Kana uri kuverenga izvi, hapana mubvunzo kuti watotanga rwendo rwako mukudzidza kwakadzama. Kana iwe uri mutsva kunyaya iyi, kudzidza kwakadzama ndeyekuwedzera-pamwe kunoshandisa akasiyana uropi-senge zvimiro zvinonzi artificial neural network kugadzira makomputa akafanana nevanhu anogadzirisa nyaya dzepasirese.
Kubatsira mukusimudzira magadzirirwo aya, matekinoroji ematekinoroji akaita seGoogle, Facebook, uye Uber akagadzira akasiyana masisitimu ePython yakadzika nharaunda yekudzidza, zvichiita kuti zvive nyore kunzwisisa, kugadzira, uye kudzidzisa akasiyana neural network.
Hurongwa hwekudzidza hwakadzama ichidimbu chesoftware inoshandiswa nevadzidzi uye nesainzi yedata kugadzira nekudzidzisa mamodheru ekudzidza akadzama.
Chinangwa chezvirongwa izvi ndechekuita kuti vanhu vakwanise kudzidzisa mamodheru avo vasinganzwisise maitiro ari kumashure kudzidza zvakadzika, neural network, uye kudzidza muchina.
Kuburikidza neiyo yepamusoro-level programming interface, aya masisitimu anopa zvivharo zvekuvaka, kudzidzisa, uye yekusimbisa mhando.
Tichatarisa TensorFlow, Keras, Apache MXNet, Microsoft CNTK, uye DeepLearing4j sedzimwe nzira dzePyTorch, iyo inoshandiswa zvakanyanya. hurongwa hwekudzidza hwakadzama.
Chii chinonzi Pytorch?
PyTorch iraibhurari yemahara, yakavhurika-sosi yemuchina yekudzidza yakavakwa neTorch Python raibhurari.
Yakagadzirwa neFacebook's AI Research group uye yakaburitswa seyemahara uye yakavhurika-sosi raibhurari muna Ndira 2016 ine maapplication ari muchiratidzo chekombuta, kudzidza kwakadzama, uye kugadzirwa kwemutauro wechisikigo.
Iyo ine yakakosha uye Pythonic programming mutauro inotsigira kodhi semuenzaniso, inofambisa debugging, uye inofambirana nemamwe anozivikanwa sainzi komputa maraibhurari, zvese zvichiramba zvichibudirira uye zvichigonesa hardware accelerators seGPUs.
PyTorch yakakura mukuzivikanwa pakati pevaongorori vekudzidza kwakadzama nekuda kwekutarisa kwayo kwekushandisa uye kunyatsoita kufunga kwekuita.
Iyo ine yakakosha data chimiro, Tensor, inova yakawanda-dimensional array yakafanana neNumpy arrays, iyo inobvumira vanogadzira zvirongwa nyore kugadzira yakaoma. neural network.
Iri kuita mukurumbira muzvikamu zvazvino uye munharaunda yezvidzidzo nekuda kwekuchinjika kwayo, kukurumidza, uye nyore kuita, zvichiita kuti ive imwe yezvinonyanya kufarirwa zvakadzika zvekushandisa.
Pytorch Key Features
- PyTorch iPython-centric, kana kuti "pythonic," mukuti inoitirwa kubatanidza kwakadzama nePython programming pane kushanda sechiratidziro kuraibhurari yakagadzirwa mune mumwe mutauro.
- Nyore Kudzidza - PyTorch inotevera chimiro chakafanana neyechinyakare hurongwa uye yakanyorwa zvine hungwaru, nenharaunda yekuvandudza inogara ichiedza kuisimudzira. Saka zviri nyore kudzidza kune vese vanogadzira uye vasiri-programmers.
- PyTorch inogona kugovera computational basa pane akati wandei CPU kana GPU cores vachishandisa data parallelism kugona. Kunyangwe kufanana kwakafanana kuchigona kuitwa nemamwe matekiniki ekudzidza muchina, PyTorch inoita kuti zvive nyore.
- Debugging: Imwe yeakawanda anowanikwa anowanikwa Python debugging maturusi (semuenzaniso, Python's pdb uye ipdb zvishandiso) inogona kushandiswa kugadzirisa PyTorch.
