class=\"wp-block-list\"><li><a href=\"https:\/\/scikit-learn.org\/stable\/modules\/tree.html#\">Scikit-learn &#8211; otsustuspuud<\/a><\/li><\/ul>\n\n\n\n<p>Scikit-learn on laialt kasutatav Pythoni teek, mille abil on v\u00f5imalik rakendada paljusid masin\u00f5ppealgoritme. Sellel lehel antakse l\u00fchike \u00fclevaade otsustuspuudest koos praktiliste n\u00f5uannetega Scikit-learni implementatsiooni kasutamiseks.<\/p>\n\n\n\n<ul class=\"wp-block-list\"><li><a href=\"https:\/\/scikit-learn.org\/stable\/modules\/ensemble.html#\">Scikit-learn &#8211; otsustusmetsad ja <em>gradient boosting<\/em><\/a><\/li><\/ul>\n\n\n\n<p>Lisaks tavalisele otsustuspuule toetab Scikit-learn ka otsustusmetsa ja <em>gradient boosting <\/em>meetodite implementeerimist \u2013 siin kirjeldatakse Scikit-learnis saadaolevaid ansambel\u00f5ppemeetodeid ja tuuakse n\u00e4iteid erinevate meetodite kasutamise kohta.<\/p>\n\n\n\n<ul class=\"wp-block-list\"><li><a href=\"https:\/\/xgboost.readthedocs.io\/en\/latest\/\">XGBoost &#8211; <em>gradient boosting<\/em><\/a><\/li><\/ul>\n\n\n\n<p>Nimi XGBoost on l\u00fchend fraasist <em>eXtreme Gradient Boosting<\/em>, mille peamised eelised on kiirus ja skaleeritavus. XGBoost kasutab spetsiaalset algoritmi, mis t\u00f6\u00f6tab h\u00e4sti h\u00f5redatel andmetel (ingl<em> sparse data<\/em>) ehk andmetel, kus on palju puuduvaid v\u00e4\u00e4rtusi v\u00f5i nulle, ning kasutab paralleliseerimist \u00fche puu treenimise l\u00f5ikes.<\/p>\n\n\n\n<ul class=\"wp-block-list\"><li><a href=\"https:\/\/lightgbm.readthedocs.io\/en\/latest\/\">LightGBM &#8211; <em>gradient boosting<\/em><\/a><\/li><\/ul>\n\n\n\n<p>LightGBM teegis on astutud veel \u00fcks samm edasi kiirema treenimise suunas, kasutades kaht uuenduslikku meetodit: gradiendip\u00f5hine treeningandmete valik ja \u00fcksteist v\u00e4listavate tunnuste \u00fchendamine. Esimese eesm\u00e4rk on v\u00e4hendada kirjete arvu, millel \u00fcksikuid mudeleid treenitakse, ning teise eesm\u00e4rk on v\u00e4hendada tunnuste arvu, mille vahel treenitav mudel saab valida. Lisaks treenitakse LightGBM-is mudeleid mitte kihthaaval, vaid lehthaaval ehk j\u00e4rgmine s\u00f5lm, millest puu edasi hargneb, v\u00f5ib asuda \u00fcksk\u00f5ik millisel s\u00fcgavusel \u2013 peaasi, et sealt j\u00e4tkamine tooks v\u00f5imalikult palju kasu.<\/p>\n\n\n\n<ul class=\"wp-block-list\"><li><a href=\"https:\/\/catboost.ai\/\">CatBoost &#8211; <em>gradient boosting<\/em><\/a><\/li><\/ul>\n\n\n\n<p>CatBoost (fraasist <em>categorical boosting<\/em>) toetab mudelite treenimist nii CPU kui ka GPU toel, pakkudes viimasega v\u00f5imalust ressursside olemasolul treenimist kiirendada. Nagu teegi nimi viitab, on selle implementatsiooni juures pandud r\u00f5hku kategooriliste tunnuste toetamisele &#8211; nimelt v\u00f5ib CatBoosti rakendada otse andmestikul, kus on kategoorilisi tunnuseid, ilma et kasutajal oleks tarvis need ise sobivale kujule teisendada. Samuti on selle teegi puhul t\u00f5en\u00e4oline, et h\u00fcperparameetrite (kasutaja poolt seadistatavate parameetrite) vaikev\u00e4\u00e4rtused t\u00f6\u00f6tavad juba \u00fcsna h\u00e4sti, sest need leitakse vastavalt \u00fclesandele, andmestiku suurusele ja muudele n\u00e4itajatele.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"kokkuvote\">Kokkuv\u00f5te<\/h2>\n\n\n\n<p>Puup\u00f5hised masin\u00f5ppemeetodid on leidnud ja leiavad endiselt kasutust paljudes valdkondades, alates finantsvaldkonnas krediidiriski hindamisest kuni transpordis reaalajas liiklusriskide hindamiseni ja m\u00fc\u00fcgivaldkonnas tulemusn\u00e4itajate ennustamiseni. STACCis on otsustusmetsa meetodi abil ennustatud n\u00e4iteks museaalide s\u00e4ilivust, mille kohta saab pikemalt lugeda <a href=\"https:\/\/stacc.veebilahendused.ee\/et\/kuidas-aitas-stacc-muinsuskaitseametil-tehisintellekti-kasutades-museaalide-sailivust-hinnata\/\">siit<\/a>.