Uhlaka lokufunda okujulile luqukethe inhlanganisela yokuxhumana, amalabhulali namathuluzi okuchaza nokuqeqesha amamodeli Okufunda Ngomshini ngokushesha nangokunembile.
Ngenxa yokuthi ukufunda okujulile kusebenzisa inani elikhulu ledatha engahlelekile, engeyona engokombhalo, udinga uhlaka olulawula ukusebenzisana phakathi "kwezendlalelo" futhi lenze ukuthuthukiswa kwemodeli kusheshe ngokufunda kudatha yokufaka nokwenza izinqumo ezizimele.
Uma ungathanda ukufunda ngokujulile ngokufunda ngo-2021, cabanga ukusebenzisa olunye lwezinhlaka ezikhonjiswe ngezansi. Khumbula ukukhetha eyodwa ezokusiza ukufeza izinhloso zakho kanye nombono wakho.
1. I-TensorFlow
Uma ukhuluma ngokufunda okujulile, I-TensorFlow ngokuvamile uhlaka lokuqala olushiwo. Okudume kakhulu, lolu hlaka alusetshenziswa i-Google kuphela - inkampani enesibopho sokudalwa kwayo - kodwa futhi nezinye izinkampani ezifana ne-Dropbox, i-eBay, i-Airbnb, i-Nvidia, nezinye eziningi.
I-TensorFlow ingasetshenziselwa ukwakha ama-API asezingeni eliphezulu neliphansi, okukuvumela ukuthi usebenzise izinhlelo zokusebenza cishe kunoma yiluphi uhlobo lwedivayisi. Nakuba iPython kuwulimi lwayo oluyinhloko, ukuxhumana kwe-Tensoflow kungafinyelelwa futhi kulawulwe kusetshenziswa ezinye izilimi zokuhlela ezifana ne-C++, Java, Julia, ne-JavaScript.
Njengoba ingumthombo ovulekile, i-TensorFlow ikuvumela ukuthi wenze ukuhlanganiswa okumbalwa namanye ama-API futhi uthole ukwesekwa okusheshayo nezibuyekezo ezivela emphakathini. Ukuthembela kwayo "kumagrafu amile" ukuze wenze izibalo kukuvumela ukuthi wenze izibalo ngokushesha noma ugcine imisebenzi ukuze uyifinyelele ngesinye isikhathi. Lezi zizathu, ezingezwe ekutheni kungenzeka "ubuke" ukuthuthukiswa kwenethiwekhi yakho ye-neural nge-TensorBoard, wenze i-TensorFlow uhlaka oludume kakhulu lokufunda ngokujulile.
Izici Key
- Umthombo ovulekile
- Ukuvumelana nezimo
- Ukulungisa iphutha
2. I-PyTorch
I-PyTorch wuhlaka olwakhiwe yi-Facebook ukuze lusekele ukusebenza kwezinsizakalo zayo. Selokhu laba umthombo ovulekile, lolu hlaka lusetshenziswe izinkampani ngaphandle kwe-Facebook, njenge-Salesforce ne-Udacity.
Lolu hlaka lusebenza ngamagrafu avuselelwe ngokushintshashintshayo, okukuvumela ukuthi wenze izinguquko ekwakhiweni kwedathasethi yakho njengoba uyicubungula. Nge-PyTorch kulula ukuthuthukisa nokuqeqesha inethiwekhi ye-neural, ngisho nangaphandle kwesipiliyoni ekufundeni okujulile.
Ukuba ngumthombo ovulekile futhi kusekelwe kuPython, ungenza ukuhlanganiswa okulula nokusheshayo ku-PyTorch. Futhi kuwuhlaka olulula lokufunda, ukulusebenzisa, nokulungisa iphutha. Uma unemibuzo, ungathembela ekusekelweni okukhulu nasezibuyekezweni ezivela emiphakathini yomibili - umphakathi wePython kanye nomphakathi wePyTorch.
Izici Key
- Kulula ukufunda
- Isekela i-GPU ne-CPU
- Isethi ecebile yama-API ukuze inwebe amalabhulali
3. I-Apache MXNet
Ngenxa yokukala kwayo okuphezulu, ukusebenza okuphezulu, ukuxazulula izinkinga ngokushesha, nokusekelwa kwe-GPU okuthuthukisiwe, lolu hlaka lwakhiwe yi-Apache ukuze lusetshenziswe kumaphrojekthi amakhulu ezimboni.
I-MXNet ihlanganisa isixhumi esibonakalayo se-Gluon esivumela onjiniyela bawo wonke amazinga wamakhono ukuthi qala ngokufunda okujulile emafini, kumadivayisi asemaphethelweni, nasezinhlelweni zokusebenza zeselula. Ngemigqa embalwa nje yekhodi ye-Gluon, ungakha ukuhlehla komugqa, amanethiwekhi okuguqula kanye nama-LSTM ajwayelekile ukutholwa kwento, ukuqashelwa kwenkulumo, isincomo, nokwenza kube ngokwakho.