- PyTorch inotsigira dynamic computational graphs, izvo zvinoreva kuti maitiro etiweki anogona kuchinjwa zvine simba panguva yekumhanya.
- PyTorch inouya neakasiyana akagadzirwa mamodule, akadai torchtext, torchvision, uye torchaudio, iyo inogona kushandiswa kubata nenzvimbo dzakasiyana dzekudzidza kwakadzama, senge NLP, kuona komputa, uye kugadzirisa izwi.
Pytorch Limitations
- Yakaganhurirwa yekutarisa uye yekuona maratidziro: Nepo TensorFlow ichisanganisira ine simba rekuona chishandiso kugadzira iyo modhi girafu (TensorBoard), PyTorch parizvino haina chinhu ichi. Nekuda kweizvozvo, vanogadzira vanogona kubatana neTensorBoard kunze kana kushandisa imwe yeakawanda aripo Python. zvishandiso zvekuona data.
- PyTorch haisi yekupedzisira-ku-kuguma machine learning Development platform; inotumira zvikumbiro kumaseva, nzvimbo dzekushandira, uye nharembozha.
Nekuda kwezvikonzero zvese izvi, kutsvaga nzira dzakanakisa dzePytorch chingave sarudzo yehungwaru.
Yakanyanya kufarirwa dzimwe nzira dzePytorch
Heino rondedzero yezvakanakisa dzimwe nzira dzePytorch.
1. Tensorflow
TensorFlow idzidzo yakadzama-yakatarisana, yakavhurika-sosi chimiro chakagadzirwa neGoogle. Inotsigirawo mwero machine learning. TensorFlow yakagadzirwa ine hombe manhamba ekuverenga mupfungwa, pane kudzidza kwakadzama.
Uyezve, yakaratidza kuve yakakosha pakukura kwekudzidza kwakadzama zvakare, saka Google yakaita kuti iwanikwe mahara. TensorFlow inotora data muchimiro cheakawanda-dimensional arrays ane hukuru hukuru, hunozivikanwa sematensor. Paunenge uchibata neakawanda mavhoriyamu e data, akawanda-dimensional arrays anouya anobatsira.
TensorFlow yakavakirwa pane node-kumucheto data kuyerera magirafu. Nekuti nzira yekuuraya inotora chimiro chegirafu, zviri nyore kuita TensorFlow kodhi pamusoro peboka remakomputa uchishandisa maGPU.
C#, Haskell, Julia, R, Ruby, Rust, uye Scala ndedzimwe mitauro iyo nharaunda yeTensorFlow yakagadzira rutsigiro. TensorFlow inopa bhenefiti yekuve nehuwandu hukuru hwenzvimbo dzekuwana.
Kunze kwemitauro, TensorFlow ine huwandu hukuru hwezvishandiso zvinobatana nazvo kana zvakavakwa pamusoro payo.
Advantages
- Iri nyore kushandisa. Kana iwe uchijairana nePython, zvichave nyore kutora.
- Rutsigiro kubva munharaunda. TensorFlow inovandudzwa mazuva ese neGoogle nemamwe masangano vanogadzira.
- TensorFlow Lite inogona kushandiswa kuita TensorFlow modhi pane nharembozha.
- Tensorboard chishandiso chekutarisa uye kuona data. Kana iwe uchida kuona yako yakadzama yekudzidza modhi mukuita, ichi chishandiso chakanakisa kushandisa.
- Tensorflow.js inokutendera kuti ushandise JavaScript kumhanya chaiyo-nguva yakadzama modhi mubrowser.
payakaipira
- TensorFlow ine yakasarudzika chimiro, zvichiita kuti zvinyanye kunetsa kuwana uye kugadzirisa zvikanganiso.
- Iko hakuna OpenCL rutsigiro.
- TensorFlow haipe akawanda masimba kune vashandisi veWindows inoshanda system. Inovhura kuwanda kwekugona kwevashandisi veLinux. Nekudaro, vashandisi veWindows vanogona kudhawunirodha TensorFlow vachishandisa anaconda kukurumidza kana pip package.
- TensorFlow inowira kumashure maererano nekupa zvishwe zvekufananidzira kune dzisingaperi kutevedzana. Iine mashandisirwo chaiwo ekutevedzana kwakasiyana, zvichiita kuti ive inoshandisika sisitimu. Nekuda kweizvozvo, inoonekwa seyakaderera-level API.