<\/p>\n\n\n\n<p>Igal masin\u00f5ppemeetodil on omad ise\u00e4rasused, mist\u00f5ttu tasub muu hulgas l\u00e4bi m\u00f5elda, millised on andmed, n\u00f5uded treenitud mudelile ning saadaolevad ressursid. Seej\u00e4rel saab asuda rakendama nendest sobivamaid ja avastada, mida masin\u00f5pe andmetega teha suudab.<\/p>\n\n\n\n<p><strong>M\u00e4rks\u00f5nad:<\/strong> masin\u00f5pe, masin\u00f5ppemeetodid, otsustuspuu, puumudelid, otsustusmets, juhuslik mets, gradient boosting, XGBoost,&nbsp;Scikit-learn, LightGBM, CatBoost, ansambel\u00f5pe.<\/p>\n<div class=\"speaker-mute footnotes_reference_container\"> <div class=\"footnote_container_prepare\"><p><span role=\"button\" tabindex=\"0\" class=\"footnote_reference_container_label pointer\" onclick=\"footnote_expand_collapse_reference_container_12684_1();\">Viited<\/span><span role=\"button\" tabindex=\"0\" class=\"footnote_reference_container_collapse_button\" style=\"display: none;\" onclick=\"footnote_expand_collapse_reference_container_12684_1();\">[<a id=\"footnote_reference_container_collapse_button_12684_1\">+<\/a>]<\/span><\/p><\/div> <div id=\"footnote_references_container_12684_1\" style=\"\"><table class=\"footnotes_table footnote-reference-container\"><caption class=\"accessibility\">Viited<\/caption> <tbody> \r\n\r\n<tr class=\"footnotes_plugin_reference_row\"> <th scope=\"row\" class=\"footnote_plugin_index_combi pointer\"  onclick=\"footnote_moveToAnchor_12684_1('footnote_plugin_tooltip_12684_1_1');\"><a id=\"footnote_plugin_reference_12684_1_1\" class=\"footnote_backlink\"><span class=\"footnote_index_arrow\">&#8593;<\/span>1<\/a><\/th> <td class=\"footnote_plugin_text\"><a href=\"https:\/\/www.mdpi.com\/1424-8220\/15\/7\/15974\/htm\"><em>\u201cSmart City Mobility Application\u2014Gradient Boosting Trees for Mobility Prediction and Analysis Based on Crowdsourced Data\u201d<\/em> <\/a>(Semanjski 2015) <\/td><\/tr>\r\n\r\n<tr class=\"footnotes_plugin_reference_row\"> <th scope=\"row\" class=\"footnote_plugin_index_combi pointer\"  onclick=\"footnote_moveToAnchor_12684_1('footnote_plugin_tooltip_12684_1_2');\"><a id=\"footnote_plugin_reference_12684_1_2\" class=\"footnote_backlink\"><span class=\"footnote_index_arrow\">&#8593;<\/span>2<\/a><\/th> <td class=\"footnote_plugin_text\"><a href=\"https:\/\/www.sciencedirect.com\/science\/article\/pii\/S0957417416306819?casa_token=Jc23xnqTRuUAAAAA:CDmO8jlPJEOEi_YgSlpVvn5eYu_vHOfYYROPGZBs6k31Mw6Jz9DCkMEkMqBOJ3ZxIpI951Ks\"><em>\u201cAutomatic Selection of Molecular Descriptors Using Random Forest: Application to Drug Discovery\u201d<\/em><\/a> (Cano 2017) <\/td><\/tr>\r\n\r\n<tr class=\"footnotes_plugin_reference_row\"> <th scope=\"row\" class=\"footnote_plugin_index_combi pointer\"  onclick=\"footnote_moveToAnchor_12684_1('footnote_plugin_tooltip_12684_1_3');\"><a id=\"footnote_plugin_reference_12684_1_3\" class=\"footnote_backlink\"><span class=\"footnote_index_arrow\">&#8593;<\/span>3<\/a><\/th> <td class=\"footnote_plugin_text\"><a href=\"https:\/\/arxiv.org\/abs\/1705.08504\">\u201c<em>Interpreting Blackbox Models via Model Extraction<\/em>\u201d <\/a>(Bastani 2017) <\/td><\/tr>\r\n\r\n<tr class=\"footnotes_plugin_reference_row\"> <th scope=\"row\" class=\"footnote_plugin_index_combi pointer\"  onclick=\"footnote_moveToAnchor_12684_1('footnote_plugin_tooltip_12684_1_4');\"><a id=\"footnote_plugin_reference_12684_1_4\" class=\"footnote_backlink\"><span class=\"footnote_index_arrow\">&#8593;<\/span>4<\/a><\/th> <td class=\"footnote_plugin_text\"><a href=\"https:\/\/arxiv.org\/pdf\/1312.1121.pdf\"><em>\u201cInterpreting Random Forest Classification Models Using a Feature Contribution Method\u201d<\/em> <\/a>(Palczewska 2014) <\/td><\/tr>\r\n\r\n<tr class=\"footnotes_plugin_reference_row\"> <th scope=\"row\" class=\"footnote_plugin_index_combi pointer\"  onclick=\"footnote_moveToAnchor_12684_1('footnote_plugin_tooltip_12684_1_5');\"><a