{"id":3821,"date":"2020-09-30T10:12:08","date_gmt":"2020-09-30T01:12:08","guid":{"rendered":"https:\/\/www.kagoya.jp\/howto\/?p=3821"},"modified":"2023-07-18T17:47:04","modified_gmt":"2023-07-18T08:47:04","slug":"gpu-container3","status":"publish","type":"post","link":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/","title":{"rendered":"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011"},"content":{"rendered":"
\u3010\u9023\u8f09\u4f01\u753b\u3011GPU\u30b3\u30f3\u30c6\u30ca\u6d3b\u7528 \u3010\u51686\u56de\u3011<\/th> <\/tr> <\/thead>
GPU\u30b3\u30f3\u30c6\u30ca\u3068\u306f\u4f55\u304b\uff1f\u4f55\u304c\u4fbf\u5229\u306a\u306e\u304b\uff1f\u3010\u7b2c1\u56de\uff1aGPU\u30b3\u30f3\u30c6\u30ca\u3067\u901f\u653b\u74b0\u5883\u69cb\u7bc9\u3011<\/a> <\/td> <\/tr>
TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2460\u3010\u7b2c2\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u6e96\u5099\u7de8\u301c\u3011<\/a> <\/td> <\/tr>
TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011 <\/td> <\/tr>
Chainer\u3092\u4f7f\u3063\u305f\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3010\u7b2c4\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u6a5f\u68b0\u5b66\u7fd2\u3059\u308b\u3011<\/a> <\/td> <\/tr>
PyTorch\u3067\u6a5f\u68b0\u5b66\u7fd2\u3010\u7b2c5\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u30c6\u30f3\u30bd\u30eb\u306e\u57fa\u672c\u3092\u7406\u89e3\u3059\u308b\u3011<\/a> <\/td> <\/tr> <\/tbody> <\/table><\/div>
\n
\"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\"<\/figure>\n<\/div>\n\n\n

AI\u521d\u5b66\u8005\u306b\u3068\u3063\u3066\u30cf\u30fc\u30c9\u30eb\u306e\u3072\u3068\u3064\u306b\u306a\u3063\u3066\u3044\u308bGPU\u74b0\u5883\u69cb\u7bc9\u3092\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306b\u884c\u3046\u305f\u3081\u306b\u3001\u30b3\u30f3\u30c6\u30ca\u3092\u6709\u52b9\u6d3b\u7528\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u306b\u9023\u8f09\u3092\u30b9\u30bf\u30fc\u30c8\u3057\u307e\u3057\u305f\u3002\u524d\u56de\u306f\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3068\u3057\u3066\u30e1\u30b8\u30e3\u30fc\u306aTensorFlow\u3092Dokcer\u30b3\u30f3\u30c6\u30ca\u306e\u4e2d\u304b\u3089\u8d77\u52d5\u3055\u305b\u308b\u3068\u3053\u308d\u307e\u3067\u884c\u3044\u307e\u3057\u305f\u3002\u4eca\u56de\u306fTensorFlow\u3068\u3044\u3063\u3057\u3087\u306b\u4f7f\u308f\u308c\u308b\u3053\u3068\u304c\u591a\u3044\u3001Keras\u3092\u53d6\u308a\u4e0a\u3052\u307e\u3059\u3002<\/p>\n\n\n\n

\u3053\u306e\u9023\u8f09\u306b\u3064\u3044\u3066<\/h2>\n\n\n\n

\u672c\u9023\u8f09\u306f\u3001\u51686\u56de\u306e\u30b7\u30ea\u30fc\u30ba\u3092\u901a\u3057\u3066\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306b\u3001GPU\u306e\u74b0\u5883\u69cb\u7bc9\u3092\u884c\u3046\u305f\u3081\u306b\u30b3\u30f3\u30c6\u30ca\u306e\u6d3b\u7528\u3092\u884c\u3063\u3066\u3044\u304d\u307e\u3059\u3002\u300c\u6a5f\u68b0\u5b66\u7fd2\u3084\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3092GPU\u3067\u5b9f\u884c\u3057\u3066\u307f\u305f\u3044\u3051\u3069\u96e3\u3057\u305d\u3046\u2026\u300d\u306a\u3069\u5c0e\u5165\u306b\u30cf\u30fc\u30c9\u30eb\u3092\u611f\u3058\u3089\u308c\u3066\u3044\u308b\u65b9\u306b\u3001\u30b3\u30f3\u30c6\u30ca\u3092\u6d3b\u7528\u3059\u308b\u3053\u3068\u3067\u3001\u74b0\u5883\u69cb\u7bc9\u306b\u8981\u3059\u308b\u5de5\u6570\u3092\u5727\u5012\u7684\u306b\u524a\u6e1b\u3057\u3001\u5373\u5ea7\u306b\u8ab2\u984c\u306b\u53d6\u308a\u7d44\u3080\u3053\u3068\u304c\u3067\u304d\u308b\u30e1\u30ea\u30c3\u30c8\u3092\u611f\u3058\u3066\u3044\u305f\u3060\u304d\u307e\u3059\u3002\u305d\u306e\u305f\u3081\u306b\u5fc5\u8981\u306a\u77e5\u8b58\u3084\u64cd\u4f5c\u65b9\u6cd5\u3092\u3001\u5f53\u793e\u306eGPU\u30b5\u30fc\u30d0\u30fc\u3092\u4f7f\u3044\u89e3\u8aac\u3057\u3066\u3044\u304d\u307e\u3059\u3002
\u9023\u8f09\u3092\u8aad\u307f\u7d42\u3048\u308b\u3053\u308d\u306b\u306f\u3001TensorFlow\u3084PyTorch\u306a\u3069\u306e\u30e1\u30b8\u30e3\u30fc\u306a\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3092\u4f7f\u3063\u305f\u6f14\u7fd2\u304c\u3067\u304d\u308b\u3088\u3046\u306b\u306a\u3063\u3066\u3044\u308b\u306f\u305a\u3067\u3059\u3002<\/p>\n\n\n

\n
\"\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3092\u4f7f\u3063\u305f\u6f14\u7fd2\"<\/figure>\n<\/div>\n\n\n

\u672c\u9023\u8f09\u306f\u3001\u3053\u3061\u3089\u306e\u624b\u9806\u3067\u9032\u3081\u3066\u3044\u307e\u3059<\/h2>\n\n\n\n

\u30fbGPU\u30b3\u30f3\u30c6\u30ca\u3068\u306f\u4f55\u304b\uff1f\u4f55\u304c\u4fbf\u5229\u306a\u306e\u304b\uff1f\uff08\u7b2c1\u56de\uff09<\/h3>\n\n\n\n

AI\u521d\u5b66\u8005\u304cGPU\u3092\u4f7f\u3063\u3066\u6a5f\u68b0\u5b66\u7fd2\u3084\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306b\u53d6\u308a\u7d44\u307f\u305f\u3044\u5834\u5408\u3001\u74b0\u5883\u69cb\u7bc9\u306b\u60f3\u50cf\u4ee5\u4e0a\u306e\u5de5\u6570\u304c\u767a\u751f\u3059\u308b\u3053\u3068\u304c\u3042\u308a\u307e\u3059\u3002\u30bb\u30c3\u30c8\u30a2\u30c3\u30d7\u4f5c\u696d\u306b\u8981\u3059\u308b\u6642\u9593\u3092\u6975\u529b\u524a\u6e1b\u3059\u308b\u305f\u3081\u306b\u30b3\u30f3\u30c6\u30ca\u6280\u8853\u3092\u9069\u7528\u3057\u3001\u30b3\u30f3\u30c6\u30ca\u5185\u304b\u3089GPU\u3092\u5229\u7528\u3059\u308b\u305f\u3081\u306e\u6e96\u5099\u3068\u624b\u9806\u306b\u3064\u3044\u3066\u7d39\u4ecb\u3057\u307e\u3059\u3002<\/p>\n\n\n\t\t

\n\t\t\t\t\t\t\t
\n\t\t\t\t\t\"GPU\u30b3\u30f3\u30c6\u30ca\u3068\u306f\u4f55\u304b\uff1f\u4f55\u304c\u4fbf\u5229\u306a\u306e\u304b\uff1f\u3010\u7b2c1\u56de\uff1aGPU\u30b3\u30f3\u30c6\u30ca\u3067\u901f\u653b\u74b0\u5883\u69cb\u7bc9\u3011\"\t\t\t\t<\/figure>\n\t\t\t\t\t\t
\n\t\t\t\t

