2017年度30大最驚艷的開源機(jī)器學(xué)習(xí)項(xiàng)目在過去的一年里，MybridgeAI比較了近8,800個(gè)開源機(jī)器學(xué)習(xí)項(xiàng)目，選擇了前30名（概率只有0.3％）。這是一個(gè)競(jìng)爭(zhēng)激烈的名單，精挑細(xì)選了2017年1月到12月之間發(fā)布的最佳開源機(jī)器學(xué)習(xí)庫(kù)、數(shù)據(jù)集和應(yīng)用程序。MybridgeAI通過考量受歡迎程度、參與度和新近度來等指標(biāo)來評(píng)估這些參選項(xiàng)目的質(zhì)量。這些項(xiàng)目在Github得星的數(shù)量平均為3,558個(gè)，其質(zhì)量相當(dāng)不錯(cuò)。開源項(xiàng)目對(duì)數(shù)據(jù)科學(xué)家來說可能大有裨益。通過閱讀源代碼并在現(xiàn)有項(xiàng)目之上構(gòu)建一些東西來可以加快學(xué)習(xí)步伐。是該拿出時(shí)間來玩玩過去一年中可能錯(cuò)過的機(jī)器學(xué)習(xí)項(xiàng)目啦！推薦閱讀文章在看排名清單之前，建議首先看看下面2篇文章：A）神經(jīng)網(wǎng)絡(luò)深度學(xué)習(xí)A-Z?：動(dòng)手的人工神經(jīng)網(wǎng)絡(luò)（DeepLearningA-Z?:Hands-OnArtificialNeuralNetworks）http://bit.ly/2CH1WcQ[68,745條推薦，4.5/5星]B)TensorFlow用Python進(jìn)行深度學(xué)習(xí)的TensorFlow完整指南(CompleteGuidetoTensorFlowforDeepLearningwithPython)http://bit.ly/2EatVy7[17,834條推薦，4.6/5星]開源項(xiàng)目排行榜第1名FastText：用于快速文本表示和分類的庫(kù)(FastText:Libraryforfasttextrepresentationandclassification)。https://goo.gl/VWGfCs[Github上的11786顆星]由FacebookResearch提供…[Muse:基于快速文本的多語言無監(jiān)督或監(jiān)督詞嵌入(Muse:MultilingualUnsupervisedorSupervisedwordEmbeddings,basedonFastText)。Github上的695顆星]https://goo.gl/stbhjZ第2名深色照片風(fēng)格轉(zhuǎn)換：論文“深度照片風(fēng)格轉(zhuǎn)移”代碼和數(shù)據(jù)(Deep-photo-styletransfer:Codeanddataforpaper“DeepPhotoStyleTransfer ”)/luanfujun/deep-photo-styletransfer?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上的9747顆星星]由康奈爾大學(xué)FujunLuan博士提供第3名Python和世界上最簡(jiǎn)單的面部識(shí)別api和命令行(Theworld'ssimplestfacialrecognitionapiforPythonandthecommandline)/ageitgey/face_recognition?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上8672個(gè)星]由AdamGeitgey提供第4名洋紅(Magenta:)：機(jī)器智能生成音樂和藝術(shù)(Magenta:MusicandArtGenerationwithMachineIntelligence )。/tensorflow/magenta?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[GitHub上8113顆星]第5名Sonnet：基于TensorFlow的神經(jīng)網(wǎng)絡(luò)庫(kù)(Sonnet:TensorFlow-basedneuralnetworklibrary )/deepmind/sonnet?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上5731顆星]。由Deepmind公司的MalcolmReynolds(/@malcolm.reynolds)提供第6名deeplearn.js：為Web提供硬件加速的機(jī)器智能庫(kù)(/PAIR-code/deeplearnjs?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上5462顆星]。由GoogleBrain的NikhilThorat提供第7名TensorFlow的快速風(fēng)格遷移(FastStyleTransferinTensorFlow)/lengstrom/fast-style-transfer?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上4843顆星]由MIT的LoganEngstrom提供第8名Pysc2：星際爭(zhēng)霸II學(xué)習(xí)環(huán)境(Pysc2:StarCraftIILearningEnvironment)/deepmind/pysc2?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上3683顆星星]由DeepMind的TimoEwalds提供第9名AirSim：基于UnrealEngine的開源模擬器，用于微軟AI＆Research自主車輛(AirSim:OpensourcesimulatorbasedonUnrealEngineforautonomousvehiclesfromMicrosoftAI&Research)/Microsoft/AirSim?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上3861顆星]。由Microsoft的ShitalShah提供第10名Facets:機(jī)器學(xué)習(xí)數(shù)據(jù)集的可視化(Facets:Visualizationsformachinelearningdatasets)/PAIR-code/facets?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上3371顆星]由GoogleBrain提供第11名Style2Paints：圖像的AI彩色化(Style2Paints:AIcolorizationofimages)/lllyasviel/style2paints?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上3310顆星]第12名Tensor2Tensor：GoogleResearch的廣義序列到序列模型庫(kù)(Tensor2Tensor:Alibraryforgeneralizedsequencetosequencemodels?—?GoogleResearch)/tensorflow/tensor2tensor?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上3087顆星]由GoogleBrain的RyanSepassi提供第13名PyTorch中的圖像到圖像轉(zhuǎn)換(例如horse2zebra，edges2cats等)[Image-to-imagetranslationinPyTorch(e.g.horse2zebra,edges2cats,andmore)]/junyanz/pytorch-CycleGAN-and-pix2pix?