Mhoroi mose! Zita rangu ndiSasha, ndiri CTO & Co-Muvambi paLoyaltyLab. Makore maviri apfuura, shamwari dzangu neni, sevadzidzi vose varombo, takaenda manheru kunotenga doro pachitoro chaiva pedyo pedyo nepamba pedu. Takagumbuka zvikuru kuti mutengesi, achiziva kuti taizouya kudoro, haana kupa mutero pamachipisi kana kuti crackers, kunyange zvazvo izvi zvakanga zvine musoro! Hatina kunzwisisa kuti sei mamiriro ezvinhu aya ari kuitika uye takasarudza kutanga kambani yedu. Zvakanaka, sebhonasi, zvipe zvideredzwa Chishanu chega chega pamachipi mamwe chete iwayo.

Uye zvese zvakasvika padanho rekuti ndiri kuratidza zvinhu padivi rehunyanzvi rechigadzirwa NVIDIA GTC. Tinofara kugoverana basa redu nenzanga, naizvozvo ndiri kubudisa mushumo wangu muchimiro chechinyorwa.

Nhanganyaya

Kufanana nevamwe vese pakutanga kwerwendo, takatanga netarisiro yemagadzirirwo ehurongwa hwekukurudzira. Uye iyo inonyanya kufarirwa dhizaini yakave iyi inotevera mhando:

Inoumbwa nezvikamu zviviri:

1. Sampling vamiriri vekurudziro vachishandisa yakapusa uye inokurumidza modhi, kazhinji inodyidzana.
2. Chiyero chevavhoti vane yakanyanya kuomarara uye inononoka yemukati modhi, uchifunga nezve zvese zvinogoneka mu data.

Pano ini ndichashandisa mazwi anotevera:

• mukwikwidzi/mumiriri wekurudziro - mushandisi-chigadzirwa vaviri vanogona kuverengerwa mukurudziro mukugadzira.
• vanovhota kudhirowa / dhizaini / mumiriri yekubvisa nzira - maitiro kana nzira yekubvisa "vanorudziro" kubva kune iripo data.

Danho rekutanga rinowanzo sanganisira kushandiswa kwakasiyana kwakasiyana kwekusefa kwakabatana. Iyo inonyanya kufarirwa - ALS. Zvinoshamisa kuti zvinyorwa zvakawanda pamusoro pekukurudzira masisitimu zvinongoburitsa kuvandudzwa kwakasiyana kumhando dzekubatana padanho rekutanga, asi hapana anotaura zvakanyanya nezve dzimwe nzira dzesampling. Kwatiri, maitiro ekushandisa chete ekubatana modhi uye akasiyana optimizations navo haana kushanda nemhando yataitarisira, saka takachera mukutsvagisa zvakananga pane ichi chikamu. Uye pakupera kwechinyorwa ini ndicharatidza kuti takakwanisa sei kuvandudza ALS, yaive yekutanga yedu.

Ndisati ndaenderera mberi nekutsanangura maitiro edu, zvakakosha kuziva kuti mukurudziro yenguva chaiyo, kana zvichikosha kuti isu tifunge nezve data yakaitika maminitsi makumi matatu apfuura, hapana nzira dzakawanda dzinogona kushanda munguva inodiwa. Asi, kwatiri isu, isu tinofanirwa kuunganidza kurudziro kwete kanopfuura kamwe pazuva, uye kazhinji kamwechete pasvondo, izvo zvinotipa mukana wekushandisa yakaoma mhando uye kuvandudza mhando kakawanda.

Ngatitorei seyekutanga izvo metrics chete ALS inoratidza pabasa rekutora vavhoti. Iwo akakosha metrics atinotarisa ndeaya:

• Precision - chiyero chevamiriri vakasarudzwa nemazvo kubva kune avo vakatorwa.
• Recall chikamu chevavhoti chakaitika kubva kune avo vanga vari mukati menguva yakatarwa.
• F1-score - F-chiyero chakaverengerwa pamapoinzi maviri apfuura.

Isu tichatarisawo ma metrics eiyo yekupedzisira modhi mushure mekudzidzira gradient kuwedzera nekuwedzera zvemukati maficha. Kune zvakare matatu makuru metrics pano:

• precision@5 - avhareji muzana yezvigadzirwa kubva pamusoro 5 maererano nemukana wemutengi wega wega.
• mhinduro-chiyero @ 5 - shanduko yevatengi kubva pakushanya kuchitoro kuenda kukutenga kanenge kamwe chete kupihwa kwemunhu (5 zvigadzirwa mune imwe yekupihwa).
• avg roc-auc pamushandisi - pakati roc-auc kumutengi mumwe nomumwe.