GPU\u30b3\u30f3\u30c6\u30ca\u3068\u306f\u4f55\u304b\uff1f\u4f55\u304c\u4fbf\u5229\u306a\u306e\u304b\uff1f\u3010\u7b2c1\u56de\uff1aGPU\u30b3\u30f3\u30c6\u30ca\u3067\u901f\u653b\u74b0\u5883\u69cb\u7bc9\u3011<\/a><\/p>\n\t\t\t\t

\u9ad8\u901f\u63cf\u753b\u51e6\u7406\u304c\u5fc5\u8981\u3068\u3055\u308c\u308bCG\u3001VR\u30013DCAD\u7b49\u3067\u5229\u7528\u3055\u308c\u3066\u304d\u305fGPU\uff08Graphics Processing Unit\uff1a\u30b0\u30e9\u30d5\u30a3\u30c3\u30af\u30b9 \u30d7\u30ed\u30bb\u30c3\u30b7\u30f3\u30b0 \u30e6\u30cb\u30c3\u30c8\uff09\u3067\u3059\u304c\u3001\u8fd1\u5e74\u306fIoT\u3084AI\u306e\u9700\u8981\u304c\u5897\u3048\u308b\u306b\u3064\u308c\u3066\u79d1\u5b66\u6280\u8853\u8a08\u7b97\u3084\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306a\u3069\u81a8\u5927\u306a\u30c7\u30fc\u30bf\u3092\u6271\u3046\u5834\u9762\u3067\u306e\u5229\u7528\u304c\u5897\u3048\u3066\u3044\u307e\u3059\u3002\u3072\u3068\u3053\u3068\u3067GPU\u3068\u8a00\u3063\u3066\u3082\u3001\u30b2\u30fc\u30df\u30f3\u30b0PC\u306e\u30b0\u30e9\u30d5\u30a3\u30c3\u30af\u30dc\u30fc\u30c9\u306b\u642d\u8f09\u3055\u308c\u308b\u30d1\u30fc\u30bd\u30ca\u30eb\u30e6\u30fc\u30b9\u306eGPU\u304b\u3089\u3001…<\/p>\n\t\t\t<\/div>\n\t\t<\/div>\n\t\t\n\n\n

\u30fbTensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2460\uff08\u7b2c2\u56de\uff09<\/h3>\n\n\n\n

OSS\uff08\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u30bd\u30d5\u30c8\u30a6\u30a8\u30a2\uff09\u306e\u6a5f\u68b0\u5b66\u7fd2\u30e9\u30a4\u30d6\u30e9\u30ea\u306e\u4e2d\u304b\u3089TensorFlow\uff08\u30c6\u30f3\u30bd\u30eb\u30d5\u30ed\u30fc\uff09\u3001Keras\uff08\u30b1\u30e9\u30b9\uff09<\/mark>\u3092\u53d6\u308a\u4e0a\u3052\u3001\u3053\u308c\u3089\u304c\u7a3c\u50cd\u3059\u308b\u30b3\u30f3\u30c6\u30ca\u3092\u4f5c\u6210\u3057\u3001\u30b3\u30f3\u30c6\u30ca\u5185\u304b\u3089GPU\u3092\u6307\u5b9a\u3059\u308b\u65b9\u6cd5\u306b\u3064\u3044\u3066\u7d39\u4ecb\u3057\u307e\u3059\u3002TensorFlow\u306fKeras\u3092\u53d6\u308a\u8fbc\u3080\u5f62\u3067\u516c\u958b\u3055\u308c\u3066\u3044\u3066\u3001\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3092\u3059\u308b\u969b\u306e\u4f7f\u3044\u52dd\u624b\u306e\u826f\u3055\u304b\u3089\u3001\u591a\u304f\u306e\u30e6\u30fc\u30b6\u30fc\u306b\u5229\u7528\u3055\u308c\u3066\u3044\u307e\u3059\u3002\u3053\u306e\u56de\u3067\u306f\u30b3\u30f3\u30c6\u30ca\u5185\u306eTensorFlow\u3067GPU\u3092\u5229\u7528\u3067\u304d\u308b\u72b6\u614b\u307e\u3067\u78ba\u8a8d\u3057\u307e\u3059\u3002<\/p>\n\n\n\t\t

\n\t\t\t\t\t\t\t
\n\t\t\t\t\t\"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2460\u3010\u7b2c2\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u6e96\u5099\u7de8\u301c\u3011\"\t\t\t\t<\/figure>\n\t\t\t\t\t\t
\n\t\t\t\t

TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2460\u3010\u7b2c2\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u6e96\u5099\u7de8\u301c\u3011<\/a><\/p>\n\t\t\t\t

AI\u521d\u5b66\u8005\u306b\u3068\u3063\u3066\u30cf\u30fc\u30c9\u30eb\u306e\u3072\u3068\u3064\u306b\u306a\u3063\u3066\u3044\u308bGPU\u74b0\u5883\u69cb\u7bc9\u3092\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306b\u884c\u3046\u305f\u3081\u306b\u3001\u30b3\u30f3\u30c6\u30ca\u3092\u6709\u52b9\u6d3b\u7528\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u306b\u9023\u8f09\u3092\u30b9\u30bf\u30fc\u30c8\u3057\u307e\u3057\u305f\u3002\u4eca\u56de\u306f\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3068\u3057\u3066\u30e1\u30b8\u30e3\u30fc\u306aTensorFlow\u3092\u53d6\u308a\u4e0a\u3052\u307e\u3059\u3002 \u3053\u306e\u9023\u8f09\u306b\u3064\u3044\u3066 \u672c\u9023\u8f09\u306f\u3001\u51686\u56de\u306e\u30b7\u30ea\u30fc\u30ba\u3092\u901a\u3057\u3066\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306b\u3001GPU\u306e\u74b0\u5883\u69cb\u7bc9\u3092\u884c\u3046\u305f\u3081\u306b\u30b3\u30f3\u30c6\u30ca\u306e\u6d3b\u7528\u3092\u884c\u3063\u3066\u3044\u304d\u307e\u3059\u3002\u300c\u6a5f\u68b0\u5b66\u7fd2\u3084\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc…<\/p>\n\t\t\t<\/div>\n\t\t<\/div>\n\t\t\n\n\n

\u30fbTensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\uff08\u7b2c3\u56de\uff09\u4eca\u56de\u306e\u8a18\u4e8b<\/h3>\n\n\n\n

TensorFlow\uff08\u30c6\u30f3\u30bd\u30eb\u30d5\u30ed\u30fc\uff09\u3084Theano\uff08\u30c6\u30a2\u30ce\uff09\uff0fCNTK\uff08Cognitive Toolkit\uff09<\/mark>\u306e\u8907\u6570\u306e\u30d0\u30c3\u30af\u30a8\u30f3\u30c9\u3068\u3057\u3066\u5229\u7528\u53ef\u80fd\u306aKeras\uff08\u30b1\u30e9\u30b9\uff09<\/mark>\u3092\u53d6\u308a\u4e0a\u3052\u3001TensorFlow\u3068Keras\u3092\u4f7f\u3063\u305f\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3092\u884c\u3044\u307e\u3059\u3002\u3053\u3053\u3067\u306fAI\u521d\u5b66\u8005\u306e\u65b9\u304c\u89aa\u3057\u307f\u3084\u3059\u3044\u8ab2\u984c\u3092\u6271\u3046\u3053\u3068\u3092\u610f\u56f3\u3057\u3001TensorFlow\u306e\u516c\u5f0f\u30ac\u30a4\u30c9\u306b\u8a18\u8f09\u3055\u308c\u3066\u3044\u308b\u300c\u521d\u5fc3\u8005\u306e\u305f\u3081\u306e TensorFlow 2.0 \u5165\u9580\u300d\u306e\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3092\u53d6\u308a\u4e0a\u3052\u307e\u3057\u305f\u3002<\/p>\n\n\n\n

\u30fbChainer\u3092\u4f7f\u3063\u305f\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\uff08\u7b2c4\u56de\uff09<\/h3>\n\n\n\n

