Uma ufunda lokhu, ngokungangabazeki usuvele uluqalile uhambo lwakho lokufunda ngokujulile. Uma umusha kulesi sihloko, ukufunda okujulile kuyisengezo esisebenzisa izakhiwo ezihlukile ezifana nobuchopho ezibizwa ngokuthi amanethiwekhi e-artificial neural ukuze kwakhiwe amakhompyutha afana nabantu abhekana nezinkinga zomhlaba wangempela.
Ukusiza ekuthuthukisweni kwale miklamo, ama-behemoth obuchwepheshe afana ne-Google, i-Facebook, ne-Uber benze izinhlaka ezihlukahlukene zemvelo yokufunda ejulile ye-Python, okwenza kube lula ukuqonda, ukudala, nokuqeqesha amanethiwekhi emizwa ahlukahlukene.
Uhlaka lokufunda olujulile luwucezu lwesofthiwe izifundiswa nososayensi bedatha abazisebenzisayo ukuze bakhe futhi baqeqeshe amamodeli okufunda ajulile.
Inhloso yalezi zinhlaka ukwenza kube nokwenzeka ukuthi abantu ngabanye baqeqeshe amamodeli abo ngaphandle kokuqonda amasu angemuva ukufunda okujulile, amanethiwekhi emizwa, nokufunda komshini.
Ngokusebenzisa isixhumi esibonakalayo sokuhlela sezinga eliphezulu, lezi zinhlaka zihlinzeka ngamabhulokhi wokwakha, ukuqeqesha, kanye namamodeli okuqinisekisa.
Sizobheka i-TensorFlow, i-Keras, i-Apache MXNet, i-Microsoft CNTK, ne-DeepLearing4j njengezinye izindlela esikhundleni se-PyTorch, esetshenziswa kakhulu. uhlaka lokufunda olujulile.
Iyini iPytorch?
I-PyTorch iyilabhulali yokufunda yomshini yamahhala, enomthombo ovulekile eyakhiwe ngomtapo wezincwadi weTorch Python.
Yakhiwe yiqembu le-Facebook ye-AI Research futhi yanyatheliswa njengomtapo wezincwadi wamahhala nomthombo ovulekile ngoJanuwari 2016 enezinhlelo zokusebenza ezibonwa ngekhompyutha, ukufunda okujulile, kanye nokucubungula ulimi lwemvelo.
Inolimi lokuhlela olubalulekile nolwePythonic olusekela ikhodi njengemodeli, olusiza ukulungisa amaphutha, futhi luhambisana namanye amalabhulali ekhompuyutha esayensi adumile, sonke isikhathi sihlala sisebenza kahle futhi sivumela ama-accelerator wehadiwe njengama-GPU.
I-PyTorch ikhule ekudumeni phakathi kwabacwaningi bokufunda okujulile ngenxa yokugxila kwayo ekusebenziseni nasekucatshangelweni kokusebenza okuphelele.
Iqukethe ukwakheka kwedatha eyisisekelo, i-Tensor, okuyi-multi-dimensional arrays efana ne-Numpy arrays, evumela abahleli bohlelo ukuthi baklame kalula inkimbinkimbi. inethiwekhi ye-neural.
Iya idume kakhulu emikhakheni yamanje kanye nasemphakathini wezemfundo ngenxa yokuguquguquka kwayo, isivinini, kanye nokusebenziseka kalula, okuyenza ibe elinye lamathuluzi okufunda ajulile aziwa kakhulu.
Izici ezibalulekile zePytorch
- I-PyTorch iyi-Python-centric, noma “pythonic,” ngoba ihloselwe ukuhlanganiswa okujulile nohlelo lwePython esikhundleni sokusebenza njengesixhumi esibonakalayo kumtapo wolwazi othuthukiswe ngolunye ulimi.