id=\"footnote_plugin_reference_12684_1_5\" class=\"footnote_backlink\"><span class=\"footnote_index_arrow\">&#8593;<\/span>5<\/a><\/th> <td class=\"footnote_plugin_text\"><a href=\"https:\/\/www.sciencedirect.com\/science\/article\/pii\/S0957417417302397?casa_token=pG6Xp0Gz414AAAAA:f0g0PGZOJgpmvQGrKJWcT78ZZb7VLwwnarGz0GYoYXjQ6blcaFGQR4G3rp2NGzb_8TIwQSYP\"><em>\u201cAn Up-to-Date Comparison of State-of-the-Art Classification Algorithms\u201d<\/em><\/a><em> <\/em>(Zhang 2017) <\/td><\/tr>\r\n\r\n<tr class=\"footnotes_plugin_reference_row\"> <th scope=\"row\" class=\"footnote_plugin_index_combi pointer\"  onclick=\"footnote_moveToAnchor_12684_1('footnote_plugin_tooltip_12684_1_6');\"><a id=\"footnote_plugin_reference_12684_1_6\" class=\"footnote_backlink\"><span class=\"footnote_index_arrow\">&#8593;<\/span>6<\/a><\/th> <td class=\"footnote_plugin_text\"><a href=\"https:\/\/www.sciencedirect.com\/science\/article\/pii\/S1386505617302368?casa_token=U3zCfbpeoJcAAAAA:URCBRY9aUxhahYz5TXRGDP39hfOuzYKLDobkND6KoM72Ij_krurqguxiKtNWpGy00wfHDCvx\"><em>\u201cPrediction of Lung Cancer Patient Survival via Supervised Machine Learning Classification Techniques\u201d<\/em><\/a> (Lynch 2017) <\/td><\/tr>\r\n\r\n <\/tbody> <\/table> <\/div><\/div><script type=\"text\/javascript\"> function footnote_expand_reference_container_12684_1() { jQuery('#footnote_references_container_12684_1').show(); jQuery('#footnote_reference_container_collapse_button_12684_1').text('\u2212'); } function footnote_collapse_reference_container_12684_1() { jQuery('#footnote_references_container_12684_1').hide(); jQuery('#footnote_reference_container_collapse_button_12684_1').text('+'); } function footnote_expand_collapse_reference_container_12684_1() { if (jQuery('#footnote_references_container_12684_1').is(':hidden')) { footnote_expand_reference_container_12684_1(); } else { footnote_collapse_reference_container_12684_1(); } } function footnote_moveToReference_12684_1(p_str_TargetID) { footnote_expand_reference_container_12684_1(); var l_obj_Target = jQuery('#' + p_str_TargetID); if (l_obj_Target.length) { jQuery( 'html, body' ).delay( 0 ); jQuery('html, body').animate({ scrollTop: l_obj_Target.offset().top - window.innerHeight * 0.2 }, 380); } } function footnote_moveToAnchor_12684_1(p_str_TargetID) { footnote_expand_reference_container_12684_1(); var l_obj_Target = jQuery('#' + p_str_TargetID); if (l_obj_Target.length) { jQuery( 'html, body' ).delay( 0 ); jQuery('html, body').animate({ scrollTop: l_obj_Target.offset().top - window.innerHeight * 0.2 }, 380); } }<\/script>","protected":false},"excerpt":{"rendered":"<p>Masin\u00f5ppemeetodi valiku juures m\u00e4ngivad rolli mitmed aspektid, n\u00e4iteks see, millised on andmed, millel meetodit rakendada plaanitakse.<\/p>\n","protected":false},"author":2,"featured_media":12722,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[156,143,79,100],"tags":[133,134],"class_list":["post-12684","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-assistendid","category-energeetika","category-kaubandus","category-toostus","tag-lahendused","tag-suvavaade"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v25.1 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid - STACC<\/title>\n<meta name=\"description\" content=\"Masin\u00f5ppemeetodi valiku juures m\u00e4ngib rolli, millistel andmetel meetodit rakendada plaanitakse ja milliseid ressursse on v\u00f5imalik kasutada.\" \/>\n<meta name=\"robots\" content=\"noindex, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<meta property=\"og:locale\" content=\"et_EE\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid - STACC\" \/>\n<meta property=\"og:description\" content=\"Masin\u00f5ppemeetodi valiku juures m\u00e4ngib rolli, millistel andmetel meetodit rakendada plaanitakse ja milliseid ressursse on v\u00f5imalik kasutada.