Hupenyu hwangu hwese hwevanhu vakuru, ndanga ndiri chinwiwa chesimba (kwete, ikozvino hatisi kutaura nezvechinwiwa chine dubious properties).
Handisati ndambonyanya kufarira nyika yeruzivo rwemichina, uye handikwanise kana kuwanza matrices pabepa. Uye ini handina kumbobvira ndada izvi, kuti iwe unzwisise zvishoma nezve chaiyo yebasa rangu, ndinogona kugovera nyaya inoshamisa. Ndakambokumbira vandaishanda navo kuti vaite basa muExcel spreadsheet, hafu yezuva rekushanda rakanga rapfuura, ndakaenda kwavari, uye vakanga vakagara uye vachipfupisa data pane calculator, hongu, pane imwe calculator yakasviba ine mabhatani. Zvakanaka, ndedzipi mhando dzeneural network dzatinogona kutaura nezvazvo mushure meizvi? Asi, sezvavanotaura, "kwakanaka kwatisiri," shamwari dzangu dzakaridza nzeve dzangu nezve chokwadi chakawedzerwa, nezve neural network, nezve mitauro yekuronga (kunyanya nezvePython).
Mumashoko zvaiita sezviri nyore, uye ndakasarudza kuti sei ndisingagone unyanzvi uhwu hwemashiripiti kuti ndihushandise mubasa rangu.
Muchinyorwa chino, ini ndichasvetuka kuedza kwangu kugona izvo zvekutanga zvePython uye kugovana newe zvandinoona zveyemahara TensorFlow kosi kubva kuUdacity.
Nhanganyaya
Kutanga, zvakakosha kuziva kuti mushure memakore gumi nerimwe muindasitiri yesimba, kana iwe uchiziva uye uchigona kuita zvese uye kunyange zvishoma (maererano nemabasa ako), kudzidza zvakanyanya zvinhu zvitsva - kune rumwe rutivi, kunokonzera kushingaira kukuru, asi kune imwe - inoshanduka kuva kurwadziwa kwemuviri "giya mumusoro mangu."
Ini handisati ndanyatsonzwisisa ese ekutanga mazano ehurongwa uye kudzidza muchina, saka haufanirwe kunditonga nehasha. Ndinovimba chinyorwa changu chichave chinonakidza uye chinobatsira kune vanhu vakaita seni vari kure nekuvandudzwa kwesoftware.
Ndisati ndaenda kuongororo yekosi, ini ndichati kuti uidzidze iwe uchada kashoma ruzivo rwePython. Unogona kuverenga akati wandei emabhuku emadummies (ini ndatangawo kutora kosi paStepic, asi handisati ndazvigona zvizere).
Iyo TensorFlow kosi pachayo haizove neyakaomesesa zvivakwa, asi zvichave zvakakodzera kuti unzwisise kuti sei maraibhurari achiunzwa kunze, kuti basa rinotsanangurwa sei, uye nei chimwe chinhu chichitsiviwa mairi.
Sei TensorFlow uye Udacity?
Chinangwa chikuru chekudzidziswa kwangu chaive chishuwo chekuziva mafoto emagetsi ekuisa zvinhu uchishandisa neural network.
Ndakasarudza TensorFlow nekuti ndakanzwa nezvazvo kubva kushamwari dzangu. Uye sezvandinozvinzwisisa, kosi iyi yakakurumbira.
Ndakaedza kutanga kudzidza kubva kune mukuru .
Uye ipapo ndakasangana nematambudziko maviri.
- Kune zvakawanda zvekushandisa zvekudzidzisa, uye zvinouya mumhando dzakasiyana. Zvakanga zvakandiomera kuti ndigadzire mufananidzo wakawanda kana wakaderera wekugadzirisa dambudziko rekuzivikanwa kwemufananidzo.
- Zvizhinji zvezvinyorwa zvandinoda hazvina kushandurirwa muchiRussian. Zvakangoitika kuti ndakadzidza chiGerman ndichiri mwana uye zvino, kufanana nevana vakawanda veSoviet, handizivi chiGerman kana kuti Chirungu. Ehe, muhupenyu hwangu hwese hwevanhu vakuru, ndakaedza kugona Chirungu, asi zvakazoitika senge pamufananidzo.
Mushure mekuchera kutenderedza pawebhusaiti yepamutemo, ndakawana kurudziro yekupfuura nayo .
Sezvandinozvinzwisisa, kosi yeCoursera yakabhadharwa, uye kosi yacho kwaibvira kupfuudza “mahara, ndiko kuti, pasina.”
Kosi yemukati
Kosi yacho ine zvidzidzo 9.
Chikamu chekutanga chaicho nhanganyaya, kwavanozokuudza kuti nei ichidikanwa musimboti.
Chidzidzo #2 chakava chandinoda. Yakanga iri nyore kunzwisisa uye yakaratidzawo zvinoshamisa zvesayenzi. Muchidimbu, muchidzidzo chino, mukuwedzera kune ruzivo rwekutanga nezve neural network, vagadziri vanoratidza mashandisiro e-single-layer neural network kugadzirisa dambudziko rekushandura tembiricha kubva kuFahrenheit kuenda kuCelsius.
Chokwadi uyu muenzaniso wakajeka. Ndichiri kugara pano ndichifunga nezvekuuya nekugadzirisa dambudziko rakafanana, asi kune vemagetsi chete.
