Kuphunzira pamakina kumapangidwa ngati "zovuta zochepetsera" za ntchito yotayika motsutsana ndi zitsanzo zoperekedwa (seti yophunzitsira). Ntchitoyi ikuwonetsa kusiyana pakati pa zomwe zimanenedweratu ndi chitsanzo chomwe chikuphunzitsidwa ndi zomwe zimayembekezeredwa pa chitsanzo chilichonse.
Cholinga chachikulu ndikuphunzitsa chitsanzo luso lolosera molondola pazochitika zomwe sizikupezeka mu maphunziro.
A njira imene n'zotheka kusiyanitsa magulu osiyanasiyana a aligorivimu ndi mtundu wa linanena bungwe kuyembekezera dongosolo lina la makina kuphunzira.
Pakati pa magulu akuluakulu timapeza:
Chitsanzo cha magawidwe ndi ntchito yokhala ndi zilembo chimodzi kapena zingapo pazithunzi zozikidwa pazinthu kapena zinthu zomwe zilimo;
Chitsanzo cha kukonzanso ndikuyerekeza kuya kwa mawonekedwe kuchokera pakuyimiriridwa monga mawonekedwe a chithunzi.
M'malo mwake, gawo lazoyang'anira zomwe zafunsidwa ndizopanda malire, ndipo sizimangolekeredwa pazotheka zina;
Linear regression ndi mmitundu yogwiritsidwa ntchito kwambiri kuyerekezera zinthu zenizeni monga:
ndipo amatsatira muyeso wa zosinthika mosalekeza:
Mukuwongolera mzere, ubale pakati pa zosinthika zodziimira mosiyana ndi zina zotengera zotsatiridwa zimatsatiridwa kudzera pamzere womwe nthawi zambiri umaimira ubale wapakati pa zosiyana ziwirizi.
Mzere wokwanira umadziwika kuti regression line ndipo umaimiridwa ndi mzere wa equation monga Y = a * X + b.
Fomulo imakhazikitsidwa pakuphatikiza deta kuti muzigwirizanitsa mawonekedwe awiri kapena angapo ndi wina ndi mnzake. Mukapatsa algorithm mawonekedwe a kulowetsa, kukonzanso kumabweza mawonekedwe enawo.
Tikakhala ndi zophatikizira zingapo pawokha, ndiye kuti timalankhula za mitundu ingapo, kutengera chitsanzo chonga ichi:
y =b0 + b1x1 + b2x2 +… + Bnxn
Kwenikweni kufananaku kumafotokoza ubale pakati pa kusinthika kosadalira kosasintha (y) ndi mitundu iwiri kapena kupitirira palokha (x1, x2, x3…).
Mwachitsanzo, ngati tinkafuna kuyerekezera mpweya wa CO2 wa galimoto (yodalira variable y) poganizira mphamvu ya injini, kuchuluka kwa masilinda ndi kugwiritsa ntchito mafuta. Zotsirizirazi ndizosiyana zodziyimira pawokha x1, x2 ndi x3. The constants bi ndi manambala enieni ndipo amatchedwa ma regression coefficients a chitsanzo.Y ndi mtundu wodalira mosalekeza, mwachitsanzo kukhala chiŵerengero cha b0, b1 x1, b2 x2, ndi zina zotero. y adzakhala nambala yeniyeni.
Kusanthula kangapo ndi njira yomwe imagwiritsidwa ntchito pozindikira zotsatira zomwe mitundu yodziyimira payokha imakhala nayo pakusintha kodalira.
Kumvetsetsa momwe zosinthika zosinthika zimasinthira ngati zosinthika zakumasinthika zimatilola kulosera zotsatira kapena zovuta zakusintha mu zochitika zenizeni.
Kugwiritsa ntchito mzera wambiri wosiyanasiyana ndikotheka kumvetsetsa momwe kuthamanga kwa magazi kumasinthira pamene kuchuluka kwa minyewa ya thupi kumasintha poganizira zinthu monga zaka, kugonana, ndi zina zambiri, poganiza zomwe zingachitike.
Ndi kusinthika kangapo titha kudziwa za momwe mitengo ikukhalira, monga zamtsogolo zamafuta kapena golide.
Pomaliza, kuwongolera mzera wambiri ndikupeza chidwi chochulukirapo pantchito yophunzirira makina ndi luntha lochita kupanga popeza zimaloleza kupeza mitundu yophunzirira ngakhale pamtundu wa mbiri yambiri kuti iwunikidwe.
Logistic regression ndi chida chowerengera chomwe chimafuna kuwonetsa zotsatira za binomial ndi chimodzi kapena zingapo zofotokozera.