\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3068\u3057\u3066\u6709\u540d\u306aChainer\uff08\u30c1\u30a7\u30a4\u30ca\u30fc\uff09<\/mark>\u306e\u5229\u7528\u65b9\u6cd5\u3092\u7d39\u4ecb\u3057\u307e\u3059\u3002\u30b3\u30f3\u30c6\u30ca\u304b\u3089GPU\u3092\u5229\u7528\u3059\u308b\u624b\u9806\u3068\u3001GPU\u306b\u3088\u3063\u3066\u3069\u308c\u3060\u3051\u9ad8\u901f\u5316\u306b\u5bc4\u4e0e\u3067\u304d\u305f\u304b\u306b\u3064\u3044\u3066\u3001Python\u306e\u30d7\u30ed\u30b0\u30e9\u30e0\u5b9f\u884c\u7d50\u679c\u3092\u901a\u3057\u3066\u78ba\u8a8d\u3057\u307e\u3059\u3002C\u8a00\u8a9e\u306b\u6bd4\u3079\u3066\u51e6\u7406\u6642\u9593\u304c\u304b\u304b\u308b\u3068\u8a00\u308f\u308c\u3066\u3044\u308bPython\u3067\u3059\u304c\u3001\u6570\u5024\u8a08\u7b97\u3092\u52b9\u7387\u7684\u306b\u884c\u3046\u305f\u3081\u306e\u62e1\u5f35\u30e2\u30b8\u30e5\u30fc\u30eb\u3067\u3042\u308bNumPy\uff08\u30ca\u30e0\u30d1\u30a4\uff09<\/mark>\u306b\u3064\u3044\u3066\u3082\u53d6\u308a\u4e0a\u3052\u307e\u3059\u3002<\/p>\n\n\n\n

\u30fbPytorch\u3067\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\uff08\u7b2c5\u56de\uff09<\/h3>\n\n\n\n

Python\u306e\u6a5f\u68b0\u5b66\u7fd2\u7528\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3042\u308bPytorch\uff08\u30d1\u30a4\u30c8\u30fc\u30c1\uff09<\/mark>\u3092\u53d6\u308a\u4e0a\u3052\u307e\u3059\u3002PyTorch\u3067\u306fTensor\uff08\u30c6\u30f3\u30bd\u30eb\uff09<\/mark>\u3068\u3044\u3046\u578b\u3067\u884c\u5217\u3092\u8868\u73fe\u3057\u307e\u3059\u3002Tensor\u306f\u591a\u6b21\u5143\u914d\u5217\u3092\u6271\u3046\u305f\u3081\u306e\u30c7\u30fc\u30bf\u69cb\u9020\u3067\u3042\u308a\u3001GPU\u3092\u30b5\u30dd\u30fc\u30c8\u3057\u3066\u3044\u308b\u3053\u3068\u304b\u3089\u3001Pytorch\u304c\u7a3c\u50cd\u3059\u308b\u30b3\u30f3\u30c6\u30ca\u3092\u5229\u7528\u3057\u3001GPU\u306b\u3088\u308b\u9ad8\u901f\u51e6\u7406\u3092\u884c\u3046\u624b\u9806\u306b\u3064\u3044\u3066\u7d39\u4ecb\u3057\u307e\u3059\u3002<\/p>\n\n\n\n

\u30fbOpenPose\u306b\u3088\u308b\u95a2\u7bc0\u70b9\u62bd\u51fa\u30fb\u59ff\u52e2\u63a8\u5b9a\uff08\u7b2c6\u56de\uff09<\/h3>\n\n\n\n

\u30ab\u30e1\u30e9\u753b\u50cf\u306eAI\u753b\u50cf\u8a8d\u8b58\u3068\u8a00\u3048\u3070\u300c\u9854\u8a8d\u8a3c\u300d\u3092\u601d\u3044\u6d6e\u304b\u3079\u308b\u4eba\u304c\u591a\u3044\u3068\u601d\u3044\u307e\u3059\u304c\u3001\u6700\u8fd1\u306f\u4e00\u6b69\u9032\u307f\u3001\u4eba\u304c\u6620\u3063\u305f\u9759\u6b62\u753b\u3084\u52d5\u753b\u304b\u3089\u95a2\u7bc0\u70b9\u62bd\u51fa\u30fb\u59ff\u52e2\u63a8\u5b9a\u306b\u53d6\u308a\u7d44\u3080\u30b1\u30fc\u30b9\u304c\u5897\u3048\u3066\u3044\u307e\u3059\u3002\u4eba\u4f53\u3001\u9854\u3001\u624b\u8db3\u306a\u3069\u306e\u30ad\u30fc\u30dd\u30a4\u30f3\u30c8\u3092\u753b\u50cf\u304b\u3089\u691c\u51fa\u3059\u308b\u6280\u8853\u304c\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306b\u3088\u308a\u3001\u5b9f\u7528\u30ec\u30d9\u30eb\u307e\u3067\u5411\u4e0a\u3057\u3066\u3044\u308b\u304b\u3089\u3067\u3059\u3002\u3053\u306e\u56de\u3067\u306fOpenPose\uff08\u30aa\u30fc\u30d7\u30f3\u30dd\u30fc\u30ba\uff09<\/mark>\u3068\u3044\u3046\u30e9\u30a4\u30d6\u30e9\u30ea\u3092GPU\u4e0a\u3067\u52d5\u304b\u3059\u30b3\u30f3\u30c6\u30ca\u3092\u4f7f\u3044\u3001\u52d5\u753b\u30d5\u30a1\u30a4\u30eb\u306e\u95a2\u7bc0\u70b9\u62bd\u51fa\u624b\u9806\u3092\u7d39\u4ecb\u3057\u307e\u3059\u3002<\/p>\n\n\n\n

\u524d\u56de(\u7b2c2\u56de)\u306e\u632f\u308a\u8fd4\u308a<\/h2>\n\n\n\n

\u524d\u56de\u306f\u3001Docker\u3092\u4f7f\u3063\u3066GPU\u30b3\u30f3\u30c6\u30ca\u3067TensroFlow\u3092\u52d5\u304b\u3059\u3068\u3053\u308d\u307e\u3067\u884c\u3044\u307e\u3057\u305f\u3002
\u3010 \u7b2c2\u56de\u306e\u9023\u8f09\u8a18\u4e8b\u306f\u3053\u3061\u3089\u3000https:\/\/www.kagoya.jp\/howto\/cloud\/gpu-container2\/ \u3011<\/a><\/p>\n\n\n\n

\u5fa9\u7fd2\u3092\u517c\u306d\u3066\u3001GPU\u30b3\u30f3\u30c6\u30ca\u304b\u3089Python\u306e\u30d7\u30ed\u30b0\u30e9\u30e0\u3092\u52d5\u304b\u3057\u307e\u3059\u3002
\uff08\u30b3\u30f3\u30c6\u30ca\u5185\u3067bash\u30b7\u30a7\u30eb\u3092\u305f\u3061\u3042\u3052\u3001python\u3092\u8d77\u52d5\u3057\u3066\u304b\u3089\u3001TensorFlow\u3068tf.Keras\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u78ba\u8a8d\u3059\u308b\u7c21\u5358\u306a\u30b9\u30af\u30ea\u30d7\u30c8\u3092\u6d41\u3057\u307e\u3059\uff09
\u307e\u305a\u306f\u3001gpu\u30b3\u30f3\u30c6\u30ca\u3092\u8d77\u52d5\u3057\u3001\u30b7\u30a7\u30eb\u306b\u5165\u308a\u307e\u3059\u3002<\/p>\n\n\n