Izvo zvakakosha kuti uzive kuti ese metrics aya anoyerwa pa nguva-yakatevedzana kuyambuka-kusimbisa, ndiko kuti, kudzidziswa kunoitika mumavhiki ekutanga k, uye k + 1 vhiki inotorwa se data rekuedza. Nokudaro, kukwira kwemwaka uye kuderera kwaiva nemigumisiro shoma pakududzirwa kwehutano hwemhando. Kupfuurirazve pamagirafu ese, iyo abscissa axis icharatidza nhamba yevhiki mukuchinjika-kusimbisa, uye iyo yakarongeka axis icharatidza kukosha kweiyo metric yakatsanangurwa. Ese magirafu akavakirwa pane data rekutengesa kubva kune mumwe mutengi kuitira kuti kuenzanisa pakati peumwe neumwe kwakarurama.

Tisati tatanga kutsanangura maitiro edu, tinotanga tatarisa pahwaro hwekutanga, inova muenzaniso wakadzidziswa neALS.
Mametriki ekudzoreredza mumiriri:

Final metrics:

Ini ndinobata zvese kuitiswa kwealgorithms seimwe mhando yebhizinesi hypothesis. Saka, zvakanyanyisa, chero mhando yemubatanidzwa inogona kutorwa sefungidziro yekuti "vanhu vanowanzotenga zvinotengwa nevanhu vakafanana navo." Sezvandambotaura, isu hatina kugumira kune semantics yakadai, uye heano mamwe mafungidziro anoshanda nemazvo pane data mukutengesa kunze kwenyika:

1. Izvo ndakatotenga kare.
2. Zvakafanana nezvandakatenga kare.
3. Nguva yekutenga kwenguva refu yapfuura.
4. Yakakurumbira nechikamu/mhando.
5. Kumwe kutenga kwezvinhu zvakasiyana vhiki nevhiki (Markov cheni).
6. Zvigadzirwa zvakafanana kune vatengi, zvinoenderana nehunhu hwakavakwa nemhando dzakasiyana (Word2Vec, DSSM, nezvimwewo).

Wakambotenga chii?

Iyo yakanyanya kujeka heuristic inoshanda zvakanyanya mukutengesa grocery. Pano tinotora zvinhu zvose zvakatengwa nekadhi rekuvimbika mumazuva ekupedzisira eK (kazhinji mavhiki 1-3), kana K mazuva apfuura. Tichishandisa nzira iyi chete, tinowana anotevera metrics:

Apa zviripachena kuti kana tichitora nguva yakareba, ndipo patinonyanya kuyeuka uye nekushomeka kwatinako uye neimwe nzira. Paavhareji, "mavhiki maviri ekupedzisira" anopa mhedzisiro iri nani kune vatengi.

Zvakafanana nezvandakatenga kare

Hazvishamisi kuti kune grocery rekutengesa "zvandakatenga kare" zvinoshanda zvakanaka, asi kutora vavhoti chete kubva kune izvo mushandisi atotenga hazvina kutonhorera zvakanyanya, nekuti hazvigoneke kushamisa mutengi nechimwe chigadzirwa chitsva. Naizvozvo, isu tinokurudzira kuvandudza zvishoma iyi heuristic tichishandisa iwo mamodheru ekubatana. Kubva kumavekita atakagamuchira panguva yekudzidziswa kweALS, tinogona kuwana zvigadzirwa zvakafanana kune zvakatotengwa nemushandisi. Pfungwa iyi yakafanana ne "mavhidhiyo akafanana" mumasevhisi ekuona zvemukati zvevhidhiyo, asi sezvo isu tisingazive izvo mushandisi ari kudya / kutenga pane imwe nguva, tinogona kungotsvaga akafanana neaakatotenga, kunyanya. sezvo isu Tinotoziva kuti inoshanda sei. Tichishandisa nzira iyi pakuchinjana kwevashandisi mukati memavhiki maviri apfuura, tinowana anotevera metrics:

zviri k - huwandu hwezvigadzirwa zvakafanana izvo zvinodzoserwa kune chimwe nechimwe chakatengwa nemutengi mukati memazuva gumi nemana apfuura.
Iyi nzira yakashanda zvakanyanya kumutengi wedu, uyo zvaive zvakakosha kuti asakurudzira chero chinhu chaive chatova munhoroondo yekutenga yemushandisi.

