Yog tias koj nyeem qhov no, koj twb tsis ntseeg tau pib koj txoj kev mus rau kev kawm tob. Yog tias koj yog tus tshiab rau lub ntsiab lus no, kev kawm tob yog ib qho ntxiv uas siv cov qauv zoo li lub hlwb hu ua cov khoom siv neural networks los tsim cov khoos phis tawj zoo li tib neeg uas daws teeb meem hauv ntiaj teb tiag.
Txhawm rau pab txhim kho cov qauv tsim no, tech behemoths zoo li Google, Facebook, thiab Uber tau tsim ntau yam haujlwm rau Python qhov chaw kawm sib sib zog nqus, ua kom yooj yim rau kev nkag siab, tsim, thiab cob qhia ntau hom neural networks.
Lub hauv paus kev kawm tob yog ib qho ntawm cov software uas cov kws tshawb fawb thiab cov ntaub ntawv tshawb fawb siv los tsim thiab cob qhia cov qauv kev kawm tob.
Lub hom phiaj ntawm cov txheej txheem no yog ua kom cov tib neeg tuaj yeem cob qhia lawv cov qauv yam tsis tas yuav nkag siab txog cov txheej txheem tom qab kawm tob, neural networks, thiab machine learning.
Los ntawm cov txheej txheem kev sib txuas lus siab, cov txheej txheem no muab lub tsev thaiv kev tsim, kev cob qhia, thiab kev txheeb xyuas cov qauv.
Peb yuav saib TensorFlow, Keras, Apache MXNet, Microsoft CNTK, thiab DeepLearing4j ua lwm txoj hauv kev rau PyTorch, uas yog siv dav. kev kawm tob.
Pytorch yog dab tsi?
PyTorch yog lub tsev qiv ntawv dawb, qhib lub tshuab kev kawm tsim nrog lub tsev qiv ntawv Torch Python.
Nws tau tsim los ntawm Facebook's AI Research pab pawg thiab luam tawm raws li lub tsev qiv ntawv pub dawb thiab qhib rau lub Ib Hlis 2016 nrog cov ntawv thov hauv computer tsis pom kev, kev kawm tob, thiab kev ua cov lus ntuj.
Nws muaj qhov tseem ceeb thiab Pytonic programming lus uas txhawb nqa code ua tus qauv, pab txhawb kev debugging, thiab tau sib xws nrog lwm cov tsev qiv ntawv suav nrog nrov, txhua lub sijhawm ua haujlwm tau zoo thiab ua rau cov khoom siv nrawm xws li GPUs.
PyTorch tau loj hlob hauv kev muaj koob npe ntawm cov kws tshawb fawb tob tob ua tsaug rau nws txoj kev tsom mus rau kev siv tau thiab kev xav txog kev ua tau zoo.
Nws muaj cov qauv ntaub ntawv yooj yim, Tensor, uas yog ntau qhov sib txawv zoo ib yam li Numpy arrays, uas tso cai rau cov neeg ua haujlwm tau yooj yim tsim qhov nyuaj. neural network.
Nws tau dhau los ua neeg nyiam nyob hauv cov haujlwm tam sim no thiab hauv zej zog kev kawm vim nws qhov yooj yim, nrawm, thiab yooj yim ntawm kev siv, ua rau nws yog ib qho ntawm cov cuab yeej kawm tob tshaj plaws.
Cov yam ntxwv tseem ceeb ntawm Pytorg
- PyTorch yog Python-centric, los yog "pythonic," nyob rau hauv uas nws yog txhais tau tias rau sib sib zog nqus kev koom ua ke nrog Python programming es tsis ua hauj lwm raws li ib tug interface rau lub tsev qiv ntawv tsim nyob rau hauv lwm yam lus.
- Yooj yim Kawm - PyTorch ua raws li cov qauv zoo ib yam li kev ua haujlwm ib txwm muaj thiab tau sau tseg kom zoo, nrog cov neeg tsim tawm hauv zej zog ib txwm sim txhim kho nws. Yog li nws yooj yim kawm rau ob tus programmers thiab tsis yog programmers.
- PyTorch tuaj yeem faib ua haujlwm suav nrog ntau lub CPU lossis GPU cores siv cov ntaub ntawv parallelism muaj peev xwm. Txawm hais tias kev sib piv zoo sib xws tuaj yeem ua tiav nrog lwm cov kev kawm tshuab, PyTorch ua kom yooj yim dua.
- Kev debugging: Ib qho ntawm cov cuab yeej siv tau dav dav Python debugging (piv txwv li, Python's pdb thiab ipdb cov cuab yeej) tuaj yeem siv los kho PyTorch.