I-MXNet ingasetshenziswa kumadivayisi ahlukahlukene futhi isekelwa ngabaningana izilimi zokuhlela njengeJava, R, JavaScript, Scala neGo. Nakuba inani labasebenzisi namalungu emphakathini walo liphansi, i-MXNet inemibhalo ebhalwe kahle namandla amakhulu okukhula, ikakhulukazi njengoba i-Amazon isikhethe lolu hlaka njengethuluzi eliyinhloko Lokufunda Ngomshini ku-AWS.
Izici Key
- 8 izilimi ezibophezelayo
- Ukuqeqeshwa Okusabalalisiwe, okusekela ama-multi-CPU nama-multi-GPU amasistimu
- I-hybrid front-end, evumela ukushintsha phakathi kwezindlela ezibalulekile nezingokomfanekiso
4. I-Microsoft Cognitive Toolkit
Uma ucabanga ukwenza izinhlelo zokusebenza noma izinsiza ezisebenza ku-Azure (izinsizakalo zefu ze-Microsoft), i-Microsoft Cognitive Toolkit iwuhlaka lokukhetha lwamaphrojekthi akho okufunda ajulile. Lona umthombo ovulekile, futhi usekelwa izilimi zokuhlela ezifana nePython, C++, C#, Java, phakathi kwezinye. Lolu hlaka luklanyelwe “ukucabanga njengobuchopho bomuntu”, ukuze lukwazi ukucubungula idatha enkulu engahlelekile, kuyilapho lunikeza ukuqeqeshwa okusheshayo kanye nezakhiwo ezinembile.
Ngokukhetha lolu hlaka - okufanayo ngemuva kwe-Skype, i-Xbox, ne-Cortana - uzothola ukusebenza okuhle ezinhlelweni zakho zokusebenza, ukuqina nokuhlanganiswa okulula ne-Azure. Kodwa-ke, uma kuqhathaniswa ne-TensorFlow noma i-PyTorch, inani lamalungu emphakathini walo nokusekelwa liyehla.
Ividiyo elandelayo inikeza isingeniso esiphelele nezibonelo zohlelo lokusebenza:
Izici Key
- Sula imibhalo
- Ukusekelwa okuvela eqenjini le-Microsoft
- Ukubonwa kwegrafu okuqondile
5. UKeras
Njenge-PyTorch, i-Keras iwumtapo wezincwadi osekwe kwiPython wamaphrojekthi adinga idatha. I-keras API isebenza ezingeni eliphezulu futhi ivumela ukuhlanganiswa nama-API wezinga eliphansi njenge-TensorFlow, i-Theano, ne-Microsoft Cognitive Toolkit.
Ezinye izinzuzo zokusebenzisa ama-keras wukuba lula kwawo ukufunda - ukuba wuhlaka olunconywayo lwabaqalayo ekufundeni okujulile; ijubane layo lokuthumela; enokwesekwa okukhulu okuvela emphakathini wezinhlwathi kanye nasemiphakathini yezinye izinhlaka ehlanganiswe nazo.
I-Keras iqukethe ukusetshenziswa okuhlukahlukene kwe- amabhlogo wokwakha amanethiwekhi we-neural njengezendlalelo, imisebenzi eyinhloso, imisebenzi yokuvula, nezilungiseleli zezibalo. Ikhodi yayo isingathwa ku-GitHub futhi kukhona izinkundla kanye nesiteshi sokusekela i-Slack. Ngaphezu kokusekela okujwayelekile amanethiwekhi we-neural, i-Keras inikezela ngosekelo lwe-Convolutional Neural Networks kanye namaNethiwekhi e-Recurrent Neural.
I-Keras ivumela amamodeli okufunda ajulile azokhiqizwa kuma-smartphone kukho kokubili i-iOS ne-Android, emshinini we-Java Virtual, noma kuwebhu. Iphinde ivumele ukusetshenziswa kokuqeqeshwa okusabalalisiwe kwamamodeli okufunda okujulile kumaqoqo Amayunithi Okucubungula Imidwebo (GPU) kanye Neyunithi Yokucubungula I-Tensor (TPU).
Izici Key
- Amamodeli aqeqeshwe ngaphambilini
- Ukusekelwa kwe-backend okuningi
- Ukusekelwa okusebenziseka kalula nomphakathi omkhulu
6. I-Apple Core ML
I-Core ML yathuthukiswa i-Apple ukuze isekele i-ecosystem yayo - i-IOS, i-Mac OS, ne-iPad OS. I-API yayo isebenza ezingeni eliphansi, isebenzisa kahle izinsiza ze-CPU kanye ne-GPU, okuvumela amamodeli nezinhlelo zokusebenza ezidalwe ukuthi ziqhubeke nokusebenza ngisho nangaphandle kokuxhumeka kwe-inthanethi, okunciphisa “inkumbulo yenkumbulo” kanye nokusetshenziswa kwamandla kwedivayisi.