Kunonoka kutenga nguva

Sezvatakatoona, nekuda kwehuwandu hwakawanda hwekutenga zvinhu, nzira yekutanga inoshanda zvakanaka kune zvatinoda chaizvo. Asi zvakadini nezvinhu zvakaita sehupfu yekuwachisa/shampoo/etc. Ndiko kuti, nezvigadzirwa zvisingaite kudikanwa vhiki imwe neimwe kana maviri uye izvo zvakapfuura nzira hazvigone kubvisa. Izvi zvinotungamira kune inotevera zano - inokurudzirwa kuverenga nguva yekutenga kwechigadzirwa chimwe nechimwe paavhareji kune vatengi vakatenga chigadzirwa zvakanyanya. k kamwe. Wobva wabvisa izvo mutengi angangove atopererwa. Iyo yakaverengerwa nguva yezvinhu inogona kutariswa nemeso ako kukwana:

Uye tobva tatarisa kana kupera kwenguva yechigadzirwa kuchiwira mukati menguva iyo kurudziro inenge iri mukugadzirwa uye kuenzanisa zvinoitika. Iyo nzira inogona kuenzanisirwa seizvi:

Pano isu tine 2 nyaya huru dzinogona kutariswa:

1. Izvo zvinodikanwa here kuyedza zvigadzirwa kubva kune vatengi vakatenga chigadzirwa zvisingasviki K nguva.
2. Izvo zvinodikanwa here kuenzanisa chigadzirwa kana kupera kwenguva yacho kuchiwira pamberi pekutanga kwechinangwa chepakati.

Girafu rinotevera rinoratidza mhedzisiro nzira iyi inowana nema hyperparameter akasiyana:

ft - Tora chete vatengi vakatenga chigadzirwa kanokwana K (pano K = 5) nguva
tm - Tora chete vavhoti vanowira mukati menguva yakatarwa

Hazvishamisi kuti anokwanisa (0, 0) chikuru yeuka uye mudukusa zvakarurama, sezvo pasi pemamiriro ezvinhu aya vavhoti vakawanda vanotorwa. Nekudaro, mibairo yakanakisa inowanikwa kana isu tikasatevedzera zvigadzirwa zvevatengi vakatenga chimwe chigadzirwa zvishoma pane k nguva uye kubvisa, kusanganisira zvinhu, kupera kwenguva iyo inowira pamberi penguva yakatarwa.

Yakakurumbira nechikamu

Imwe pfungwa iri pachena ndeyekuyedza zvigadzirwa zvakakurumbira mumapoka akasiyana kana mhando. Pano tinoverengera kune mutengi wega wega pamusoro-k "favorite" zvikamu / mhando uye bvisa "yakakurumbira" kubva muchikamu ichi / mhando. Muchiitiko chedu, isu tichasarudza "inoda" uye "yakakurumbira" nehuwandu hwekutenga kwechigadzirwa. Imwezve mukana weiyi nzira ndeyekushandiswa kwayo mune inotonhora yekutanga kesi. Izvi zvinoreva, kune vatengi vakatenga zvishoma zvishoma, kana vasina kuenda kuchitoro kwenguva yakareba, kana kuti vachangobva kupa kadhi rekuvimbika. Kwavari, zviri nyore uye zvirinani kutenga zvinhu zvinofarirwa nevatengi uye zvine nhoroondo. Izvo zvinoguma metrics ndezvi:

Pano nhamba iri shure kwezwi rokuti "category" rinoreva nhanho yekugara kwechikwata.

Pakazere, hazvishamise zvakare kuti mapoka akamanikana anowana mhedzisiro iri nani, sezvo ivo vachitora zvakanyanya "zvavanoda" zvigadzirwa zvevatengi.

