Yandex iri kuvhura chirongwa chekugara muchina kudzidza kune vane ruzivo vanogadzira backend. Kana iwe wakanyora zvakawanda muC ++/Python uye uchida kushandisa ruzivo urwu kuML, saka tichakudzidzisa maitiro ekuita tsvakiridzo inoshanda uye nekupa varairidzi vane ruzivo. Iwe uchashanda pane akakosha Yandex masevhisi uye uwane hunyanzvi munzvimbo dzakadai semitsetse modhi uye gradient kuwedzera, kurudziro masisitimu, neural network yekuongorora mifananidzo, zvinyorwa uye ruzha. Iwe zvakare unozodzidza maitiro ekuongorora nemazvo mamodheru ako uchishandisa metrics offline uye online.
Nguva yechirongwa igore rimwe chete, panguva iyo vatori vechikamu vachashanda mudhipatimendi rehungwaru uye tsvakiridzo yeYandex, pamwe nekupinda hurukuro nemasemina. Kutora chikamu kunobhadharwa uye kunosanganisira basa renguva yakazara: maawa makumi mana pasvondo, kutanga Chikunguru 40 wegore rino. uye ichagara kusvika May 1.
Uye zvino mune zvimwe zvakadzama - nezve rudzi rwevateereri vatiri kumirira, kuti basa rei richange riri uye, kazhinji, kuti nyanzvi yekumashure inogona sei kushandura kuita basa muML.
Kutungamirira
Makambani mazhinji ane Residency Zvirongwa, kusanganisira, semuenzaniso, Google uye Facebook. Ivo vanonyanya kunangana nevadiki uye vepakati-level nyanzvi vari kuyedza kutora nhanho kuenda kuML research. Chirongwa chedu ndechevateereri vakasiyana. Tinokoka vanogadzira backend vakatowana ruzivo rwakakwana uye vanoziva chokwadi kuti muhunyanzvi hwavo vanofanirwa kuchinjika vakananga kuML, kuti vawane hunyanzvi hunoshanda - uye kwete hunyanzvi hwesainzi - mukugadzirisa matambudziko ekudzidza muchina weindasitiri. Izvi hazvirevi kuti hatitsigire vatsvakurudzi vechidiki. Takavagadzirira chirongwa chakasiyana - yakatumidzwa zita reIlya Segalovich, iyo inokubvumirawo kushanda muYandex.
Mugari achashanda kupi?
MuDhipatimendi reMachina Intelligence uye Tsvagiridzo, isu pachedu tinogadzira mazano eprojekiti. Mhedziso huru yekurudziro ndeyezvinyorwa zvesainzi, zvinyorwa, uye mafambiro munharaunda yekutsvagisa. Ini nevandinoshanda navo tinoongorora zvatinoverenga, tichitarisa kuti tingavandudza sei kana kuwedzera nzira dzakataurwa nemasayendisiti. Panguva imwecheteyo, mumwe nomumwe wedu anofunga nezvenzvimbo yake yezivo uye zvaanofarira, anoronga basa racho zvichienderana nenzvimbo dzaanoona dzakakosha. Pfungwa yepurojekiti inowanzozvarwa pamharadzano yemhedzisiro yekutsvagisa kwekunze uye kugona kwake.
Iyi sisitimu yakanaka nekuti inonyanya kugadzirisa matambudziko ehunyanzvi eYandex masevhisi kunyange asati asimuka. Kana sevhisi yakatarisana nedambudziko, vamiriri vayo vanouya kwatiri, vangangotora matekinoroji atakatogadzirira, ayo anosara anofanirwa kushandiswa nemazvo muchigadzirwa. Kana chimwe chinhu chisina kugadzirira, isu tichakurumidza kuyeuka kwatinga "tanga kuchera" uye mune zvipi zvinyorwa zvekutsvaga mhinduro. Sezvatinoziva, nzira yesainzi ndeyekumira pamapfudzi ehofori.
Zvekuita
PaYandex - uye kunyanya muhutungamiriri hwedu - nzvimbo dzose dzakakodzera dzeML dziri kuvandudzwa. Chinangwa chedu ndechekuvandudza mhando yezvakasiyana-siyana zvezvigadzirwa, uye izvi zvinoshanda sechikurudzira chekuyedza zvese zvitsva. Pamusoro pezvo, masevhisi matsva anoonekwa nguva nenguva. Saka chirongwa chemharidzo chine ese makiyi (akanyatsopupurirwa) nzvimbo dzemuchina kudzidza mukusimudzira maindasitiri. Pakuronga chikamu changu chekosi, ndakashandisa ruzivo rwangu rwekudzidzisa paChikoro cheData Analysis, pamwe chete nemidziyo nebasa revamwe vadzidzisi veSHAD. Ndinoziva kuti vandaishanda navo vakaita zvimwe chetezvo.