- Kulula Ukufunda - I-PyTorch ilandela isakhiwo esifanayo nesohlelo lwendabuko futhi ibhalwe ngokucophelela, umphakathi wonjiniyela uhlale uzama ukuyithuthukisa. Ngakho-ke kulula ukufunda kubo bobabili abahleli bezinhlelo nabangezona izinhlelo.
- I-PyTorch ingahlukanisa umsebenzi wokubala nge-CPU eminingana noma GPU ama-cores asebenzisa amandla e-data parallelism. Nakuba ukufana okufanayo kungafezwa ngamanye amasu okufunda ngomshini, i-PyTorch ikwenza kube lula kakhulu.
- Ukulungisa iphutha: Elinye lamathuluzi amaningi okulungisa iphutha ePython afinyeleleka kabanzi (isibonelo, amathuluzi ePython's pdb kanye ne-ipdb) lingasetshenziswa ukulungisa iphutha le-PyTorch.
- I-PyTorch isekela amagrafu ekhompyutha aguquguqukayo, okusho ukuthi ukuziphatha kwenethiwekhi kungashintshwa ngokuguqukayo phakathi nesikhathi sokusebenza.
- I-PyTorch iza namamojula ahlukahlukene adalwe ngokukhethekile, njenge umbhalo wethoshi, ithoshivision, nethoshi yomsindo, engasetshenziswa ukubhekana nemikhakha eyahlukene yokufunda okujulile, njenge-NLP, umbono wekhompyutha, nokucubungula izwi.
Imikhawulo yePytorch
- Ukuqapha okulinganiselwe nokubonwa kubonwa: Nakuba i-TensorFlow ihlanganisa ithuluzi lokubonisa elinamandla lokukhiqiza igrafu eyimodeli (i-TensorBoard), i-PyTorch okwamanje ayinaso lesi sici. Njengomphumela, abathuthukisi bangaxhumeka ku-TensorBoard ngaphandle noma basebenzise enye yenqwaba yePython ekhona. amathuluzi wokubona idatha.
- I-PyTorch ayiyona ukuphela kokuphela ukufunda imishini inkundla yokuthuthukisa; ithumela izinhlelo zokusebenza kumaseva, izindawo zokusebenza, namadivayisi eselula.
Kuzo zonke lezi zizathu, ukufuna ezinye izindlela ezingcono kakhulu zePytorch kungaba yisinqumo esihlakaniphile.
Ezinye izindlela ezidume kakhulu zePytorch
Nalu uhlu lwezinye izindlela ezingcono kakhulu ze-Pytorch.
1. I-Tensorflow
I-TensorFlow iwuhlaka olujulile olugxile ekufundeni, olunomthombo ovulekile oludalwe i-Google. Iphinde isekele izinga ukufunda imishini. I-TensorFlow yakhelwe kucatshangwa ngezibalo ezinkulu zezinombolo, kunokufunda okujulile.
Ngaphezu kwalokho, ibonakale iwusizo kakhulu ekuthuthukisweni kokufunda okujulile, ngakho i-Google ikwenze kutholakale mahhala. I-TensorFlow ithatha idatha ngesimo samalungu afanayo anezinhlangothi eziningi ezinobukhulu obukhulu, aziwa ngokuthi ama-tensor. Lapho usebenzisana namavolumu amakhulu wedatha, amalungu afanayo anezinhlangothi eziningi eza ngosizo.
I-TensorFlow isuselwe kumagrafu okugeleza kwedatha ye-node-edge. Ngenxa yokuthi indlela yokwenza ithatha uhlobo lwamagrafu, kulula kakhulu ukwenza ikhodi ye-TensorFlow phezu kweqoqo lamakhompyutha kuyilapho usebenzisa ama-GPU.
I-C#, i-Haskell, i-Julia, i-R, i-Ruby, i-Rust, ne-Scala ziphakathi kwezilimi umphakathi we-TensorFlow ozidalele ukusekelwa. I-TensorFlow inikeza inzuzo yokuba nenani elikhulu lamaphoyinti okufinyelela.