\n$ docker run --gpus all -it --rm tensorflow\/tensorflow:latest-gpu\n Subsystem: NVIDIA Corporation Device 11d9\n________                               _______________\n___  __\/__________________________________  ____\/__  \/________      __\n__  \/  _  _ \\_  __ \\_  ___\/  __ \\_  ___\/_  \/_   __  \/_  __ \\_ | \/| \/ \/\n_  \/   \/  __\/  \/ \/ \/(__  )\/ \/_\/ \/  \/   _  __\/   _  \/ \/ \/_\/ \/_ |\/ |\/ \/\n\/_\/    \\___\/\/_\/ \/_\/\/____\/ \\____\/\/_\/    \/_\/      \/_\/  \\____\/____\/|__\/\n\nWARNING: You are running this container as root, which can cause new files in\nmounted volumes to be created as the root user on your host machine.\nTo avoid this, run the container by specifying your user's userid:\n$ docker run -u $(id -u):$(id -g) args...\nroot@bec5aca64bf7:\/#\n<\/pre><\/div>\n\n\n
\u4ee5\u964d\u3001\u30b3\u30f3\u30c6\u30ca\u5185\u3067\u884c\u3063\u3066\u3044\u308b\u5185\u5bb9\u306b\u3064\u3044\u3066\u7d30\u304b\u304f\u898b\u3066\u3044\u304d\u305f\u3044\u306e\u3067\u3001bash\u30b7\u30a7\u30eb\u304b\u3089\u64cd\u4f5c\u3057\u3066\u3044\u307e\u3059\u3002
Python\u8d77\u52d5\u5f8c\u306btensorflow\u3092\u30a4\u30f3\u30dd\u30fc\u30c8\u3057\u3001\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u78ba\u8a8d\u3057\u307e\u3059\u3002<\/p>\n\n\n
\nroot@bec5aca64bf7:\/# python\nPython 3.6.9 (default, Apr 18 2020, 01:56:04)\n[GCC 8.4.0] on linux\nType "help", "copyright", "credits" or "license" for more information.\n>>> import tensorflow as tf\n>>> print('TensorFlow', tf.__version__)\nTensorFlow 2.2.0\n<\/pre><\/div>\n\n\n
tensorflow\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u306f2.2.0\u3067\u3042\u308b\u3053\u3068\u304c\u78ba\u8a8d\u3067\u304d\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
\u7d9a\u3044\u3066\u3001tf.keras\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u78ba\u8a8d\u3057\u307e\u3059\u3002<\/p>\n\n\n
\n>>> print('TensorFlow', tf.keras.__version__)\nTensorFlow 2.3.0-tf\n<\/pre><\/div>\n\n\n
tf.keras\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u306f2.3.0\u3067\u3042\u308b\u3053\u3068\u304c\u78ba\u8a8d\u3067\u304d\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
GPU\u306b\u5bfe\u5fdc\u3057\u3066\u3044\u308b\u30b3\u30f3\u30c6\u30ca\u5185\u3067\u30a4\u30f3\u30dd\u30fc\u30c8\u3057\u305ftensorflow\u3068tf.keras\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u78ba\u8a8d\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3057\u305f\u3002\u524d\u56de\u306e\u5fa9\u7fd2\u306f\u4ee5\u4e0a\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
TensorFlow\u3068Keras\u3092\u4f7f\u3063\u305f\u753b\u50cf\u8a8d\u8b58<\/h2>\n\n\n\n
2\u7a2e\u985e\u306eKeras\u306b\u3064\u3044\u3066<\/h3>\n\n\n\n
\u4eca\u56de\u3001\u3068\u308a\u3042\u3052\u308b keras \u306b\u306f\u300c\u30b9\u30bf\u30f3\u30c9\u30a2\u30ed\u30f3keras\u300d<\/mark>\u3068\u300cTensorFlow\u4ed8\u5c5ekeras\u300d<\/mark>\u306e2\u7a2e\u985e\u304c\u3042\u308a\u307e\u3059\u3002\u3082\u3068\u3082\u3068keras\u306fTensorFlow\u4ee5\u5916\u306b\u3082\u3001Theano\uff08\u30c6\u30a2\u30ce\uff09\u3084CNTK\uff08\u30b3\u30b0\u30cb\u30c6\u30a3\u30d6\u30c4\u30fc\u30eb\u30ad\u30c3\u30c8\uff09\u306a\u3069\u8907\u6570\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3092\u30d0\u30c3\u30af\u30a8\u30f3\u30c9\u306b\u5229\u7528\u3067\u304d\u308b\u6c4e\u7528\u6027\u306e\u3042\u308b\u3082\u306e\u3067\u3057\u305f\u304c\u3001\u3053\u306e\u6587\u7ae0\u3092\u66f8\u3044\u3066\u3044\u308b2020\u5e748\u6708\u6642\u70b9\u3067\u300cTensorFlow\u4ed8\u5c5ekeras\u300d\u3064\u307e\u308a\u3001tf.keras\u306b\u4e00\u672c\u5316\u3055\u308c\u307e\u3057\u305f\u3002<\/mark>
\u30de\u30eb\u30c1\u30d0\u30c3\u30af\u30a8\u30f3\u30c9Keras\uff08\u30b9\u30bf\u30f3\u30c9\u30a2\u30ed\u30f3keras\uff09\u3067\u4f5c\u6210\u3057\u305f\u30b9\u30af\u30ea\u30d7\u30c8\u306f\u3001\u305d\u306e\u6642\u70b9\u3067\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u4f7f\u3046\u3053\u3068\u3067\u7a3c\u50cd\u3055\u305b\u308b\u3053\u3068\u304c\u53ef\u80fd\u3067\u3059\u3002
\u305f\u3060\u3001\u3053\u306e\u9023\u8f09\u306e\u8da3\u65e8\u306f\u3001\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306bGPU\u3092\u52d5\u304b\u3059\u3053\u3068\u306b\u3042\u308b\u306e\u3067\u3001\u516c\u5f0f\u306b\u300cTensorFlow\u4ed8\u5c5ekeras\u300d\u4e00\u672c\u5316\u304c\u8868\u660e\u3055\u308c\u305f\u3053\u3068\u3092\u3046\u3051\u3001tf.keras\u3092\u5229\u7528\u3059\u308b\u3053\u3068\u306b\u3057\u307e\u3059\u3002<\/p>\n\n\n\n
tf.keras\u3068\u8a18\u8ff0\u3059\u308b\u7406\u7531<\/h3>\n\n\n\n
\u524d\u8ff0\u306e\u30b9\u30af\u30ea\u30d7\u30c8\u5185\u3067\u3001<\/p>\n\n\n
\n>>> print('TensorFlow', tf.keras.__version__)\n<\/pre><\/div>\n\n\n
\u3068\u8a18\u8ff0\u3057\u3066\u3001keras\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u78ba\u8a8d\u3057\u3066\u3044\u307e\u3059\u3002\u3053\u3053\u3092 tf.keras.__version__\u3067\u306a\u304f\u3001\u4e0b\u8a18\u306e\u3088\u3046\u306bkeras.__version__\u3068\u3057\u3066\u5b9f\u884c\u3059\u308b\u3068<\/p>\n\n\n
\n>>> print('TensorFlow', keras.__version__)\n<\/pre><\/div>\n\n
\nTraceback (most recent call last):\nFile "&lt;stdin>", line 1, in &lt;module>\nNameError: name 'keras' is not defined\n<\/pre><\/div>\n\n\n
\u3053\u306e\u3088\u3046\u306a\u30a8\u30e9\u30fc\u304c\u8868\u793a\u3055\u308c\u307e\u3059\u3002\u305d\u308c\u306f\u3001\u3053\u306e\u30b3\u30f3\u30c6\u30ca\u3067\u306f\u300c\u30b9\u30bf\u30f3\u30c9\u30a2\u30ed\u30f3keras\u300d\u3092\u30a4\u30f3\u30dd\u30fc\u30c8\u3057\u3066\u3001keras \u304c\u5229\u7528\u3067\u304d\u308b\u72b6\u614b\u306b\u306a\u3063\u3066\u3044\u306a\u3044\u304b\u3089\u3067\u3059\u3002<\/p>\n\n\n\n