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/\" \/>\n<meta property=\"og:site_name\" content=\"STACC\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/staccest\" \/>\n<meta property=\"article:published_time\" content=\"2022-04-26T12:10:01+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2025-01-16T19:19:58+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/blogi-pilt-stacc-varvid.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1024\" \/>\n\t<meta property=\"og:image:height\" content=\"593\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Merili Auser\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@staccest1\" \/>\n<meta name=\"twitter:site\" content=\"@staccest1\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Merili Auser\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"13 minutit\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/\"},\"author\":{\"name\":\"Merili Auser\",\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/#\/schema\/person\/10197711b76fb6c7f2c5b32014c3dfbb\"},\"headline\":\"Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid\",\"datePublished\":\"2022-04-26T12:10:01+00:00\",\"dateModified\":\"2025-01-16T19:19:58+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/\"},\"wordCount\":2367,\"publisher\":{\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/#organization\"},\"image\":{\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/blogi-pilt-stacc-varvid.png\",\"keywords\":[\"lahendused\",\"s\u00fcvavaade\"],\"articleSection\":[\"AI-assistendid\",\"Energeetika\",\"Kaubandus\",\"T\u00f6\u00f6stus\"],\"inLanguage\":\"et\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/\",\"url\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/\",\"name\":\"Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid - STACC\",\"isPartOf\":{\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/blogi-pilt-stacc-varvid.png\",\"datePublished\":\"2022-04-26T12:10:01+00:00\",\"dateModified\":\"2025-01-16T19:19:58+00:00\",\"description\":\"Masin\u00f5ppemeetodi valiku juures m\u00e4ngib rolli, millistel andmetel meetodit rakendada plaanitakse ja milliseid ressursse on v\u00f5imalik kasutada.\",\"breadcrumb\":{\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#breadcrumb\"},\"inLanguage\":\"et\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"et\",\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#primaryimage\",\"url\":\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/blogi-pilt-stacc-varvid.png\",\"contentUrl\":\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/blogi-pilt-stacc-varvid.png\",\"width\":1024,\"height\":593},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Avaleht\",\"item\":\"https:\/\/stacc.veebilahendused.ee\/et\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"lahendused\",\"item\":\"https:\/\/stacc.veebilahendused.ee\/et\/tag\/lahendused\/\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/#website\",\"url\":\"https:\/\/stacc.veebilahendused.ee\/et\/\",\"name\":\"STACC\",\"description\":\"Juhtiv masin\u00f5ppe ja andmeteaduse ettev\u00f5te\",\"publisher\":{\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/stacc.veebilahendused.ee\/et\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"et\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/#organization\",\"name\":\"STACC\",\"url\":\"https:\/\/stacc.veebilahendused.ee\/et\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"et\",\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2021\/09\/stacc.jpg\",\"contentUrl\":\"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2021\/09\/stacc.jpg\",\"width\":176,\"height\":176,\"caption\":\"STACC\"},\"image\":{\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/#\/schema\/logo\/image\/\"},\"sameAs\":[\"https:\/\/www.facebook.com\/staccest\",\"https:\/\/x.com\/staccest1\",\"https:\/\/www.instagram.com\/staccest\/\",\"https:\/\/ee.linkedin.com\/company\/staccest\",\"https:\/\/www.youtube.com\/channel\/UCnn59bCz9vSM3lC0S9j6Ccw\"]},{\"@type\":\"Person\",\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/#\/schema\/person\/10197711b76fb6c7f2c5b32014c3dfbb\",\"name\":\"Merili Auser\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"et\",\"@id\":\"https:\/\/stacc.veebilahendused.ee\/et\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/dadc0a881bc7e4aa0559507a4f900136c956049d810c537bad0e30d7127d7e10?