Alternate kutenga zvinhu zvakasiyana vhiki nevhiki

Nzira inonakidza yandisati ndaona muzvinyorwa pamusoro pekukurudzira masisitimu iri nyore uye panguva imwechete inoshanda yenhamba nzira yeMarkov maketani. Pano tinotora mavhiki maviri akasiyana, zvino kune mutengi wega wega tinovaka mapeya ezvigadzirwa [akatengwa muvhiki i]-[akatengwa muvhiki j], apo j > i, uye kubva pano tinoverengera kune chimwe nechimwe chigadzirwa mukana wekuchinja kune chimwe chigadzirwa svondo rinouya. Ndiko kuti, peya imwe neimwe yezvinhu producti-productj Tinoverenga nhamba yavo muzvikamu zvakawanikwa uye tinoparadzanisa nenhamba yezviviri, kupi zvigadzirwa rakanga riri muvhiki rokutanga. Kuti titore vavhoti, tinotora risiti yekupedzisira yemutengi uye kubvisa pamusoro-k zvingangoitika zvigadzirwa zvinotevera kubva kushanduko matrix yatakagamuchira. Maitiro ekugadzira shanduko matrix anoita seizvi:

Kubva pamienzaniso chaiyo mushanduko yemukana matrix tinoona zvinotevera zvinonakidza zviitiko:

Pano iwe unogona kuona zvinofadza zvinotsamira zvinoratidzwa mukuzvibata kwevatengi: semuenzaniso, vanoda michero yecitrus kana rudzi rwemukaka kubva kwavanogona kuchinjira kune imwe. Izvo zvakare hazvishamise kuti zvigadzirwa zvine yakakwira frequency yekudzokorora kutenga, senge butter, zvakare inoguma pano.

Iwo metrics mune nzira neMarkov cheni ndeaya anotevera:

k - nhamba yezvigadzirwa zvinodzoserwa kune chimwe nechimwe chakatengwa kubva kune yekupedzisira kutengeserana kwemutengi.
Sezvatinoona, mhedzisiro yakanakisa inoratidzwa nekumisikidzwa ne k = 4. Iyo spike muvhiki 4 inogona kutsanangurwa nemwaka maitiro kutenderedza zororo. 

Zvigadzirwa zvakafanana kune vatengi, zvinoenderana nehunhu hwakavakwa nemhando dzakasiyana

Iye zvino tauya kune yakanyanya kuoma uye inonakidza chikamu - kutsvaga vavakidzani vepedyo zvichienderana nemavekita evatengi uye zvigadzirwa zvakavakwa zvinoenderana nemhando dzakasiyana. Mubasa redu tinoshandisa 3 mhando dzakadai:

• ALS
• Word2Vec (Item2Vec yemabasa akadaro)
• DSSM

Isu takatobata neALS, unogona kuverenga nezve madzidzire ayo pano. Panyaya yeShoko2Vec, isu tinoshandisa iyo inozivikanwa kwazvo yekumisikidza modhi kubva gensim. Nekufananidza nemagwaro, tinotsanangura chipo serisiti yekutenga. Nokudaro, kana uchigadzira vector yechigadzirwa, muenzaniso unodzidza kufanotaura nezvechigadzirwa chiri murisiti "mamiriro" ayo (zvakasara zvigadzirwa murisiti). Mune ecommerce data, zviri nani kushandisa chikamu chemutengi pane risiti; vakomana vanobva ozone. Uye DSSM inonyanya kunakidza kutsanangura. Pakutanga, yakanyorwa nevakomana veMicrosoft semuenzaniso wekutsvaga, Unogona kuverenga bepa rekutanga rekutsvagisa pano. Kuvakwa kweiyo modhi kunoratidzika seizvi:

zviri Q - mubvunzo, mushandisi wekutsvaga mubvunzo, D[i] - gwaro, peji rewebhu. Iko kupinza kune modhi ndiyo hunhu hwekukumbira uye mapeji, zvichiteerana. Mushure mechikamu chega chega chekuisa kune nhamba yezvikamu zvakazara zvakabatana (multilayer perceptron). Tevere, iyo modhi inodzidza kuderedza iyo cosine pakati pemavekita anowanikwa muzvikamu zvekupedzisira zvemuenzaniso.
Kurudziro mabasa anoshandisa chaizvo zvivakwa zvakafanana, chete pachinzvimbo chekukumbira pane mushandisi, uye panzvimbo yemapeji pane zvigadzirwa. Uye kwatiri, chivakwa ichi chinoshandurwa kuita zvinotevera:

Iye zvino, kutarisa mhedzisiro, inoramba ichivhara iyo yekupedzisira poindi - kana mune iyo ALS neDSSM isu takanyatso tsanangura mushandisi mavheji, saka mune iyo Word2Vec isu tine zvigadzirwa vectors chete. Pano, kuvaka mushandisi vector, isu takatsanangura nzira nhatu huru:

1. Ingo wedzera mavheji, ipapo kune cosine chinhambwe chinozoitika kuti isu takangoenzanisa zvigadzirwa munhoroondo yekutenga.
2. Vector summation ine imwe nguva kurema.
3. Kuyera zvinhu neTF-IDF coefficient.