- PyTorch txhawb nqa cov duab kos duab dynamic, uas txhais tau hais tias tus cwj pwm ntawm lub network tuaj yeem hloov pauv dynamically thaum lub sijhawm ua haujlwm.
- PyTorch los nrog ntau yam tshwj xeeb tsim modules, xws li torchtext, torchvision, thiab torchaudio, uas tuaj yeem siv los cuam tshuam nrog ntau hom kev kawm sib sib zog nqus, xws li NLP, khoos phis tawj pom, thiab ua suab.
Pytorch txwv
- Kev saib xyuas tsis pub dhau thiab pom kev cuam tshuam: Thaum TensorFlow suav nrog cov cuab yeej pom muaj zog rau kev tsim cov qauv duab (TensorBoard), PyTorch tam sim no tsis muaj qhov tshwj xeeb no. Yog li ntawd, cov neeg tsim khoom tuaj yeem txuas rau TensorBoard sab nraud lossis siv ib qho ntawm ntau tus Python uas twb muaj lawm. cov ntaub ntawv kom pom cov cuab yeej.
- PyTorch tsis yog qhov kawg-rau-kawg tshuab kev kawm kev txhim kho platform; nws xa cov ntawv thov rau servers, chaw ua haujlwm, thiab cov khoom siv mobile.
Rau tag nrho cov laj thawj no, nrhiav txoj hauv kev zoo tshaj plaws rau Pytorg yuav yog qhov kev txiav txim siab zoo.
Cov kev xaiv nrov tshaj plaws Pytarch
Nov yog cov npe ntawm cov kev xaiv zoo tshaj plaws rau Pytorg.
1. Tensorflow
TensorFlow yog ib qho kev kawm sib sib zog nqus-tsim, qhib-qhov chaw tsim los ntawm Google. Nws kuj txhawb tus qauv tshuab kev kawm. TensorFlow tau tsim nrog kev suav lej loj hauv siab, tsis yog kev kawm tob.
Tsis tas li ntawd, nws tau ua pov thawj tias muaj txiaj ntsig zoo rau kev txhim kho kev kawm sib sib zog nqus, yog li Google ua rau nws muaj pub dawb. TensorFlow siv cov ntaub ntawv nyob rau hauv daim ntawv ntawm ntau qhov arrays nrog ntau qhov ntev, hu ua tensors. Thaum cuam tshuam nrog cov ntaub ntawv loj loj, ntau qhov arrays tuaj yeem pab tau.
TensorFlow yog raws li cov ntaub ntawv ntawm cov ntaub ntawv ntws. Vim tias txoj kev ua tiav siv cov duab kos, nws yooj yim dua los ua TensorFlow code hla ib pawg ntawm cov khoos phis tawj thaum siv GPUs.
C#, Haskell, Julia, R, Ruby, Rust, thiab Scala yog cov lus uas TensorFlow lub zej zog tau tsim kev txhawb nqa rau. TensorFlow muab cov txiaj ntsig ntawm kev muaj ntau cov ntsiab lus nkag.
Ib cag ntawm cov lus, TensorFlow muaj ntau yam cuab yeej uas txuas nrog nws lossis ua rau saum nws.
zoo
- Nws yog neeg siv-phooj ywg. Yog tias koj paub nrog Python, nws yuav yooj yim los khaws.
- Kev txhawb nqa los ntawm zej zog. TensorFlow tau txhim kho qhov ua tau zoo txhua hnub los ntawm Google thiab lwm lub koom haum cov kws tshaj lij tsim tawm.
- TensorFlow Lite tuaj yeem siv los ua TensorFlow qauv ntawm cov khoom siv txawb.
- Tensorboard yog ib qho cuab yeej rau kev saib xyuas thiab pom cov ntaub ntawv. Yog tias koj xav saib koj cov qauv kev kawm tob hauv kev nqis tes ua, qhov no yog ib qho cuab yeej zoo siv.
- Tensorflow.js tso cai rau koj siv JavaScript los khiav cov qauv kev kawm tob hauv lub sijhawm.
tsis zoo
- TensorFlow muaj cov qauv tshwj xeeb, ua rau nws nyuaj rau nrhiav pom thiab kho qhov tsis raug.
- Tsis muaj kev txhawb nqa OpenCL.
- TensorFlow tsis muab ntau lub peev xwm rau cov neeg siv ntawm Windows operating system. Nws qhib ntau lub peev xwm rau cov neeg siv Linux. Txawm li cas los xij, cov neeg siv Windows tseem tuaj yeem rub tawm TensorFlow siv anaconda prompt lossis pip pob.
- TensorFlow poob qab nyob rau hauv cov nqe lus ntawm muab cov cim voj voog rau kev ua tsis tiav. Nws muaj kev siv tshwj xeeb rau cov kab ke tshwj xeeb, ua rau nws siv tau. Raws li qhov tshwm sim, nws raug xa mus suav tias yog API qib qis.