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/dadc0a881bc7e4aa0559507a4f900136c956049d810c537bad0e30d7127d7e10?s=96&d=mm&r=g\",\"caption\":\"Merili Auser\"},\"sameAs\":[\"https:\/\/stacc.veebilahendused.ee\"],\"url\":\"https:\/\/stacc.veebilahendused.ee\/et\/author\/merili\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid - STACC","description":"Masin\u00f5ppemeetodi valiku juures m\u00e4ngib rolli, millistel andmetel meetodit rakendada plaanitakse ja milliseid ressursse on v\u00f5imalik kasutada.","robots":{"index":"noindex","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"og_locale":"et_EE","og_type":"article","og_title":"Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid - STACC","og_description":"Masin\u00f5ppemeetodi valiku juures m\u00e4ngib rolli, millistel andmetel meetodit rakendada plaanitakse ja milliseid ressursse on v\u00f5imalik kasutada.","og_url":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/","og_site_name":"STACC","article_publisher":"https:\/\/www.facebook.com\/staccest","article_published_time":"2022-04-26T12:10:01+00:00","article_modified_time":"2025-01-16T19:19:58+00:00","og_image":[{"width":1024,"height":593,"url":"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/blogi-pilt-stacc-varvid.png","type":"image\/png"}],"author":"Merili Auser","twitter_card":"summary_large_image","twitter_creator":"@staccest1","twitter_site":"@staccest1","twitter_misc":{"Written by":"Merili Auser","Est. reading time":"13 minutit"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#article","isPartOf":{"@id":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/"},"author":{"name":"Merili Auser","@id":"https:\/\/stacc.veebilahendused.ee\/et\/#\/schema\/person\/10197711b76fb6c7f2c5b32014c3dfbb"},"headline":"Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid","datePublished":"2022-04-26T12:10:01+00:00","dateModified":"2025-01-16T19:19:58+00:00","mainEntityOfPage":{"@id":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/"},"wordCount":2367,"publisher":{"@id":"https:\/\/stacc.veebilahendused.ee\/et\/#organization"},"image":{"@id":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#primaryimage"},"thumbnailUrl":"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/blogi-pilt-stacc-varvid.png","keywords":["lahendused","s\u00fcvavaade"],"articleSection":["AI-assistendid","Energeetika","Kaubandus","T\u00f6\u00f6stus"],"inLanguage":"et"},{"@type":"WebPage","@id":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/","url":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/","name":"Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid - STACC","isPartOf":{"@id":"https:\/\/stacc.veebilahendused.ee\/et\/#website"},"primaryImageOfPage":{"@id":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#primaryimage"},"image":{"@id":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#primaryimage"},"thumbnailUrl":"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/blogi-pilt-stacc-varvid.png","datePublished":"2022-04-26T12:10:01+00:00","dateModified":"2025-01-16T19:19:58+00:00","description":"Masin\u00f5ppemeetodi valiku juures m\u00e4ngib rolli, millistel andmetel meetodit rakendada plaanitakse ja milliseid ressursse on v\u00f5imalik