Mumwedzi yekutanga, kudzidziswa zvinoenderana nechirongwa chekosi kuchaverengera ingangoita 30% yenguva yako yekushanda, zvino ingangoita gumi%. Nekudaro, zvakakosha kuti unzwisise kuti kushanda nemamodeli eML pachawo kucharamba kuchitora kanenge kana zvishoma pane ese ane chekuita maitiro. Izvi zvinosanganisira kugadzirira backend, kugamuchira data, kunyora pombi ye preprocessing, optimizing kodhi, kuchinjika kune chaiyo hardware, etc. An ML injiniya, kana uchida, yakazara-stack developer (chete nekunyanya kukoshesa pakudzidza muchina) , anokwanisa kugadzirisa dambudziko kubva pakutanga kusvika pakupedzisira. Kunyangwe neyakagadzirirwa-yakagadzirwa modhi, iwe ungangoda kuita akati wandei zviito: kuenzanisa maitirwo ayo pamichina yakati wandei, gadzirira kuita muchimiro chemubato, raibhurari, kana zvikamu zvesevhisi pachayo.
Sarudzo yevadzidzi
Kana iwe wanga uri pasi pekufungidzira kuti zviri nani kuve ML injiniya nekutanga kushanda semugadziri webackend, ichi hachisi chokwadi. Kunyoresa muSHAD imwechete pasina ruzivo chairwo mukugadzira masevhisi, kudzidza uye kuve mukudiwa zvakanyanya pamusika isarudzo yakanaka. Nyanzvi dzakawanda dzeYandex dzakaguma munzvimbo dzavo dzazvino nenzira iyi. Kana chero kambani yakagadzirira kukupa basa mumunda weML nekukurumidza mushure mekupedza kudzidza, iwe unofanirwa kugamuchirawo kupihwa. Edza kupinda muchikwata chakanaka nemudzidzisi ane ruzivo uye gadzirira kudzidza zvakawanda.
Chii chinowanzokutadzisa kuita ML?
Kana backender achishuvira kuva ML injiniya, anogona kusarudza kubva munzvimbo mbiri dzebudiriro - asingatarise chirongwa chekugara.
Chekutanga, dzidza sechikamu cheimwe kosi yedzidzo. Coursera ichakuswededza padyo nekunzwisisa hunyanzvi hunyanzvi, asi kuti unyure muhunyanzvi kusvika padanho rakakwana, unofanirwa kushandisa nguva yakawanda kwairi. Semuenzaniso, pedza kudzidza kubva kuSHAD. Nekufamba kwemakore, ShaD yaive nenhamba yakasiyana yezvidzidzo zvakanangana nekudzidza muchina - paavhareji, ingangoita sere. Chimwe nechimwe chazvo chakakosha uye chinobatsira, kusanganisira mumaonero evakapedza kudzidza.
Chechipiri, iwe unogona kutora chikamu mumapurojekiti ekurwisa kwaunoda kuita imwe kana imwe ML algorithm. Nekudaro, kune mashoma mapurojekiti akadaro pamusika wekuvandudza IT: kudzidza kwemichina hakushandiswe mumabasa mazhinji. Kunyangwe mumabhanga ari kushingaira kutsvaga mikana ine chekuita neML, vashoma chete ndivo vari kuita ongororo yedata. Kana iwe usina kukwanisa kujoina chimwe chezvikwata izvi, yako chete sarudzo ndeye kutanga yako purojekiti (apo, zvakanyanya, iwe uchaisa yako wega nguva, uye izvi zvine chekuita nemabasa ekugadzira kurwisa), kana kutanga kukwikwidza. Kaggle.
Chokwadi, batana nedzimwe nhengo dzenharaunda uye uzviedze iwe mumakwikwi - kunyanya kana iwe ukatsigira hunyanzvi hwako nekudzidziswa uye makosi ataurwa paCoursera. Makwikwi ega ega ane nguva yekupedzisira - ichashanda sekukurudzira iwe uye kukugadzirira iwe kune yakafanana system mumakambani eIT. Iyi inzira yakanaka - iyo, zvisinei, zvakare yakarambana zvishoma kubva kune chaiwo maitiro. PaKaggle unopihwa pre-yakagadziriswa, kunyangwe isiri nguva dzose yakakwana, data; usape kufunga nezve mupiro kune chigadzirwa; uye zvakanyanya kukosha, hazvidi mhinduro dzakakodzera kugadzirwa. Algorithms yako ingangoshanda uye ichave yakarurama kwazvo, asi mamodheru nekodhi yako ichafanana neFrankenstein yakasonanidzwa pamwechete kubva kuzvikamu zvakasiyana - mupurojekiti yekugadzira, chimiro chese chichashanda zvishoma nezvishoma, zvichanetsa kugadzirisa nekuwedzera (semuenzaniso, mutauro uye izwi algorithms zvinogara zvichinyorwa patsva sezvo mutauro uchikura). Makambani anofarira kuti basa rakanyorwa rinogona kuitwa kwete chete newe pachako (zviri pachena kuti iwe, semunyori wemhinduro, unogona kuita izvi), asiwo nemumwe wako. Musiyano pakati pemitambo uye maindasitiri mapurogiramu anokurukurwa , uye Kaggle anodzidzisa chaizvo "vatambi" - kunyangwe achizviita nemazvo, zvichivabvumira kuwana ruzivo.
Ndakatsanangura mitsara miviri inogoneka yebudiriro - kudzidziswa kuburikidza nezvirongwa zvedzidzo uye kudzidziswa "mukurwa", semuenzaniso paKaggle. Chirongwa chekugara musanganiswa weidzi nzira mbiri. Hurukuro nemasemina padanho reSHAD, pamwe nemapurojekiti anorwa zvechokwadi, akakumirira.
Source: www.habr.com