2. Kev
Keras yog lub tsev qiv ntawv Python-raws li kev kawm sib sib zog nqus, uas txawv nws los ntawm lwm yam kev kawm tob.
Nws yog ib hom lus programming siab uas txhais tau tias a neural network API txhais. Nws tuaj yeem siv tau ob qho tib si raws li tus neeg siv interface thiab txhawm rau txhim kho lub peev xwm ntawm cov txheej txheem kev kawm tob uas nws khiav.
Nws yog lub moj khaum minimalist uas yog lub teeb yuag thiab yooj yim siv. Vim li no, Keras yog ib feem ntawm TensorFlow's core API. Lub Keras pem hauv ntej kawg tso cai rau kev nrawm nrawm ntawm cov qauv neural network hauv kev tshawb fawb.
API yog ncaj nraim rau kev nkag siab thiab siv, nrog rau cov nyiaj ntxiv ntawm kev tso cai rau cov qauv kom yooj yim pauv ntawm lub moj khaum.
zoo
- Keras API yooj yim siv. API yog tsim tau zoo, cov khoom taw qhia, thiab hloov tau, ua rau muaj kev lom zem ntau dua rau cov neeg siv.
- Kev them nyiaj yug rau kev cob qhia faib thiab multi-GPU parallelism yog built-in.
- Keras yog Python haiv neeg module uas muab kev nkag tau yooj yim rau qhov ua tiav Python cov ntaub ntawv tshawb fawb ib puag ncig. Keras qauv, piv txwv li, tuaj yeem siv tau siv Python scikit-kawm API.
- Keras suav nrog cov luj uas tau kawm ua ntej rau ntau tus qauv kev kawm tob. Peb tuaj yeem siv cov qauv no ncaj qha los ua kev kwv yees lossis rho tawm cov yam ntxwv.
tsis zoo
- Nws tuaj yeem ua rau muaj kev ntxhov siab heev kom tau txais cov teeb meem qis backend tsis tu ncua. Cov teeb meem no tshwm sim thaum peb sim ua cov hauj lwm uas Keras tsis tau ua kom tiav.
- Thaum piv rau nws cov backends, nws yuav qeeb qeeb ntawm GPUs thiab siv sijhawm ntev dua los suav. Raws li qhov tshwm sim, peb yuav tsum muaj kev cuam tshuam nrawm rau kev siv tus phooj ywg.
- Thaum piv rau lwm cov pob khoom xws li sci-kit-kawm, Keras cov ntaub ntawv-preprocessing muaj peev xwm tsis txaus siab.
3. Apache MX Net
Lwm qhov tseem ceeb Deep Learning moj khaum yog MXNet. MXNet, uas tau tsim los ntawm Apache Software Foundation, txhawb ntau hom lus, suav nrog JavaScript, Python, thiab C ++.
Amazon Web Services kuj txhawb MXNet hauv kev txhim kho cov qauv kev kawm tob. Nws yog qhov loj heev, tso cai rau kev cob qhia tus qauv nrawm, thiab nws tau sib xws nrog ntau hom lus hauv computer.
Txhawm rau txhim kho qhov nrawm thiab tsim tau, MXNet tso cai rau koj los muab cov lus cim thiab qhov tseem ceeb ntawm kev ua haujlwm. Nws yog raws li lub sijhawm muaj kev vam meej dynamic uas sib piv cov cim thiab cov haujlwm tseem ceeb hauv lub sijhawm.
Nyob rau sab saum toj ntawm qhov ntawd, txheej txheej graph optimization ua rau cov cim ua tiav sai thiab nco kev lag luam. MXNet yog lub tsev qiv ntawv nqa tau yooj yim thiab hnav.
Nws yog powered los ntawm NVIDIA PascalTM GPUs thiab scalable tshaj ob peb GPUs thiab nodes, tso cai rau koj mus cob qhia cov qauv sai dua.
zoo
- Txhawb GPUs thiab muaj ntau hom GPU.
- Efficiency, scalable, thiab xob laim-ceev.
- Txhua lub platform loj yog nyob ntawm lub nkoj.
- Kev pabcuam qauv yog yooj yim, thiab API yog ceev.
- Scala, R, Python, C ++, thiab JavaScript yog cov lus programming txhawb.
tsis zoo
- MXNet muaj qhov me me Qhib qhov chaw zej zog tshaj TensorFlow.
- Kev txhim kho, kho kab laum, thiab lwm yam kev txhim kho yuav siv sijhawm ntev dua los siv vim tsis muaj kev txhawb nqa hauv zej zog tseem ceeb.