Ngaphandle kwezilimi, i-TensorFlow inohlu olukhulu lwamathuluzi axhumeka kuyo noma akhelwe phezu kwayo.
Izinzuzo
- Kuyasebenziseka. Uma ujwayelene nePython, kuzoba lula ukuyithatha.
- Ukwesekwa okuvela emphakathini. I-TensorFlow ithuthukiswa cishe nsuku zonke ngabathuthukisi abangochwepheshe be-Google nezinye izinhlangano.
- I-TensorFlow Lite ingasetshenziswa ukusebenzisa amamodeli e-TensorFlow kumadivayisi eselula.
- I-Tensorboard iyithuluzi lokuqapha kanye ukubona idatha. Uma ufuna ukubuka amamodeli akho okufunda okujulile esebenza, leli ithuluzi elihle kakhulu ongalisebenzisa.
- I-Tensorflow.js ikuvumela ukuthi usebenzise i-JavaScript ukuze usebenzise amamodeli okufunda okujulile ngesikhathi sangempela esipheqululini.
Okumbi
- I-TensorFlow inesakhiwo esiyingqayizivele, okwenza kube nzima ukuthola nokulungisa amaphutha.
- Alukho usekelo lwe-OpenCL.
- I-TensorFlow ayinikezi amandla amaningi kubasebenzisi bohlelo lokusebenza lweWindows. Ivula inqwaba yamakhono kubasebenzisi be-Linux. Kodwa-ke, abasebenzisi be-Windows basengakwazi ukulanda i-TensorFlow besebenzisa i-anaconda prompt noma iphakethe lepayipi.
- I-TensorFlow iwela ngemuva mayelana nokunikeza izihibe ezingokomfanekiso zokulandelana okungapheli. Inokusetshenziswa okuqondile kokulandelana okuthile, okuyenza ibe uhlelo olusebenzisekayo. Ngenxa yalokho, kubhekwa njenge-API yezinga eliphansi.
2. I-Keras
UKeras iyilabhulali yokufunda ejulile esekwe kwiPython, eyihlukanisa kwezinye izinhlaka zokufunda ezijulile.
Ulimi lokuhlela olusezingeni eliphezulu oluchaza a inethiwekhi ye-neural Incazelo ye-API. Ingasetshenziswa kokubili njengesixhumi esibonakalayo somsebenzisi kanye nokuthuthukisa amakhono ezinhlaka zokufunda ezijulile esebenza kuzo.
Kuwuhlaka lwe-minimalist olungasindi futhi olusebenziseka kalula. Ngalezi zizathu, i-Keras iyingxenye ye-TensorFlow's core API. Isiphetho sangaphambili se-Keras sivumela ukwenziwa kwe-prototyping esheshayo yamamodeli enethiwekhi ye-neural ocwaningweni.
I-API iqondile ukuthi uyibambe futhi uyisebenzise, ngebhonasi eyengeziwe yokuvumela amamodeli ukuthi adluliswe kalula phakathi kwezinhlaka.
Izinzuzo
- I-Keras API ilula ukuyisebenzisa. I-API yakhelwe kahle, igxile entweni, futhi iyavumelana nezimo, okuholela ekuhlangenwe nakho okujabulisayo komsebenzisi.
- Ukusekelwa kokuqeqeshwa okusabalalisiwe kanye nokufana kwe-GPU eminingi yakhelwe ngaphakathi.
- I-Keras iyimojula yomdabu yePython enikeza ukufinyelela okulula endaweni ephelele yesayensi yedatha yePython. Amamodeli we-Keras, isibonelo, angasetshenziswa kusetshenziswa i-Python scikit-learn API.
- I-Keras ihlanganisa izisindo eziqeqeshwe kusengaphambili zamamodeli amaningana okufunda okujulile. Singasebenzisa lawa mamodeli ngokuqondile ukuze senze izibikezelo noma sikhiphe izici.