Amagwiritsidwa ntchito nthawi zambiri pamavuto a binary, pomwe pali magulu awiri okha, mwachitsanzo Inde kapena ayi, 0 kapena 1, wamwamuna kapena wamkazi etc.
Mwanjira imeneyi ndikothekera kufotokozera deta ndikufotokozera ubale womwe uli pakati pa zosankha zotsalira pamsika ndi zosinthika zingapo kapena zingapo.
Zotsatira zake zimatsimikiziridwa chifukwa chogwiritsa ntchito njira yolumikizirana, yomwe imayerekezera mwayi ndiyeno defiimamaliza kalasi yoyandikira kwambiri (zabwino kapena zoyipa) ku mtengo womwe wapezeka.
Titha kuona kusinthika kwa zinthu monga njira yogawa banja la kuyang'anira ma algorithms ophunzirira.
Pogwiritsa ntchito njira zowerengera, kusinthika kwazinthu kumapangitsa kukhala kotheka kupanga chotsatira chomwe, kwenikweni, chimayimira mwayi woti mtengo womwe waperekedwa ndi wa gulu lomwe laperekedwa.
M'mabvuto a binomial logistic regression, mwayi woti zotulukazo ndi za gulu limodzi udzakhala P, pomwe wa gulu lina udzakhala 1-P (pomwe P ndi nambala pakati pa 0 ndi 1 chifukwa ikuwonetsa kuthekera).
Kusintha kwazinthu za binomial kumagwira bwino nthawi zonse zomwe zosinthika zomwe tikufuna kunena ndizosankha, ndiye kuti, zingangoganiza zinthu ziwiri zokha: mtengo 1 womwe umayimira gulu loyimira, kapena mtengo 0 womwe umayimira gulu loyipa.
Zitsanzo zamavuto omwe amatha kuthana ndi kukonzanso kwa zinthu ndi izi:
Ndi malingaliro oyendetsera zinthu titha kusanthula kolosera, kuyeza mgwirizano pakati pazomwe tikufuna kulosera (zosintha modalira) ndi zosinthika chimodzi kapena zingapo zodziyimira, mwachitsanzo. Kuyerekezera koyenera kumachitika pogwiritsa ntchito ntchito.
Vutoli limasinthidwa kenako kukhala lingaliro lamabizinesi, ndipo kuti zonenerazo zitheke, zotsatirazo zimaperekedwa kwa gulu lomwe zimachokera, kutengera ngati lili pafupi ndi kalasiyo.
Mwachitsanzo, ngati kugwiritsidwa ntchito kwa ntchitoyo kumabwereranso ndi 0,85, ndiye kuti zikutanthauza kuti zoperekazo zidapanga gulu labwino powapatsa gulu 1. Mosiyana ndi izi ngati atapeza phindu monga 0,4 kapena kupitilira apo <0,5 ..
Kusunthika kwa zinthu kumagwiritsa ntchito ntchito kuti tiwunikenso gulu la zomwe akuika.
Ntchito yolumikizidwa, yomwe imatchedwanso sigmoid, ndi njira yokhotakhota yomwe imatha kutenga nambala iliyonse ya mtengo weniweni ndikuiika pamtengo wapatali pakati pa 0 ndi 1, kupatula owonjezera. Ntchitoyi ndi:
Kumene:
Kukongoletsa kwa zinthu kumagwiritsa ntchito equation ngati choyimira, monganso kutsata kwa mzere
Mitengo yolowera (x) imaphatikizidwa mosiyanasiyana pogwiritsa ntchito zolemera kapena zolondola, kulosera mtengo watulutsa (y). Kusiyanitsa kofunikira kuchokera pamakina ogwiritsira ntchito mzere ndikuti mtengo wotumizidwa ndi mtengo wa binary (0 kapena 1) m'malo mwa kuchuluka kwa manambala.
Nachi chitsanzo cha logistic regression equation:
y = e^(b0 + b1 * x) / (1 + e^(b0 + b1 * x))
Kumene:
Chingwe chilichonse chomwe chili mu data yolowera chimakhala ndi cholumikizira (chomwe chimakhala chenicheni) chomwe chimaphunzitsidwa kuchokera ku zomwe ophunzira aphunzitsidwa.
Choyimira chenicheni cha mtundu womwe mungasunge kukumbukira kapena fayilo ndi ma coefficients omwe ali mu equation (beta kapena b).
Mitundu yoyendetsera zochitika ikusonyeza kuthekera kwa kalasi lophweka.