\u300ctf.keras\u300d\u3092\u3069\u3046\u3084\u3063\u3066\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u305f\uff1f<\/h3>\n\n\n\n
tf.keras\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u306f\u3001pip install tensorflow\u3000\u306b\u306a\u308a\u307e\u3059\u3002
\u300ctf.keras\u300d\u306fTensorFlow\u306b\u4ed8\u5c5e\u3059\u308b\uff08\u8a00\u3044\u65b9\u3092\u5909\u3048\u308c\u3070\u540c\u68b1\u3055\u308c\u3066\u3044\u308b\uff09keras\u306a\u306e\u3067\u3001\u6700\u65b0\u306eTensorFlow\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3059\u308b\u3053\u3068\u3067keras\u3082\u5229\u7528\u3067\u304d\u308b\u3088\u3046\u306b\u306a\u3063\u3066\u3044\u307e\u3059\u3002
\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u4e2d\u304b\u3089TensorFlow\u4ed8\u5c5ekeras\u3092\u547c\u3073\u51fa\u3059\u306b\u306f\u3001\u524d\u8ff0\u306e<\/p>\n\n\n
\nprint('TensorFlow', tf.keras.__version__)\n<\/pre><\/div>\n\n\n
\u306e\u3088\u3046\u306a\u65b9\u6cd5\u3067\u53ef\u80fd\u3067\u3059\u3002
\u3061\u306a\u307f\u306b\u300c\u30b9\u30bf\u30f3\u30c9\u30a2\u30ed\u30f3keras\u300d\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u306f\u3001pip install keras\u3000\u306b\u306a\u308b\u3053\u3068\u304b\u3089\u60f3\u50cf\u3067\u304d\u308b\u3088\u3046\u306b\u300c\u30b9\u30bf\u30f3\u30c9\u30a2\u30ed\u30f3keras\u300d\u3068\u300cTensorFlow\u4ed8\u5c5ekeras\u300d\u306f\u307e\u3063\u305f\u304f\u5225\u7269\u306b\u306a\u3063\u3066\u3044\u307e\u3059\u306e\u3067\u3001\u30d7\u30ed\u30b0\u30e9\u30e0\u304b\u3089\u306e\u547c\u3073\u51fa\u3057\u65b9\u304c\u9055\u3046\u3053\u3068\u304c\u3042\u308a\u307e\u3059\u3002
\u3068\u8a00\u3063\u3066\u3082\u300ctf.keras\u300d\u3068\u300c\u30b9\u30bf\u30f3\u30c9\u30a2\u30ed\u30f3keras\u300d\u306f\u4e92\u63db\u6027\u304c\u9ad8\u3044\u305f\u3081\u3001\u30d7\u30ed\u30b0\u30e9\u30e0\u5185\u306eimport\u90e8\u3092\u66f8\u304d\u63db\u3048\u308c\u3070\u300ctf.keras\u300d\u306b\u5bfe\u5fdc\u53ef\u80fd\u3067\u3059\u3002
\u3053\u3061\u3089\u306e\u8a18\u4e8b\u3067\u306f\u300cTensorFlow\u4ed8\u5c5ekeras\u300d\u3092\u5229\u7528\u3057\u3066\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3092\u884c\u3044\u307e\u3059\u3002
\u4ee5\u964d\u306e\u6587\u7ae0\u3067keras\u3068\u66f8\u3044\u3066\u3044\u308b\u90e8\u5206\u306f\u3001tf.keras\u3092\u5229\u7528\u3057\u3066\u3044\u308b\u306e\u3060\u306a\u3068\u601d\u3063\u3066\u3001\u8aad\u307f\u9032\u3081\u3066\u304f\u3060\u3055\u3044\u3002
\u300ctf.keras\u300d\u306b\u3064\u3044\u3066\u306e\u8a73\u7d30\u306fTensorFlow\u306e\u516c\u5f0f\u30b5\u30a4\u30c8<\/mark>\u306b\u63b2\u8f09\u3055\u308c\u3066\u3044\u307e\u3059\u306e\u3067\u3001\u53c2\u7167\u3055\u308c\u308b\u3053\u3068\u3092\u304a\u85a6\u3081\u3057\u307e\u3059\u3002
\u3010 https:\/\/www.tensorflow.org\/guide\/keras<\/a> <\/a> \u3011<\/p>\n\n\n\n
MNIST(\u30a8\u30e0\u30cb\u30b9\u30c8)\u3092\u4f7f\u3063\u305f\u624b\u66f8\u304d\u6587\u5b57\u306e\u5224\u5225\u51e6\u7406<\/h3>\n\n\n\n
\u3053\u3053\u304b\u3089\u672c\u984c\u3067\u3042\u308bGPU\u30b3\u30f3\u30c6\u30ca\u304b\u3089\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0<\/strong>\u3092\u884c\u3044\u307e\u3059\u304c\u3001AI\u521d\u5b66\u8005\u306e\u65b9\u304c\u89aa\u3057\u307f\u3084\u3059\u3044\u8ab2\u984c\u3092\u6271\u3046\u3053\u3068\u3092\u610f\u56f3\u3057\u3001TensorFlow\u306e\u516c\u5f0f\u30b5\u30a4\u30c8\u306b\u8a18\u8f09\u3055\u308c\u3066\u3044\u308b\u300c\u521d\u5fc3\u8005\u306e\u305f\u3081\u306e TensorFlow 2.0 \u5165\u9580\u300d<\/mark>\u306e\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3092\u53d6\u308a\u4e0a\u3052\u307e\u3057\u305f\u3002
\u3010\u3000https:\/\/www.tensorflow.org\/tutorials\/quickstart\/beginner?hl=ja<\/a>\u3000\u3011<\/p>\n\n\n\n
\u516c\u5f0f\u30ac\u30a4\u30c9\u3067\u306f\u8aac\u660e\u3092\u7aef\u6298\u3063\u3066\u3044\u308b \u90e8\u5206\u3092\u88dc\u8db3\u3059\u308b\u304b\u305f\u3061\u3067\u9032\u3081\u3066\u3044\u304d\u305f\u3044\u3068\u601d\u3044\u307e\u3059\u3002
\u3053\u306e\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3067\u884c\u3063\u3066\u3044\u308b\u3053\u3068\u306e\u6982\u8981\u306f\u4ee5\u4e0b\u306b\u306a\u308a\u307e\u3059<\/p>\n\n\n\n
    \n
  1. \u624b\u66f8\u304d\u6587\u5b57\u753b\u50cf\u3092\u5206\u985e\u3059\u308b\u305f\u3081\u306e\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306e\u30e2\u30c7\u30eb\u3092\u5b9a\u7fa9\u3059\u308b<\/li>\n\n\n\n
  2. \u69cb\u7bc9\u3057\u305f\u30e2\u30c7\u30eb\u306b\u5bfe\u3057\u3001\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u7528\u753b\u50cf\u3092\u4f7f\u3044\u3001\u5b66\u7fd2\u3092\u884c\u3046<\/li>\n\n\n\n
  3. \u30c6\u30b9\u30c8\u7528\u753b\u50cf\u3092\u4f7f\u3044\u3001\u30e2\u30c7\u30eb\u3092\u6027\u80fd\u8a55\u4fa1\u3059\u308b<\/li>\n<\/ol>\n\n\n\n
    MNIST\u3068\u306f<\/h3>\n\n\n
    \n
    \"MNIST\"<\/figure>\n<\/div>\n\n\n
    \uff10\u304b\u3089\uff19\u306e\u624b\u66f8\u304d\u6570\u5b57\u6587\u5b57\u306e\u753b\u50cf\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3067\u3059\u3002\u6a5f\u68b0\u5b66\u7fd2\u306e\u9818\u57df\u3067\u30b5\u30f3\u30d7\u30eb\u3068\u3057\u3066\u5229\u7528\u3055\u308c\u308b\u3053\u3068\u304c\u591a\u3044\u4f7f\u3044\u52dd\u624b\u306e\u3088\u3044\u30c7\u30fc\u30bf\u96c6\u3067\u3042\u308a\u3001\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u7528\u30c7\u30fc\u30bf60,000\u679a\u3068\u30c6\u30b9\u30c8\u7528\u30c7\u30fc\u30bf10,000\u679a\u304c\u542b\u307e\u308c\u307e\u3059\u3002MNIST\u306e\u30c7\u30fc\u30bf\u306f\u3001\u624b\u66f8\u304d\u6587\u5b57\u306e\u753b\u50cf\u30c7\u30fc\u30bf\u304c\u4f55\u306e\u6570\u5b57\u304b\u3092\u793a\u3059\u300c\u30e9\u30d9\u30eb\u30c7\u30fc\u30bf\u300d\u3068\u7e26\u30fb\u6a2a \u5404\u300528\u30d4\u30af\u30bb\u30eb\u306e\u300c\u753b\u50cf\u30c7\u30fc\u30bf\u300d\u304b\u3089\u69cb\u6210\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