Okumbi
- Kungase kucasule ngendlela emangalisayo ukuthola izinkinga ezisezingeni eliphansi njalo. Lezi zinkinga ziphakama uma sizama ukwenza imisebenzi uKeras obengahloselwe ukuyifeza.
- Uma iqhathaniswa nezingemuva zayo, ingase ibe buthaka kuma-GPU futhi ithathe isikhathi eside ukubala. Ngenxa yalokho, kungase kudingeke ukuthi sehlise isivinini ukuze sisebenziseke kalula.
- Uma kuqhathaniswa namanye amaphakheji afana ne-sci-kit-learn, amandla e-Keras okucubungula idatha awakhangi kangako.
3. I-Apache MXNet
Enye evelele Uhlaka lokufunda okujulile iMXNet. I-MXNet, eyakhiwe i-Apache Software Foundation, isekela izilimi ezihlukahlukene, okuhlanganisa i-JavaScript, i-Python, ne-C++.
I-Amazon Web Services iphinde isekele i-MXNet ekuthuthukiseni amamodeli okufunda ajulile. Iyakala ngokwedlulele, ivumela ukuqeqeshwa kwemodeli esheshayo, futhi ihambisana nezinhlobonhlobo zezilimi zekhompyutha.
Ukuze uthuthukise isivinini nokukhiqiza, i-MXNet ikuvumela ukuthi uhlanganise izilimi zokuhlela ezingokomfanekiso nezibalulekile. Isekelwe kusihleli sokuncika esiguqukayo esihambisana nemisebenzi engokomfanekiso nebalulekile ngesikhathi sangempela.
Ngaphezu kwalokho, isendlalelo sokuthuthukisa igrafu senza ukubulawa okungokomfanekiso kusheshe nenkumbulo ukonga. I-MXNet iwumtapo wezincwadi ophathwayo futhi ongasindi.
Inikwa amandla yi-NVIDIA PascalTM GPUs futhi iyakala ngama-GPU amaningana namanodi, okukuvumela ukuthi uqeqeshe amamodeli ngokushesha okukhulu.
Izinzuzo
- Isekela ama-GPU futhi inemodi ye-GPU eminingi.
- Isebenza kahle, iyanwebeka, futhi iyashesha kakhulu.
- Wonke amapulatifomu amakhulu akhona.
- Ukusebenzisa imodeli kulula, futhi i-API iyashesha.
- I-Scala, i-R, i-Python, i-C++, ne-JavaScript ziphakathi kwezilimi zokuhlela ezisekelwayo.
Okumbi
- I-MXNet inokuncane umthombo ovulekile umphakathi kune-TensorFlow.
- Ukuthuthukiswa, ukulungiswa kweziphazamisi, nokunye ukuthuthukiswa kuthatha isikhathi eside ukuqaliswa ngenxa yokuntuleka kosekelo olubalulekile lomphakathi.
- I-MxNet, nakuba isetshenziswa kabanzi amafemu amaningi embonini ye-IT, awaziwa kakhulu njenge-Tensorflow.
4. I-Microsoft CNTK
I-Microsoft Cognitive Toolkit (CNTK) iwuhlaka lomthombo ovulekile olusebenza ngokwezentengiselwano lokusabalalisa ukufunda okujulile. Ngokuvamile isetshenziselwa ukudala amanethiwekhi we-neural, kodwa futhi ingasetshenziselwa ukufunda ngomshini kanye nekhompuyutha yokuqonda.
Isekela izilimi ezihlukahlukene futhi kulula ukuyisebenzisa efwini. Ngenxa yalezi zimfanelo, i-CNTK ifaneleka ezinhlobonhlobo zezinhlelo zokusebenza ze-AI. Yize singasebenzisa i-C++ ukuze sicele imisebenzi yayo, inketho evame kakhulu ukusebenzisa uhlelo lwePython.