Panyaya yekuremerwa kwemutsetse wemutengi vector, tinoenderera kubva kufungidziro yekuti chigadzirwa chakatengwa nemushandisi nezuro chine simba guru pamaitiro ake kupfuura chigadzirwa chaakatenga mwedzi mitanhatu yapfuura. Saka isu tinofunga svondo rapfuura remutengi ane kusawirirana kwe1, uye chii chakazoitika nekusawirirana kwe½, ⅓, zvichingodaro:

Kune TF-IDF coefficients, tinoita zvakangofanana neTF-IDF yezvinyorwa, isu chete isu tinoona mutengi segwaro, uye cheki sechipo, zvichiteerana, izwi chigadzirwa. Nenzira iyi, vector yemushandisi inochinja zvakanyanya kuenda kuzvinhu zvisingawanzo, nepo zvinhu zvinowanzoitika uye zvinojairika kune mutengi hazvichinje zvakanyanya. Iyo nzira inogona kuenzanisirwa seizvi:

Zvino ngatitarisei mametrics. Izvi ndizvo zvakaita mhedzisiro yeALS:

Metrics eItem2Vec ane akasiyana akasiyana ekuvaka mutengi vector:

Muchiitiko ichi, iyo modhi yakafanana inoshandiswa seyakatangira yedu. Musiyano chete ndewekuti k watichashandisa. Kuti ushandise maitiro ekubatana chete, iwe unofanirwa kutora anenge makumi mashanu-50 zvigadzirwa zvepedyo kune mumwe nemumwe mutengi.

Uye metrics zvinoenderana neDSSM:

Nzira yekubatanidza sei nzira dzose?

Kutonhorera, iwe unoti, asi chii chekuita neseti yakakura kudaro yematurusi ekubvisa maturusi? Nzira yekusarudza iyo yakakwana yekumisikidzwa kune yako data? Pano tine matambudziko akawanda:

1. Izvo zvinodikanwa neimwe nzira kudzikamisa nzvimbo yekutsvaga ye hyperparameters munzira yega yega. Iko, hongu, discrete kwese kwese, asi huwandu hwemapoinzi hunokwanisika hwakakura kwazvo.
2. Uchishandisa diki diki sampuli yemaitiro chaiwo ane chaiwo hyperparameter, ungasarudza sei yakanakisa gadziriso yemetric yako?

Isu hatisati tawana mhinduro chaiyo kumubvunzo wekutanga, saka tinoenderera mberi kubva kune inotevera: kune yega yega nzira, hyperparameter yekutsvaga space limiter inonyorwa, zvichienderana nehumwe huwandu hwe data yatinayo. Saka, tichiziva avhareji yenguva pakati pekutenga kubva kuvanhu, tinogona kufungidzira kuti ndeipi nguva yekushandisa iyo "yakatotengwa" uye "nguva yekutenga kwenguva refu" nzira.

Uye mushure mekunge tapfuura neimwe nhamba yakakwana yekusiyana kwenzira dzakasiyana, tinocherekedza zvinotevera: kuita kwega kwega kunobvisa imwe nhamba yevanokwikwidza uye kune imwe kukosha kweiyo kiyi metric kwatiri (rangarira). Tinoda kuwana huwandu hweimwe nhamba yevavhoti, zvichienderana nesimba redu rekombuta rinotenderwa, ine metric yepamusoro inogoneka. Pano dambudziko rinowira mudambudziko rebhegi.

Pano nhamba yevanoda ndiyo huremu hwe ingot, uye nzira yekuyeuka ndiyo kukosha kwayo. Nekudaro, pane 2 mamwe mapoinzi anofanirwa kuverengerwa pakuita iyo algorithm:

• Nzira dzingangove dzakapindirana muvamiriri vavanotora.
• Mune zvimwe zviitiko, zvichave zvakarurama kutora nzira imwe kaviri ine maparameter akasiyana, uye mumiriri anobuda kubva kune wekutanga haazove chikamu chechipiri.