kasutada.","breadcrumb":{"@id":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#breadcrumb"},"inLanguage":"et","potentialAction":[{"@type":"ReadAction","target":["https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/"]}]},{"@type":"ImageObject","inLanguage":"et","@id":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#primaryimage","url":"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/blogi-pilt-stacc-varvid.png","contentUrl":"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2022\/01\/blogi-pilt-stacc-varvid.png","width":1024,"height":593},{"@type":"BreadcrumbList","@id":"https:\/\/stacc.veebilahendused.ee\/et\/otsustuspuul-pohinevad-masinoppemeetodid\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Avaleht","item":"https:\/\/stacc.veebilahendused.ee\/et\/"},{"@type":"ListItem","position":2,"name":"lahendused","item":"https:\/\/stacc.veebilahendused.ee\/et\/tag\/lahendused\/"},{"@type":"ListItem","position":3,"name":"Otsustuspuul p\u00f5hinevad masin\u00f5ppemeetodid"}]},{"@type":"WebSite","@id":"https:\/\/stacc.veebilahendused.ee\/et\/#website","url":"https:\/\/stacc.veebilahendused.ee\/et\/","name":"STACC","description":"Juhtiv masin\u00f5ppe ja andmeteaduse ettev\u00f5te","publisher":{"@id":"https:\/\/stacc.veebilahendused.ee\/et\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/stacc.veebilahendused.ee\/et\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"et"},{"@type":"Organization","@id":"https:\/\/stacc.veebilahendused.ee\/et\/#organization","name":"STACC","url":"https:\/\/stacc.veebilahendused.ee\/et\/","logo":{"@type":"ImageObject","inLanguage":"et","@id":"https:\/\/stacc.veebilahendused.ee\/et\/#\/schema\/logo\/image\/","url":"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2021\/09\/stacc.jpg","contentUrl":"https:\/\/stacc.veebilahendused.ee\/wp-content\/uploads\/2021\/09\/stacc.jpg","width":176,"height":176,"caption":"STACC"},"image":{"@id":"https:\/\/stacc.veebilahendused.ee\/et\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/staccest","https:\/\/x.com\/staccest1","https:\/\/www.instagram.com\/staccest\/","https:\/\/ee.linkedin.com\/company\/staccest","https:\/\/www.youtube.com\/channel\/UCnn59bCz9vSM3lC0S9j6Ccw"]},{"@type":"Person","@id":"https:\/\/stacc.veebilahendused.ee\/et\/#\/schema\/person\/10197711b76fb6c7f2c5b32014c3dfbb","name":"Merili Auser","image":{"@type":"ImageObject","inLanguage":"et","@id":"https:\/\/stacc.veebilahendused.ee\/et\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/dadc0a881bc7e4aa0559507a4f900136c956049d810c537bad0e30d7127d7e10?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/dadc0a881bc7e4aa0559507a4f900136c956049d810c537bad0e30d7127d7e10?s=96&d=mm&r=g","caption":"Merili Auser"},"sameAs":["https:\/\/stacc.veebilahendused.ee"],"url":"https:\/\/stacc.veebilahendused.ee\/et\/author\/merili\/"}]}},"_links":{"self":[{"href":"https:\/\/stacc.veebilahendused.ee\/et\/wp-json\/wp\/v2\/posts\/12684","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/stacc.veebilahendused.ee\/et\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/stacc.veebilahendused.ee\/et\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/stacc.veebilahendused.ee\/et\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/stacc.veebilahendused.ee\/et\/wp-json\/wp\/v2\/comments?post=12684"}],"version-history":[{"count":0,"href":"https:\/\/stacc.veebilahendused.ee\/et\/wp-json\/wp\/v2\/posts\/12684\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/stacc.veebilahendused.ee\/et\/wp-json\/wp\/v2\/media\/12722"}],"wp:attachment":[{"href":"https:\/\/stacc.veebilahendused.ee\/et\/wp-json\/wp\/v2\/media?parent=12684"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/stacc.veebilahendused.ee\/et\/wp-json\/wp\/v2\/categories?post=12684"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/stacc.veebilahendused.ee\/et\/wp-json\/wp\/v2\/tags?post=12684"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}