    TensorFlow\u516c\u5f0f\u30ac\u30a4\u30c9\u306e\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3067\u3082\u3001MNIST\u3092\u4f7f\u3063\u305f\u6f14\u7fd2\u3092\u884c\u3063\u3066\u3044\u307e\u3059\u306e\u3067\u3001\u8a73\u7d30\u3092\u307f\u3066\u3044\u304d\u307e\u3057\u3087\u3046\u3002
    \u524d\u6bb5\u3067\u884c\u3063\u3066\u3044\u308bbash\u30b7\u30a7\u30eb\u304b\u3089\u306e\u4e00\u9023\u306e\u64cd\u4f5c\u306e\u7d9a\u304d\u306b\u306a\u308a\u307e\u3059\u3002\u5148\u305a\u6700\u521d\u306bmnist\u306e\u30c7\u30fc\u30bf\u3092tf.keras\u304b\u3089\u547c\u3073\u51fa\u3057\u307e\u3059\u3002<\/p>\n\n\n
    \n>>>mnist = tf.keras.datasets.mnist\n>>>(x_train, y_train), (x_test, y_test) = mnist.load_data()\n<\/pre><\/div>\n\n\n
    2\u884c\u76ee\u3067\u306fmnist\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306b\u542b\u307e\u308c\u3066\u3044\u308b\u30c7\u30fc\u30bf\u3092\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u7528\u30c7\u30fc\u30bf\u3068\u30c6\u30b9\u30c8\u7528\u30c7\u30fc\u30bf\u306b\u5206\u5272\u3057\u3066\u3044\u307e\u3059\u3002
    x_train\u306f\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u7528\u753b\u50cf\u30c7\u30fc\u30bf\u3001y_train\u306f\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u7528\u30e9\u30d9\u30eb\u30c7\u30fc\u30bf\uff08\uff10\u304b\u3089\uff19\u306e\u3069\u306e\u6570\u5b57\u306e\u753b\u50cf\u304b\u3092\u793a\u3059\u6570\u5024\u30c7\u30fc\u30bf\uff09\u3001x_test\u306f\u30c6\u30b9\u30c8\u7528\u753b\u50cf\u30c7\u30fc\u30bf\u3001y_test\u306f\u30c6\u30b9\u30c8\u7528\u30e9\u30d9\u30eb\u30c7\u30fc\u30bf\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
    \u4e0a\u8a18\u306e\u30b9\u30af\u30ea\u30d7\u30c8\u3092\u5b9f\u884c\u3059\u308b\u3068\u4e0b\u8a18\u306e\u3088\u3046\u306b\u7d0411.5M\u306e\u30c7\u30fc\u30bf\u304c\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3055\u308c\u307e\u3057\u305f\u3002<\/p>\n\n\n
    \nDownloading data from https:\/\/storage.googleapis.com\/tensorflow\/tf-keras-datasets\/mnist.npz\n11493376\/11490434 [==============================] - 1s 0us\/step\n<\/pre><\/div>\n\n\n
    \u6b21\u306b\u30c7\u30fc\u30bf\u306e\u6b63\u898f\u5316\u51e6\u7406\u3092\u884c\u3044\u307e\u3059\u3002
    MNIST\u306e 28\u30d4\u30af\u30bb\u30eb\u00d728\u30d4\u30af\u30bb\u30eb \u306e\u624b\u66f8\u304d\u6587\u5b57\u753b\u50cf\u306f\u3001\u5404\u30d4\u30af\u30bb\u30eb\u3054\u3068\u306e\u767d\u9ed2\u8ae7\u8abf\u30920\u304b\u3089255\u306e\u7bc4\u56f2\u5185\u3067\u3042\u3089\u308f\u3059\u30c7\u30fc\u30bf\u3068\u3057\u3066\u6301\u3063\u3066\u3044\u307e\u3059\u304c\u3001\u3053\u308c\u30920\u304b\u30891\u306e\u7bc4\u56f2\u5185\u306b\u53ce\u307e\u308b\u3088\u3046\u6b63\u898f\u5316\u3057\u307e\u3059\u3002\uff08\u6574\u6570\u304b\u3089\u6d6e\u52d5\u5c0f\u6570\u70b9\u306b\u5909\u63db\u3059\u308b\u3053\u3068\u3067\u3001\u30c7\u30fc\u30bf\u3092\u6271\u3044\u3084\u3059\u304f\u3059\u308b\u305f\u3081\u3067\u3059\uff09<\/p>\n\n\n
    \n>>>x_train, x_test = x_train \/ 255.0, x_test \/ 255.0\n<\/pre><\/div>\n\n\n
    \u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u69cb\u7bc9<\/h3>\n\n\n\n
    \u7d9a\u3051\u3066\u3001\u753b\u50cf\u3092\u5206\u985e\u3059\u308b\u305f\u3081\u306e\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u69cb\u7bc9\u3057\u307e\u3059\u3002\u4eca\u56de\u3068\u308a\u3042\u3052\u305fTensorFlow\u516c\u5f0f\u30b5\u30a4\u30c8\u306e\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3067\u306f\u300cSequential\u30e2\u30c7\u30eb\u300d\u3068\u547c\u3070\u308c\u308b\u3001\u5404\u5c64\u3092\u9806\u756a\u306b\u3064\u306a\u3052\u3066\u3044\u304f\u30e2\u30c7\u30eb\u3092\u5b9a\u7fa9\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n
    \n>>>model = tf.keras.models.Sequential([\ntf.keras.layers.Flatten(input_shape=(28, 28)),\ntf.keras.layers.Dense(128, activation='relu'),\ntf.keras.layers.Dropout(0.2),\ntf.keras.layers.Dense(10)\n])\n<\/pre><\/div>\n\n\n
    2\u884c\u76ee\u306e
    tf.keras.layers.Flatten(input_shape=(28, 28))<\/strong>
    \u3067\u306f\u3001\u5165\u529b\u30c7\u30fc\u30bf\u306e\u30d5\u30a9\u30fc\u30de\u30c3\u30c8\u3092\u5909\u63db\u3057\u3066\u3044\u307e\u3059\u3002Flatten\u95a2\u6570\u306fn\u6b21\u5143\u306e\u914d\u5217\u30921\u6b21\u5143\u306e\u914d\u5217\u306b\u5909\u63db\u3059\u308b\u95a2\u6570\u3067\u3001\u624b\u66f8\u304d\u6587\u5b57\u6570\u5024\u3092\u793a\u3059\u305f\u3081\u306e 28\u30d4\u30af\u30bb\u30eb X 28\u30d4\u30af\u30bb\u30eb \u306e2\u6b21\u5143\u914d\u5217\u30c7\u30fc\u30bf\u304b\u3089 754\uff0828 X 28\uff09\u306e1\u6b21\u5143\u914d\u5217\u30c7\u30fc\u30bf\u3078\u306e\u5909\u63db\u51e6\u7406\u3092\u884c\u3063\u3066\u3044\u307e\u3059\u3002\u3053\u308c\u306b\u3088\u308a\u3001\u6700\u521d\u306e\u5c64\uff08\u5165\u529b\u5c64\uff09\u304c\u3001754\u306e\u30ce\u30fc\u30c9\u3092\u6301\u3064\u30e2\u30c7\u30eb\u3067\u3042\u308b\u3053\u3068\u304c\u5b9a\u7fa9\u3067\u304d\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
    3\u884c\u76ee\u306e
    tf.keras.layers.Dense(128, activation=’relu’)<\/strong>