Nehurombo, ndakamira mberi, nekuti kudzidza zvinhu zvisinganzwisisike mumutauro wausina kujaira kwakaoma. Chakandiponesa ndicho chandakawana pana Habré .
Dudziro iyi yakaitwa nemhando yepamusoro, zvinyorwa zveColab zvakaturikirwawo, saka ndakabva ndatarisa zvese zvekutanga neshanduro yacho.
Chidzidzo Nhamba 3, chokwadi, kugadziridzwa kwezvinhu kubva kune yepamutemo TensorFlow tutori. Muchidzidzo ichi, tinoshandisa multilayer neural network kudzidza kurongedza mifananidzo yembatya (Fashion MNIST dataset).
Zvidzidzo Nhamba 4 kusvika Nhamba 7 zvakare inogadziriswa yedzidziso Asi nekuda kwekuti ivo vakarongeka nemazvo, hapana chikonzero chekunzwisisa kutevedzana kwekudzidza iwe pachako. Muzvidzidzo izvi tichaudzwa muchidimbu nezve ultra-chaiyo neural network, maitiro ekuwedzera iko kurongeka kwekudzidziswa uye kuchengetedza modhi. Panguva imwecheteyo, isu tichagadzirisa panguva imwe chete dambudziko rekurongedza katsi nembwa mumufananidzo.
Chidzidzo Nhamba 8 ikosi yakaparadzana zvachose, kune mudzidzisi akasiyana, uye kosi yacho pachayo yakakura. Chidzidzo chacho chiri maererano nenguva dzakatevedzana. Sezvo ini ndisati ndazvifarira, ndakaiongorora diagonally.
Izvi zvinopera nechidzidzo #9, chiri kukokwa kutora kosi yemahara paTensorFlow lite.
Zvawaida nezvausingade
Ndichatanga nezvakanaka:
- Kosi yacho ndeyemahara
- Dzidzo iri paTensorFlow 2. Mamwe mabhuku andakaona uye mamwe makosi epaInternet aive paTensorFlow 1. Handizivi kana pane musiyano mukuru, asi zvakanaka kudzidza shanduro yezvino.
- Vadzidzisi vari muvhidhiyo havatsamwise (kunyangwe mushanduro yeRussia havaverenge nemufaro sepakutanga)
- Kosi yacho haitore nguva yakawanda
- Kosi yacho haiiti kuti unzwe wakasuruvara kana kuti usina tariro. Mabasa ari mukosi ari nyore uye panogara paine zano muchimiro cheColab nemhinduro chaiyo kana chimwe chinhu chisina kujeka (uye hafu yakanaka yemabasa haina kujeka kwandiri)
- Iko hakuna chikonzero chekuisa chero chinhu, basa rese re laboratori yekosi rinogona kuitwa mubrowser
Zvino cons:
- Iko kunenge kusina zvinhu zvekudzora. Hapana bvunzo, hapana mabasa, hapana chekutarisa kugona kwekosi
- Haasi ese emanotipadhi angu akashanda sezvaaifanira kuita. Ini ndinofunga muchidzidzo chechitatu chekosi yepakutanga muChirungu Colab aive achikanda chikanganiso uye ndaisaziva zvekuita nazvo.
- Zviri nyore kuona chete pakombuta. Pamwe handina kuzvinzwisisa zvizere, asi handina kuwana iyo Udacity app pa smartphone yangu. Uye iyo mobile vhezheni yesaiti haipindure, ndiko kuti, ingangoita yese skrini nzvimbo inogarwa nemenyu yekufambisa, asi kuti uone iyo huru yemukati unofanirwa kupuruzira kurudyi kupfuura nzvimbo yekuona. Zvakare, vhidhiyo haigone kuonekwa parunhare. Iwe haugone kunyatsoona chero chinhu pachiratidziro chinorema inopfuura 6 inches.
- Zvimwe zvinhu mukosi zvinotsengwa kakati wandei, asi panguva imwe chete, izvo zvinonyatsodiwa zvinhu pa convolutional network pachazvo hazvina kutsengwa mukosi. Ini ndakanga ndisati ndanzwisisa chinangwa chese chemamwe maekisesaizi (semuenzaniso, iyo Max Pooling ndeyei).
Summary
Chokwadi waitofungidzira kuti chishamiso chacho hachina kuitika. Uye mushure mekupedza iyi kosi pfupi, hazvigoneke kunyatsonzwisisa kuti neural network inoshanda sei.
Ehe, mushure meizvi handina kukwanisa kugadzirisa dambudziko rangu ndega nekugadzika kwemifananidzo yekuchinja uye mabhatani mune switchgears.
Asi zvakazara kosi yacho inobatsira. Inoratidza izvo zvinhu zvinogona kuitwa neTensorFlow uye kuti ndeipi nzira yekutora inotevera.
Ini ndinofunga ini ndinofanira kutanga ndadzidza izvo zvekutanga zvePython uye kuverenga mabhuku muRussia nezve mashandiro anoita neural network, ndozotora TensorFlow.
Mukupedzisa, ndinoda kutaura kukutendai kushamwari dzangu nekundisundidzira kuti ndinyore chinyorwa chekutanga paHabr uye nekundibatsira kuigadzira.
PS Ndichafara kuona zvaunotaura uye chero kutsoropodza kunovaka.
Source: www.habr.com