Semuenzaniso, kana tikatora kushandiswa kwe "zvandatotenga" nzira ine nguva dzakasiyana dzekutora, ipapo seti yavo yevanoda ichaiswa mukati memumwe nemumwe. Panguva imwecheteyo, zvimiro zvakasiyana mu "periodic kutenga" pakubuda hazvipi mharadzano yakakwana. Naizvozvo, isu tinopatsanura nzira dzesampling nemaparamendi akasiyana kuita mabhuroko zvekuti kubva pabhuroka rega rega tinoda kutora nzira imwechete yekubvisa ine chaiwo hyperparameter. Kuti uite izvi, iwe unofanirwa kuwana hungwaru hudiki mukuita dambudziko reknapsack, asi iyo asymptotics uye mhedzisiro haizochinji.

Uku kusanganisa kwakangwara kunotitendera kuti tiwane anotevera metrics mukuenzanisa nemamodheru anodyidzana:

Muma metrics ekupedzisira tinoona mufananidzo unotevera:

Nekudaro, pano iwe unogona kuona kuti pane imwechete yakavhurika poindi yekurudziro inobatsira kune bhizinesi. Iye zvino tangodzidza maitiro ekuita basa guru rekufanotaura izvo mushandisi achatenga, semuenzaniso, svondo rinouya. Asi kungopa mutero pane chimwe chinhu chaanenge atotenga hakuna kunaka. Asi zvinotonhorera kukwidziridza tarisiro, semuenzaniso, yeanotevera metrics:

1. Margin/turnover zvichibva pane zvaunokurudzira.
2. Avhareji cheki yevatengi.
3. Kuwanda kwekushanya.

Saka isu tinowanza mikana yakawanikwa nemacoefficients akasiyana uye toamisazve kuitira kuti zvigadzirwa zvinokanganisa metrics pamusoro zvisvike pamusoro. Iko hakuna yakagadzirira-yakagadziriswa mhinduro yekuti ndeipi nzira iri nani kushandisa. Isu tinotoedza neakadaro coefficients zvakananga mukugadzira. Asi heano maitiro anonakidza ayo anowanzo kutipa mhedzisiro yakanaka:

1. Wedzera nemutengo / muganho wechigadzirwa.
2. Wanza neavhareji risiti umo chigadzirwa chinoonekwa. Saka zvinhu zvichauya, izvo zvinowanzotora chimwe chinhu.
3. Wedzera neavhareji frequency yekushanyirwa nevatengi vechigadzirwa ichi, zvichibva pane fungidziro yekuti chigadzirwa ichi chinotsamwisa vanhu kuti vadzoke kazhinji.

Mushure mekuita zviyedzo nemacoefficients, takawana anotevera metric mukugadzira:

zviri zvakazara zvigadzirwa kutendeuka - chikamu chezvakatengwa zvigadzirwa kubva kune zvese zvigadzirwa mune kurudziro yatakagadzira.

Muverengi anoteerera anozoona mutsauko wakakura pakati pemametrics ekunze uye online. Maitiro aya anotsanangurwa nenyaya yekuti haasi ese mafirita ane simba ezvigadzirwa anogona kukurudzirwa anogona kuverengerwa pakudzidzisa modhi. Kwatiri, inyaya yakajairwa kana hafu yevavhoterwa vakadzoserwa inogona kusefa; izvi zvakajairwa muindasitiri yedu.

Panyaya yemari, iyo inotevera nyaya inowanikwa, zviri pachena kuti mushure mekutangwa kwekurudziro, mari yeboka rebvunzo iri kukura zvakanyanya, ikozvino avhareji yekuwedzera kwemari nemazano edu ndeye 3-4%:

Mukupedzisa, ini ndinoda kutaura kuti kana iwe uchida kurudziro isiri-chaiyo-chaiyo, saka kuwedzera kukuru kwemhando kunogona kuwanikwa mukuyedza nekubvisa vavhoti kurudziro. Nguva yakawanda yenguva yechizvarwa chavo inoita kuti zvikwanise kusanganisa nzira dzakawanda dzakanaka, iyo yose ichapa migumisiro yakanaka yebhizimisi.

Ndichafara kutaura mumhinduro nemunhu wese anowana zvinyorwa zvinonakidza. Unogona kundibvunza mibvunzo iwe pachako pa teregiramu. Ini zvakare ndinogovera pfungwa dzangu nezve AI / kutanga mune yangu telegraph channel - Mauya :)

Source: www.habr.com