Uma isebenza kumakhompuyutha amaningana, i-Microsoft Cognitive Toolkit ibonwa ukuze inikeze ukusebenza okungcono nokuqina kunamathuluzi anjenge-Theano noma i-TensorFlow.
I-Microsoft Cognitive Toolkit isekela womabili amamodeli we-RNN kanye ne-CNN, ayenze ifanelekele izithombe, ukubhala ngesandla, nemisebenzi yokuqaphela inkulumo.
Izinzuzo
- Kulula ukuhlanganisa ne-Apache Spark, injini yokuhlaziya idatha.
- Ukulinganisa kwe-CNTK kwenze kwaba ukukhetha okudumile emabhizinisini amaningi. Kunezingxenye ezimbalwa ezithuthukisiwe.
- Inikeza ukusebenza okuzinzile nokuhle.
- Isebenza kahle nge-Azure Cloud, yomibili esekelwa yiMicrosoft.
- Ukusetshenziswa nokuphathwa kwezinsiza kusebenza kahle.
Okumbi
- Uma kuqhathaniswa ne-Tensorflow, kuncane ukwesekwa komphakathi.
- Ijika lokufunda eliwumqansa.
- Ayinalo ibhodi lokubuka kanye nokwesekwa kwe-ARM.
5. I-DeepLearning4j
Uma i-Java kuwulimi lwakho oluyinhloko lokuhlela, i-DeepLearning4j iwuhlaka oluhle ongalusebenzisa. Iwumtapo wolwazi osabalalisiwe osezingeni lokuhweba nomthombo ovulekile.
Zonke izinhlobo eziyinhloko zemiklamo yenethiwekhi ye-neural, njengama-RNN nama-CNN, ziyasekelwa. I-Deeplearning4j iyilabhulali ye-Java ne-Scala yokufunda ngokujulile.
Isebenza kahle nge-Hadoop ne-Apache Spark futhi. I-Deeplearning4j ingenye indlela emangalisayo yezixazululo zokufunda okujulile ezisuselwa ku-Java ngoba futhi isekela ama-GPU.
Uma kukhulunywa ngohlaka lokufunda okujulile lwe-Eclipse Deeplearning4j, ezinye zezici ezivelele zifaka ukuqeqeshwa okufanayo ngokunciphisa okuphindaphindayo, ukulungiswa kwezakhiwo zesevisi encane, kanye nama-CPU nama-GPU asabalalisiwe.
Izinzuzo
- Inemibhalo emihle kakhulu nosizo lomphakathi.
- Ukuhlanganiswa kwe-Apache Spark kulula.
- Iyakwazi ukukala futhi iyakwazi ukuphatha amavolumu amakhulu wedatha.
Okumbi
- Uma kuqhathaniswa ne-Tensorflow ne-PyTorch, ayidumile kangako.
- I-Java ukuphela kolimi lokuhlela olutholakalayo.
Isiphetho
Ukukhetha uhlaka lokufunda olujulile olungcono kakhulu kuwumsebenzi onzima. Kakhulu njengoba bebaningi kakhulu, uhlu luyakhula njengesidingo ukuhlakanipha okungekhona okwangempela ucwaningo kanye nezicelo zokufunda ngomshini ziyakhula. Uhlaka ngalunye lunesethi yalo yezinto ezinhle nezingezinhle.
Kufanele kucatshangelwe izinto eziningi, okuhlanganisa ukuphepha, ukukala, kanye nokusebenza. Ezinhlelweni zebanga lebhizinisi, ukwethembeka kuba okubaluleke kakhulu.
Uma usaqala, i-Tensorflow iyindawo enhle yokuqala. Khetha i-CNTK uma uthuthukisa umkhiqizo wezohwebo osuselwe ku-Windows. Uma ukhetha i-Java, sebenzisa i-DL4J.
shiya impendulo