- MxNet, txawm hais tias dav siv los ntawm ntau lub tuam txhab hauv kev lag luam IT, tsis zoo li Tensorflow.
4. Microsoft CNTK
Microsoft Cognitive Toolkit (CNTK) yog ib qho kev lag luam uas muaj peev xwm qhib tau lub hauv paus rau kev faib kev kawm tob. Nws yog feem ntau siv los tsim neural networks, tab sis kuj tseem siv tau rau kev kawm tshuab thiab kev txawj ntse.
Nws txhawb ntau hom lus thiab yooj yim siv rau ntawm huab. Vim tias cov khoom zoo li no, CNTK yog qhov haum rau ntau yam kev siv AI. Txawm hais tias peb tuaj yeem siv C ++ los hu nws cov haujlwm, qhov kev xaiv nquag tshaj plaws yog siv Python program.
Thaum khiav ntawm ntau lub khoos phis tawj, Microsoft Cognitive Toolkit tau lees paub los muab kev ua tau zoo dua thiab muaj peev xwm ua tau zoo dua li cov cuab yeej xws li Theano lossis TensorFlow.
Microsoft Cognitive Toolkit txhawb nqa ob qho tib si RNN thiab CNN neural qauv, ua rau nws haum rau cov duab, kev sau ntawv, thiab kev paub txog kev hais lus.
zoo
- Yooj yim rau kev koom ua ke nrog Apache Spark, cov ntaub ntawv txheeb xyuas lub cav.
- CNTK qhov scalability tau ua rau nws muaj kev xaiv nrov hauv ntau lub lag luam. Muaj ntau yam optimized Cheebtsam.
- Muab kev ruaj khov thiab kev ua haujlwm zoo.
- Ua haujlwm zoo nrog Azure Cloud, ob qho tib si tau txhawb nqa los ntawm Microsoft.
- Kev siv cov peev txheej thiab kev tswj xyuas tau zoo.
tsis zoo
- Hauv kev sib piv rau Tensorflow, muaj kev txhawb nqa hauv zej zog tsawg dua.
- Txoj kev kawm tob tob.
- Nws tsis muaj lub rooj tsav xwm pom pom zoo li kev txhawb nqa ARM.
5. DeepLearning4j
Yog tias Java yog koj hom lus programming, DeepLearning4j yog lub hauv paus zoo siv. Nws yog ib lub tsev qiv ntawv kev kawm sib sib zog nqus uas yog qib kev lag luam thiab qhib qhov chaw.
Txhua yam tseem ceeb ntawm neural network tsim, xws li RNNs thiab CNNs, tau txais kev txhawb nqa. Deeplearning4j yog Java thiab Scala lub tsev qiv ntawv rau kev kawm tob.
Nws ua haujlwm zoo nrog Hadoop thiab Apache Spark ib yam. Deeplearning4j yog lwm txoj hauv kev zoo rau Java-based kev kawm sib sib zog nqus kev daws teeb meem vim nws kuj txhawb GPUs.
Thaum nws los txog rau Eclipse Deeplearning4j txoj kev kawm sib sib zog nqus, qee qhov ntawm cov yam ntxwv tseem ceeb suav nrog kev cob qhia tib yam los ntawm kev txo qis, micro-service architecture adaptation, thiab faib CPUs thiab GPUs.
zoo
- Nws muaj cov ntaub ntawv zoo heev thiab kev pab hauv zej zog.
- Apache Spark kev koom ua ke yog qhov yooj yim.
- Nws yog scalable thiab muaj peev xwm tuav cov ntaub ntawv loj loj.
tsis zoo
- Hauv kev sib piv rau Tensorflow thiab PyTorch, nws tsis tshua nyiam.
- Java yog ib hom lus programming nkaus xwb.
xaus
Xaiv qhov zoo tshaj plaws kev kawm tob yog ib qho nyuaj ua. Ntau vim tias muaj ntau ntawm lawv, cov npe tau loj hlob raws li qhov xav tau artificial txawj ntse kev tshawb fawb thiab kev siv tshuab kawm loj hlob. Txhua lub moj khaum muaj nws tus kheej txheej ntawm pros thiab downsides.
Ntau qhov kev txiav txim siab yuav tsum tau ua, suav nrog kev ruaj ntseg, scalability, thiab kev ua haujlwm. Nyob rau hauv kev lag luam-qib systems, kev cia siab ua tseem ceeb dua.
Yog tias koj nyuam qhuav pib tawm, Tensorflow yog qhov chaw zoo pib. Xaiv CNTK yog tias koj tab tom tsim cov khoom lag luam hauv Windows. Yog tias koj nyiam Java, siv DL4J.
Sau ntawv cia Ncua