2. Keras
Keras iraibhurari yePython-yakavakirwa pakadzika yekudzidza, inoisiyanisa kubva kune mamwe maitiro akadzama ekudzidza.
Ndiwo mutauro wepamusoro-soro wepurogiramu unotsanangura a neural network Tsanangudzo yeAPI. Inogona kushandiswa zvese semushandisi interface uye kuvandudza kugona kweiyo yakadzama yekudzidza masisitimu painomhanyisa.
Iyo minimalist chimiro icho chakareruka uye chiri nyore kushandisa. Nezvikonzero izvi, Keras chikamu cheTensorFlow's musimboti API. Iyo Keras yekumberi yekupedzisira inobvumira kukurumidza prototyping yeneural network modhi mukutsvaga.
Iyo API yakatwasuka kubata nekushandisa, iine bhonasi yakawedzerwa yekubvumidza mamodheru kutamiswa zviri nyore pakati pezvimiro.
Advantages
- Iyo Keras API iri nyore kushandisa. Iyo API yakanyatsogadzirwa, yakatarisana nechinhu, uye inochinjika, zvichiita kuti iwedzere kunakidza mushandisi ruzivo.
- Tsigiro yekudzidziswa kwakagoverwa uye yakawanda-GPU parallelism yakavakirwa-mukati.
- Keras iPython yemuno module inopa nyore kuwana kune yakazara Python data sainzi nharaunda. Keras modhi, semuenzaniso, inogona kushandiswa uchishandisa iyo Python scikit-kudzidza API.
- Keras inosanganisira pre-yakadzidziswa uremu kune akati wandei akadzama emhando yekudzidza. Tinogona kushandisa aya mamodheru zvakananga kuita fungidziro kana kubvisa maficha.
payakaipira
- Zvinogona kushungurudza zvakanyanya kuwana yakaderera-level backend nyaya nguva dzose. Aya matambudziko anomuka patinoyedza kuita mabasa ayo Keras anga asina kuitirwa kuita.
- Kana ichienzaniswa nekumashure kwayo, inogona kuve nehusimbe paGPU uye kutora nguva yakareba kuti iverenge. Nekuda kweizvozvo, isu tingangofanira kukanganisa kumhanya kwekushandisa-hushamwari.
- Kana ichienzaniswa nemamwe mapakeji akadai se-sci-kit-dzidza, Keras data-preprocessing masimba haana kunaka.
3. Apache MX Net
Imwe yakakurumbira Deep Learning framework iri MXNet. MXNet, iyo yakagadzirwa neApache Software Foundation, inotsigira mitauro yakasiyana-siyana, kusanganisira JavaScript, Python, uye C++.
Amazon Web Services inotsigirawo MXNet mukuvandudza kwemhando dzekudzidza dzakadzika. Yakanyanya scalable, inobvumira kukurumidza kudzidziswa modhi, uye inoenderana neakasiyana mitauro yemakomputa.
Kukwirisa kumhanya uye kugadzira, MXNet inokutendera iwe kusanganisa inomiririra uye yakakosha mitauro yekuronga. Izvo zvakavakirwa pane ane dynamic dependency scheduler inofananidzira zviitiko zvekufananidzira uye zvakakosha munguva-chaiyo.
Pamusoro peizvozvo, girafu optimization layer inoita kuuraya kwekufananidzira nekukurumidza uye ndangariro hupfumi. MXNet raibhurari inotakurika uye isingaremi.
Inofambiswa neNVIDIA Pascal TM GPUs uye inogona scalable pamusoro akati wandei maGPU uye node, zvichikubvumidza iwe kudzidzisa modhi nekukurumidza.
Advantages
- Inotsigira maGPU uye ine akawanda-GPU modhi.
- Hunoshanda, scalable, uye mheni-nekukurumidza.
- Mapuratifomu makuru ese ari mubhodhi.
- Model kushumira iri nyore, uye API inokurumidza.
- Scala, R, Python, C++, uye JavaScript ndeimwe yemitauro inotsigirwa.
payakaipira
- MXNet ine diki open source nharaunda kupfuura TensorFlow.
- Kunatsiridzwa, kugadzirisa kwebug, uye kumwe kuvandudzwa kunotora nguva yakareba kuti iite nekuda kwekushaikwa kwerutsigiro rwakakosha munharaunda.