Indlela i-Core ML ekufeza ngayo lokhu akukhona nje ngokwenza omunye umtapo wolwazi womshini olungiselelwe ukusebenza kuma-iphone/ipad. Kunalokho, i-Core ML ifana ne-compiler ethatha ukucaciswa kwemodeli nemingcele eqeqeshiwe evezwa enye isofthiwe yokufunda yomshini bese iyiguqulela kufayela eliba insiza yohlelo lokusebenza lwe-iOS. Lokhu kuguqulwa kumodeli ye-Core ML kwenzeka phakathi nokuthuthukiswa kohlelo lokusebenza, hhayi ngesikhathi sangempela njengoba kusetshenziswa uhlelo lokusebenza, futhi kuqondiswa umtapo wezincwadi we-coremltools python.
I-Core ML iletha ukusebenza okusheshayo ngokuhlanganiswa okulula kwe ukufunda imishini amamodeli kuzinhlelo zokusebenza. Isekela ukufunda okujulile ngezinhlobo ezingaphezu kwezingu-30 zezendlalelo kanye nezihlahla zesinqumo, imishini esekela i-vector, nezindlela zokuhlehla ngomugqa, zonke ezakhelwe phezu kobuchwepheshe bezinga eliphansi njenge-Metal ne-Accelerate.
Izici Key
- Kulula ukuhlanganisa kuzinhlelo zokusebenza
- Ukusetshenziswa kahle kwezinsiza zasendaweni, okungadingi ukufinyelela ku-inthanethi
- Ubumfihlo: idatha akudingeki ukuthi ishiye idivayisi
7. I-ONNX
Uhlaka lokugcina ohlwini lwethu yi-ONNX. Lolu hlaka lwavela ekusebenzisaneni phakathi kwe-Microsoft ne-Facebook, ngenhloso yokwenza lula inqubo yokudlulisa nokwakha amamodeli phakathi kwezinhlaka ezahlukene, amathuluzi, izikhathi zokusebenza kanye nabahlanganisi.
I-ONNX ichaza uhlobo lwefayela oluvamile olungasebenza ezisekelweni eziningi, kuyilapho kusetshenziswa izinzuzo zama-API asezingeni eliphansi njengalawo asuka ku-Microsoft Cognitive Toolkit, MXNet, Caffe kanye (nokusebenzisa iziguquli) i-Tensorflow ne-Core ML. Umgomo we-ONNX ukuqeqesha imodeli kusitaki futhi uyisebenzise kusetshenziswa ezinye iziqondiso nezibikezelo.
I-LF AI Foundation, inhlangano engaphansi kwe-Linux Foundation, iyinhlangano ezinikezele ekwakheni i-ecosystem ukuze isekele. evulekile-umthombo emisha kubuhlakani bokwenziwa (AI), ukufunda komshini (ML), nokufunda okujulile (DL). Yengeze i-ONNX njengephrojekthi yezinga lokuthweswa iziqu ngomhla ziyi-14 kuLwezi 2019. Lokhu kunyakaza kwe-ONNX ngaphansi kwesambulela se-LF AI Foundation kubonwe njengengqophamlando ekusunguleni i-ONNX njengezinga lefomethi evulekile yomthengisi engathathi hlangothi.
I-ONNX Model Zoo iqoqo lamamodeli aqeqeshwe kusengaphambili ku-Deep Learning etholakala ngefomethi ye-ONNX. Ukuze imodeli ngayinye kukhona Jupyter notebook ukuqeqeshwa kwamamodeli nokwenza inkomba ngemodeli eqeqeshiwe. Izincwadi zokubhalela zibhalwe ngePython futhi ziqukethe izixhumanisi ku- isethi yedatha yokuqeqeshwa kanye nezinkomba zombhalo wokuqala wesayensi ochaza imodeli yezakhiwo.
Izici Key
- Ukusebenzisana kohlaka
- Ukuthuthukisa Ihadiwe
Isiphetho
Lesi isifinyezo sezinhlaka ezinhle kakhulu ukufunda okujulile. Kunezinhlaka ezimbalwa zale njongo, zamahhala noma ezikhokhelwayo. Ukukhetha okungcono kakhulu kwephrojekthi yakho, qala wazi ukuthi iyiphi inkundla ozobe uthuthukisa kuyo uhlelo lwakho lokusebenza.
Izinhlaka ezijwayelekile ezifana ne-TensorFlow ne-Keras ziyizinketho ezinhle kakhulu ongaziqala. Kodwa uma udinga ukusebenzisa i-OS noma izinzuzo eziqondene nedivayisi, i-Core ML kanye ne-Microsoft Cognitive Toolkit kungase kube izinketho ezingcono kakhulu.
Kukhona ezinye izinhlaka eziqondiswe kumadivayisi e-Android, eminye imishini, nezinjongo ezithile ezingashiwongo kulolu hlu. Uma leli qembu lokugcina likuthakasela, siphakamisa ukuthi useshe ulwazi lwalo ku-Google noma kwamanye amasayithi okufunda ngomshini.
shiya impendulo