    \u3067\u306f\u3001\u5165\u529b\u5c64\u306b\u3064\u306a\u304c\u3063\u3066\u3044\u308b\u6b21\u306e\u5c64\uff08\u4e2d\u9593\u5c64\uff0f\u96a0\u308c\u5c64\uff09\u306e\u5b9a\u7fa9\u3092\u884c\u3063\u3066\u3044\u307e\u3059\u3002\u3053\u306e\u5c64\u306f128\u306e\u30ce\u30fc\u30c9\u3092\u6301\u3061\u3001\u524d\u5f8c\u306e\u5c64\u306b\u3042\u308b\u3059\u3079\u3066\u306e\u30ce\u30fc\u30c9\u3068\u3064\u306a\u304c\u3063\u3066\u3044\u307e\u3059\u3002\u6d3b\u6027\u5316\u95a2\u6570\u3068\u3057\u3066ReLU\u95a2\u6570\uff08\u30e9\u30f3\u30d7\u95a2\u6570\uff09<\/strong>\u3092\u9069\u7528\u3059\u308b\u3068\u30010\u4ee5\u4e0b\u306a\u3089\u300c0\u300d\u3092\u30010\u3088\u308a\u5927\u304d\u3051\u308c\u3070\u300c\u5165\u529b\u5024\u305d\u306e\u3082\u306e\u300d\u304c\u8fd4\u3063\u3066\u304d\u307e\u3059\u306e\u3067\u3001\u30de\u30a4\u30ca\u30b9\u5024\u3092\u5207\u308a\u6368\u3066\u3001\u7279\u5fb4\u304c\u306f\u3063\u304d\u308a\u3068\u3057\u305f\u30c7\u30fc\u30bf\u3068\u3057\u3066\u6271\u3046\u3053\u3068\u304c\u53ef\u80fd\u306b\u306a\u308a\u307e\u3059\uff08\u753b\u50cf\u30c7\u30fc\u30bf\u306a\u306e\u3067\u30de\u30a4\u30ca\u30b9\u3092\u793a\u3059\u30c7\u30fc\u30bf\u306f\u610f\u5473\u306e\u7121\u3044\u30ce\u30a4\u30ba\u3068\u307f\u306a\u3057\u3066\u51e6\u7406\u3057\u307e\u3059\uff09<\/p>\n\n\n\n
    4\u884c\u76ee\u306e
    tf.keras.layers.Dropout(0.2)<\/strong>
    \u3067\u306f\u3001\u30c9\u30ed\u30c3\u30d7\u30a2\u30a6\u30c8\u7387\u30920.2\u3068\u8a2d\u5b9a\u3057\u3066\u3044\u307e\u3059\u3002\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3067\u904e\u5b66\u7fd2\uff08\u30aa\u30fc\u30d0\u30fc\u30d5\u30a3\u30c3\u30c6\u30a3\u30f3\u30b0\uff09\u3092\u9632\u3050\u305f\u3081\u306e\u624b\u6cd5\u306e\u3072\u3068\u3064\u3068\u3057\u3066\u30c9\u30ed\u30c3\u30d7\u30a2\u30a6\u30c8\u3092\u8ffd\u52a0\u3059\u308b\u3053\u3068\u304c\u3042\u308a\u3001tf.keras<\/strong>\u3067\u306f\u76f4\u524d\u306e\u5c64\u306b\u5bfe\u3057\u3066\u3001\u30c9\u30ed\u30c3\u30d7\u30a2\u30a6\u30c8\u7387\u3092\u8a2d\u5b9a\u3059\u308b\u3053\u3068\u3067\u3001\u30e2\u30c7\u30eb\u306e\u7cbe\u5ea6\u3092\u5411\u4e0a\u3055\u305b\u308b\u3053\u3068\u304c\u671f\u5f85\u3067\u304d\u307e\u3059\u3002\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u7528\u30c7\u30fc\u30bf\u3067\u3044\u304f\u3089\u7cbe\u5ea6\u304c\u9ad8\u3044\u30e2\u30c7\u30eb\u304c\u69cb\u7bc9\u3067\u304d\u305f\u3068\u3057\u3066\u3082\u3001\u672a\u77e5\u306e\u30c6\u30b9\u30c8\u30c7\u30fc\u30bf\u3092\u4f7f\u3063\u305f\u3068\u304d\u306b\u7cbe\u5ea6\u304c\u51fa\u306a\u3044\uff08\u6b63\u89e3\u7387\u304c\u4f4e\u3044\uff09\u30e2\u30c7\u30eb\u3067\u306f\u5b9f\u7528\u6027\u304c\u306a\u3044\u305f\u3081\u3001\u6562\u3048\u306620\uff05\u306e\u30ce\u30fc\u30c9\u3092\u7121\u52b9\u306b\u3057\u3066\u904e\u5b66\u7fd2\u3092\u9632\u3054\u3046\u3068\u3044\u3046\u610f\u56f3\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
    5\u884c\u76ee\u306e
    tf.keras.layers.Dense(10)<\/strong>
    \u3067\u306f\u3001\u6700\u5f8c\u306e\u5c64\uff08\u3053\u306e\u30e2\u30c7\u30eb\u306e\u51fa\u529b\u5c64\uff09\u3068\u3057\u306610\u500b\u306e\u30ce\u30fc\u30c9\u3092\u3082\u3064\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3068\u3057\u3066\u5b9a\u7fa9\u3057\u3066\u3044\u307e\u3059\u3002\uff08\u753b\u50cf\u30c7\u30fc\u30bf\u304c0\u304b\u30899\u306e\u3069\u306e\u6570\u5024\u3067\u3042\u308b\u3068\u5224\u5b9a\u3059\u308b\u304b\u3092\u51fa\u529b\u3059\u308b\u305f\u3081\u306b10\u500b\u306e\u30ce\u30fc\u30c9\u3092\u5229\u7528\u3059\u308b\u305f\u3081\u3067\u3059\uff09<\/p>\n\n\n\n
    \u3053\u306e\u3042\u305f\u308a\u306e\u8a73\u7d30\u3092\u77e5\u308a\u305f\u3044\u65b9\u306fTensorFlow\u516c\u5f0f\u30ac\u30a4\u30c9\u5185\u306e
    \u30e2\u30c7\u30eb\u306e\u69cb\u7bc9\u3000\uff1e\u3000\u5c64\u306e\u8a2d\u5b9a\u3000\u3068\u3044\u3046\u9805\u76ee\u3092\u53c2\u7167\u304f\u3060\u3055\u3044\u3002
    https:\/\/www.tensorflow.org\/tutorials\/keras\/classification?hl=ja<\/a>
    \u524d\u8ff0\u306e\u30b9\u30af\u30ea\u30d7\u30c8\u3092\u6d41\u3059\u3068\u4e0b\u8a18\u306e\u3088\u3046\u306b\u5b9f\u884c\u3055\u308c\u307e\u3057\u305f\u3002<\/p>\n\n\n
    \n2020-09-04 04:42:18.824202: I tensorflow\/stream_executor\/platform\/default\/dso_loader.cc:44]\n Successfully opened dynamic library libcuda.so.1\n\uff08\u4e2d\u7565\uff09\n2020-09-04 04:42:19.015061: I tensorflow\/core\/common_runtime\/gpu\/gpu_device.cc:1247]\n Created TensorFlow device (\/job:localhost\/replica:0\/task:0\/device:GPU:0 with 851 MB memory) -> \n physical GPU (device: 0, name: Tesla P40, pci bus id: 0000:03:00.0, compute capability: 6.1)\n>>>\n<\/pre><\/div>\n\n\n
    \u4e0a\u8a18\u306e\u30e2\u30c7\u30eb\u3067\u4e88\u6e2c\u3059\u308b<\/h3>\n\n\n\n
    \u524d\u6bb5\u3067\u5b9a\u7fa9\u3057\u305f\u30e2\u30c7\u30eb\u3092\u4f7f\u3063\u3066\u4e88\u6e2c\u3092\u884c\u3044\u307e\u3059\u3002\u3053\u306e\u6642\u70b9\u3067\u306f\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u524d\u3067\u3059\u3002<\/p>\n\n\n\n
    \u516c\u5f0f\u30ac\u30a4\u30c9\u306b\u306f\u300c\u30e2\u30c7\u30eb\u306f\u305d\u308c\u305e\u308c\u306e\u6a19\u672c\u306b\u3064\u3044\u3066\u3001\u30af\u30e9\u30b9\u3054\u3068\u306b”\u30ed\u30b8\u30c3\u30c8”\u3084”\u5bfe\u6570\u30aa\u30c3\u30ba\u6bd4”\u3068\u547c\u3070\u308c\u308b\u30b9\u30b3\u30a2\u3092\u7b97\u51fa\u3057\u307e\u3059\u300d\u3068\u8a18\u8f09\u3055\u308c\u3066\u3044\u307e\u3059\u304c\u3001\u30ed\u30b8\u30c3\u30c8\uff08logits\uff09\u3068\u306f\u3042\u308b\u4e8b\u8c61\u304c\u767a\u751f\u3059\u308b\u78ba\u7387p\u3092\u767a\u751f\u3057\u306a\u3044\u78ba\u7387(1-p)\u3067\u5272\u3063\u305f\u30aa\u30c3\u30ba\u306e\u81ea\u7136\u5bfe\u6570\u3067\u3059\u3002<\/p>\n\n\n\n