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上2847顆星]由伯克利的Jun-YanZhu博士提供第14名Faiss：用于密集向量的高效相似性搜索和聚類的庫(kù)(Faiss:Alibraryforefficientsimilaritysearchandclusteringofdensevectors)/facebookresearch/faiss?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上2629顆星]。由FacebookResearch提供第15名Fashion-mnist：類似MNIST的時(shí)尚產(chǎn)品數(shù)據(jù)庫(kù)(Fashion-mnist:AMNIST-likefashionproductdatabase/zalandoresearch/fashion-mnist?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上2780顆星星]由ZalandoTech的研究科學(xué)家HanXiao提供第16名ParlAI：在各種公開可用的對(duì)話數(shù)據(jù)集上訓(xùn)練和評(píng)估AI模型的框架(/facebookresearch/ParlAI?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more)[Github上2578顆星]。有Facebook研究院的AlexanderMiller提供第17名Fairseq:FacebookAl研究序列-序列工具包(Fairseq:FacebookAIResearchSequence-to-SequenceToolkit)/facebookresearch/fairseq?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上2571顆星]第18名Pyro：Python和PyTorch的深度通用概率編程(Pyro:DeepuniversalprobabilisticprogrammingwithPythonandPyTorch)/uber/pyro?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上的2387顆星]由UberEngineering提供第19名iGAN：由GAN支持的交互式圖像生成(iGAN:InteractiveImageGenerationpoweredbyGAN)/junyanz/iGAN?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上2369顆星]::__IHACKLOG_REMOTE_IMAGE_AUTODOWN_BLOCK__::21第20名Deep-image-prior：使用無需學(xué)習(xí)的神經(jīng)網(wǎng)絡(luò)進(jìn)行圖像恢復(fù)(Deep-image-prior:Imagerestorationwithneuralnetworksbutwithoutlearning)/DmitryUlyanov/deep-image-prior?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上2188顆星]由Skoltech博士DmitryUlyanov提供第21名Face_classification：使用具有kerasCNN模型和openCV的fer2013/imdb數(shù)據(jù)集的實(shí)時(shí)人臉檢測(cè)和情感/性別分類(Face_classification:Real-timefacedetectionandemotion/genderclassificationusingfer2013/imdbdatasetswithakerasCNNmodelandopenCV)/oarriaga/face_classification?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上1967星]第22名Speech-to-Text-WaveNet：使用DeepMind的WaveNet和tensorflow進(jìn)行端到端的句子級(jí)英語語音識(shí)別(Speech-to-Text-WaveNet:End-to-endsentencelevelEnglishspeechrecognitionusingDeepMind 'sWaveNetandtensorflow)/buriburisuri/speech-to-text-wavenet?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上1961星]由KakaoBrain的NamjuKim提供第23名StarGAN：用于多域圖像到圖像轉(zhuǎn)換的統(tǒng)一生成對(duì)抗網(wǎng)絡(luò)(StarGAN:UnifiedGenerativeAdversarialNetworksforMulti-DomainImage-to-ImageTranslation)/yunjey/StarGAN?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上1954顆星]由韓國(guó)大學(xué)的YunjeyChoi提供第24名Ml-agents：Unity機(jī)器學(xué)習(xí)代理(Ml-agents:UnityMachineLearningAgents)[Github上的1658顆星]DeepLearningatUnity3D的UnityJuliani提供第25名DeepVideoAnalytics：分布式可視化搜索和可視化數(shù)據(jù)分析平臺(tái)(DeepVideoAnalytics:Adistributedvisualsearchandvisualdataanalyticsplatform)/AKSHAYUBHAT/DeepVideoAnalytics/?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上1494顆星]由康奈爾大學(xué)的AkshayBhat博士提供第26名OpenNMT：火炬中的開源神經(jīng)機(jī)器翻譯(OpenNMT:Open-SourceNeuralMachineTranslationinTorch)/OpenNMT/OpenNMT?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上1490顆星]第27名Pix2pixHD:使用條件GAN合成和處理2048X1024的圖像(Pix2pixHD:Synthesizingandmanipulating2048 X1024imageswithconditionalGANs)/NVIDIA/pix2pixHD?utm_source=mybridge&utm_medium=blog&utm_campaign=read_more[Github上1283顆星]由Nvidia的AIResearch科學(xué)家Ming-YuLiu提供第28名Horovod：TensorFlow的分布式訓(xùn)練框架(Horovod:DistributedtrainingframeworkforTensorFlow)/uber/ho