Mwachitsanzo, tiyerekeze kuti tikufanizira kugonana kwa anthu ngati amuna kapena akazi kuchokera kutalika kwawo, kalasi yoyamba ikhoza kukhala yaimuna, ndipo mtundu wa zojambulazo ukhoza kulembedwa ngati mwayi woti wamwamuna apatsidwa kutalika kwa munthu, kapena zina. mwamwayi:
P ( jenda = mwamuna | kutalika)
Zolembedwa mwanjira ina, tikuwonetsa mwayi woti cholowetsa (X) ndi cha kalasi yoyambadefinite (Y = 1), tikhoza kulemba motere:
P(X) = P(Y = 1 | X)
Kuneneratu za kuthekera kuyenera kusinthidwa kukhala mabizinesi (0 kapena 1) kuti athe kulosera zazotheka.
Kusintha kwa zinthu ndi njira yofananira, koma zonenedweratu zimasinthidwa pogwiritsa ntchito ntchito. Zotsatira za izi ndikuti sitingamvetsetsenso kulosera monga kuphatikiza kwa zomangira momwe tingathere ndi mayendedwe a mzere, mwachitsanzo, kupitilira kuchokera kumtunda, chitsanzocho chikhoza kufotokozedwa ngati:
Episodio (X) (e) (b0 + b1 * X) / (1 + e ^ (b0 + b1 * X))
Tsopano titha kusintha mgwirizano motere. Kuti titembenuzire titha kupitilira pochotsa e mbali inayo powonjezera zachilengedwe mbali inayo.
ln (p (X) / 1 - p (X)) = b0 + b1 * X
Mwanjira imeneyi timazindikira kuti kuphatikizira kwadzanja kumanja ndikolingalanso (monga kutsata kwa mzere), ndipo kuyika kumanzere ndi chithunzi cha kuthekera kwa mtundu wotsika.
Zowerengedwa zimawerengedwa ngati chiyerekezo cha kuthekera kwa chochitikachi chomwe chimagawidwa ndi mwayi woti palibe chochitika, i.e. 0,8 / (1-0,8) yemwe zotsatira zake ndi 4. Chifukwa chake titha kulemba:
ln (zovuta) = b0 + b1 * X
Popeza kuthekera kusinthidwa ndi chipika, timatcha chipolopolo chakumanzere kapena zosowa.
Titha kubweretsa exponent kumanja ndikulemba motere:
mwayi = e ^ (b0 + b1 * X)
Zonsezi zimatithandiza kumvetsetsa kuti chitsanzocho chikadali chophatikizira chazolowera, koma kuti kuphatikiza kwa mzerewu kumatanthawuza kuthekera kwa chipika cha pre class.definita.
Ma coefficients (beta kapena mfundo za b) pamalingaliro amitundu yodziwika bwino amawerengeredwa mgawo la maphunziro. Kuti tichite izi, timagwiritsa ntchito kuwerengera kwakukulu.
Kuyerekeza kwa kuthekera kwakukulu ndi njira yophunzirira yomwe imagwiritsidwa ntchito ndi makina angapo ophunzirira makina. Ma coefficients obwera chifukwa cha modeli amalosera mtengo woyandikira kwambiri 1 (monga wachimuna) wa kalasi yoyambiriradefinite ndi mtengo wapafupi kwambiri ndi 0 (monga wamkazi) wa gulu lina. Kuthekera kwakukulu kwa kusinthika kwazinthu ndi njira yopezera ma coefficients (ma Beta kapena ma ob values) omwe amachepetsa zolakwika zomwe zanenedweratu ndi chitsanzocho poyerekeza ndi zomwe zili mu data (mwachitsanzo, 1 ngati deta ili kalasi yoyamba) .
Tidzagwiritsa ntchito algorithm yochepetsera kukhathamiritsa ma coefficients abwino kwambiri pamaphunziro. Izi nthawi zambiri zimagwiritsidwa ntchito pogwiritsa ntchito algorithm yowerengera bwino manambala.
Mawu akuti Smart Lock Market amatanthauza makampani ndi chilengedwe chozungulira kupanga, kugawa ndi kugwiritsa ntchito…
Mu engineering software, mapangidwe apangidwe ndi njira yabwino yothetsera mavuto omwe amapezeka pakupanga mapulogalamu. ndili ngati…
Kuyika zilembo zamafakitale ndi mawu otakata omwe amaphatikiza njira zingapo zomwe zimagwiritsidwa ntchito popanga zilembo zokhazikika pamwamba pa…
Zitsanzo zosavuta zotsatirazi za Excel zidalembedwa pogwiritsa ntchito VBA Yowerengera nthawi: Mphindi 3 Chitsanzo…