- MxNet, kunyangwe ichishandiswa zvakanyanya nemafemu akawanda muIT indasitiri, haina kunyanya kuzivikanwa seTensorflow.
4. Microsoft CNTK
Microsoft Cognitive Toolkit (CNTK) inzvimbo inotengeswa yakavhurika-sosi yekugovera kudzidza kwakadzama. Kazhinji inoshandiswa kugadzira neural networks, asi inogona zvakare kushandiswa pakudzidza muchina uye cognitive computing.
Inotsigira mitauro yakasiyana-siyana uye iri nyore kushandisa pagore. Nekuda kwehunhu uhu, CNTK inokodzera kwakasiyana siyana kweAI application. Kunyangwe isu tichigona kushandisa C ++ kukumbira mabasa ayo, yakajairika sarudzo ndeye kushandisa chirongwa chePython.
Paunenge uchimhanya pamakomputa akati wandei, iyo Microsoft Cognitive Toolkit inozivikanwa kupa zvirinani kuita uye scalability pane zvishandiso seTheano kana TensorFlow.
Iyo Microsoft Cognitive Toolkit inotsigira ese ari maviri eRNN neCNN neural modhi, ichiita kuti ive yakakodzera mufananidzo, kunyora nemaoko, uye mataurirwo ekutaura mabasa.
Advantages
- Yakareruka kusanganisa neApache Spark, data analytics injini.
- CNTK's scalability yaita kuti ive sarudzo yakakurumbira mumabhizinesi mazhinji. Pane akati wandei optimized components.
- Inopa yakagadzikana uye yakanaka kuita.
- Inoshanda zvakanaka neAzure Cloud, ese ari maviri anotsigirwa neMicrosoft.
- Kushandisa zviwanikwa uye manejimendi zvinobudirira.
payakaipira
- Mukuenzanisa neTensorflow, kune kushomeka kwerutsigiro rwenharaunda.
- Nzira yekudzidza yakadzika.
- Iyo inoshaya bhodhi rekuona pamwe nerutsigiro rweARM.
5. DeepLearning4j
Kana Java iriyo mutauro wako wekutanga wekugadzira, DeepLearning4j igadziriro yakanaka yekushandisa. Iyo yakagoverwa yakadzika-yekudzidza raibhurari iyo yekutengesa-giredhi uye yakavhurika-sosi.
Ese marudzi makuru ekugadzira neural network, senge maRNN neCNNs, anotsigirwa. Deeplearning4j iJava uye Scala raibhurari yekudzidza kwakadzama.
Inoshanda zvakanaka neHadoop uye Apache Spark zvakare. Deeplearning4j inzira inoshamisa yeJava-yakavakirwa zvakadzama mhinduro dzekudzidza nekuti zvakare inotsigira maGPU.
Kana zvasvika kune Eclipse Deeplearning4j yakadzama yekudzidza chimiro, zvimwe zvezvakamira zvinosanganisira kudzidziswa kwakafanana kuburikidza nekudzikisira iterative, micro-service architecture adaptation, uye akagovera maCPU nemaGPU.
Advantages
- Iine zvinyorwa zvakanakisa uye rubatsiro rwenharaunda.
- Iyo Apache Spark yekubatanidza iri nyore.
- Iyo ine scalable uye inokwanisa kubata yakakura mavhoriyamu e data.
payakaipira
- Mukuenzanisa neTensorflow nePyTorch, haina kufarirwa.
- Java ndiyo yega programming language iripo.
mhedziso
Kusarudza yakanakisa dhizaini yekudzidza ibasa rakaoma. Kunyanya sezvo kune akawanda acho, rondedzero iri kukura sekuda kwe chakagadzirwa njere tsvakiridzo uye mashandisirwo ekudzidza muchina anokura. Chimiro chega chega chine seti yezvakanakira uye zvakaderera.
Kufunga kwakawanda kunofanirwa kuitwa, kusanganisira chengetedzo, scalability, uye kuita. Mumabhizinesi-giredhi masisitimu, kuvimbika kunotonyanya kukosha.
Kana iwe uchangotanga kunze, Tensorflow inzvimbo yakanaka yekutanga. Sarudza CNTK kana uri kugadzira Windows-based commercial product. Kana uchida Java, shandisa DL4J.
Leave a Reply