    \u6b21\u306e\u30b9\u30af\u30ea\u30d7\u30c8\u3092\u5b9f\u884c\u3059\u308b\u3068\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u7d50\u679c\u304c\u8fd4\u3063\u3066\u304d\u307e\u3057\u305f\u3002<\/p>\n\n\n
    \n>>>predictions = model(x_train[:1]).numpy()\n2020-09-04 04:51:53.991893:\nI tensorflow\/stream_executor\/platform\/default\/dso_loader.cc:44]\nSuccessfully opened dynamic library libcublas.so.10\n>>>\n<\/pre><\/div>\n\n
    \n>>>predictions\n  array([[-0.0729934 , -0.5482341 , 0.3433068 , -0.15514854, 0.44248402,\n  \u3000\u3000\u3000\u3000\u3000 0.24744856, -0.15032545, 0.24870253, 0.16561234, -0.56604826]],\n  \u3000\u3000\u3000\u3000\u3000 dtype=float32)\n<\/pre><\/div>\n\n\n
    \u6b21\u306e\u30b9\u30af\u30ea\u30d7\u30c8\u3067\u306f\u6d3b\u6027\u5316\u95a2\u6570\u3068\u3057\u3066\u30bd\u30d5\u30c8\u30de\u30c3\u30af\u30b9\u95a2\u6570\u3092\u9069\u7528\u3057\u3066\u3044\u307e\u3059\u304c\u3001\u524d\u6bb5\u3067\u7b97\u51fa\u3057\u305f\u30ed\u30b8\u30c3\u30c8\u306e\u5024\u3092\u624b\u66f8\u304d\u6587\u5b57\u306e0\uff5e9\u3067\u3042\u308b\u78ba\u7387\u304c\u4f55\u30d1\u30fc\u30bb\u30f3\u30c8\u306a\u306e\u304b\u3092\u793a\u3059\u305f\u3081\u306b\u5229\u7528\u3057\u3066\u3044\u307e\u3059\u3002\uff08\u30bd\u30d5\u30c8\u30de\u30c3\u30af\u30b9\u95a2\u6570\u306f\u51fa\u529b\u5024\u3092\u78ba\u7387\u3068\u3057\u3066\u8868\u73fe\u3067\u304d\u308b\u6027\u8cea\u304c\u3042\u308b\u305f\u3081\u5229\u7528\u3057\u3066\u3044\u307e\u3059\uff09<\/p>\n\n\n
    \n>>>tf.nn.softmax(predictions).numpy()\n<\/pre><\/div>\n\n\n
    \u7d50\u679c\u306f\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u306a\u308a\u307e\u3057\u305f\u3002<\/p>\n\n\n
    \narray([[0.15495376, 0.12319947, 0.12194902, 0.07117186, 0.05812015,\n\u3000\u3000\u3000\u3000 0.0898917 , 0.08244707, 0.04264582, 0.13558292, 0.12003825]],\n\u3000\u3000\u3000\u3000 dtype=float32)\n<\/pre><\/div>\n\n\n
    \u30e2\u30c7\u30eb\u3092\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3055\u305b\u308b\u305f\u3081\u306b\u5fc5\u8981\u306a\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u8a2d\u5b9a\u3092\u30b3\u30f3\u30d1\u30a4\u30eb\u3068\u3044\u3046\u51e6\u7406\u306e\u969b\u306b\u884c\u3044\u307e\u3059\u3002\u305d\u306e1\u3064\u306b\u640d\u5931\u95a2\u6570\u304c\u3042\u308a\u3001\u4efb\u610f\u306e\u95a2\u6570\u3092\u5229\u7528\u3067\u304d\u307e\u3059\u304c\u3001\u3053\u306e\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3067\u306f\u4ea4\u5dee\u30a8\u30f3\u30c8\u30ed\u30d4\u30fc\u3092\u5229\u7528\u3057\u3066\u3044\u307e\u3059\u3002\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u524d\u306e\u30e2\u30c7\u30eb\u3067\u306f\u30e9\u30f3\u30c0\u30e0\u306b\u8fd1\u3044\u78ba\u7387\uff08\u305d\u308c\u305e\u308c\u306e\u30af\u30e9\u30b9\u306b\u3064\u3044\u3066 1\/10\uff09 \u3092\u51fa\u529b\u3057\u307e\u3059\u3002\u6700\u521d\u306e\u640d\u5931\u306f -tf.log\uff081\/10\uff09\uff5e\uff1d 2.3 \u306b\u8fd1\u3044\u5024\u306b\u306a\u308b\u306f\u305a\u3067\u3059\u3002<\/p>\n\n\n
    \n>>>loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)\n>>>loss_fn(y_train[:1], predictions).numpy()\n2.4091496\n<\/pre><\/div>\n\n\n
    \u30e2\u30c7\u30eb\u306e\u30b3\u30f3\u30d1\u30a4\u30eb\u3092\u4e0b\u8a18\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\u3067\u8a2d\u5b9a\u3057\u3001\u3069\u3093\u306a\u5b66\u7fd2\u3092\u884c\u3046\u304b\u3092\u5b9a\u7fa9\u3057\u307e\u3059\u3002\u3072\u3068\u3064\u3081\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\u306f\u6700\u9069\u5316\u95a2\u6570\u30012\u3064\u3081\u306f\u640d\u5931\u95a2\u6570\u30013\u3064\u3081\u306f\u30e1\u30c8\u30ea\u30af\u30b9\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n
    optimizer<\/strong>\uff1a\u6700\u9069\u5316\u95a2\u6570\u3068\u3057\u3066Adam\uff08Adaptive Moment Estimation\uff09\u3092\u6307\u5b9a\u3002SGD\u3068\u6bd4\u8f03\u3059\u308b\u3068\u53ce\u675f\u304c\u901f\u3044\u3068\u8a00\u308f\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
    loss<\/strong>\uff1a\u640d\u5931\u95a2\u6570\u306f\u524d\u6bb5\u3067\u6307\u5b9a\u3057\u305f\u201d\u4ea4\u5dee\u30a8\u30f3\u30c8\u30ed\u30d4\u30fc\u201d\u3092\u9069\u7528\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
    metrics<\/strong>\uff1a\u30e1\u30c8\u30ea\u30af\u30b9\uff08\u8a55\u4fa1\u95a2\u6570\u306e\u30ea\u30b9\u30c8\uff09\u3068\u3057\u3066\u3001\u4eca\u56de\u306e\u3088\u3046\u306a\u624b\u66f8\u304d\u6587\u5b57\u306e\u5206\u985e\u304c\u30c6\u30fc\u30de\u306e\u5834\u5408\u3001\u3088\u304f\u7528\u3044\u3089\u308c\u308baccuracy\u3092\u6307\u5b9a\u3002<\/p>\n\n\n
    \n>>>model.compile(optimizer='adam',\nloss=loss_fn,\nmetrics=['accuracy'])\n<\/pre><\/div>\n\n\n
    \u30e2\u30c7\u30eb\u306e\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\uff08\u5b66\u7fd2\uff09\u3092\u884c\u3044\u307e\u3059<\/h3>\n\n\n\n
    \u624b\u9806\u3068\u3057\u3066\u306f\u3001\u3053\u308c\u307e\u3067\u306b\u5b9a\u7fa9\u3057\u305f\u5b66\u7fd2\u30e2\u30c7\u30eb\u306b\u5bfe\u3057\u3001\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u30c7\u30fc\u30bf\u3001\u6559\u5e2b\u30e9\u30d9\u30eb\u30c7\u30fc\u30bf\u3092\u4e0e\u3048\u3001\u753b\u50cf\u3068\u30e9\u30d9\u30eb\u306e\u5bfe\u5fdc\u95a2\u4fc2\u3092\u5b66\u7fd2\u3057\u307e\u3059\u3002
    \u30e2\u30c7\u30eb\u306b\u5bfe\u3057\u3001\u30c6\u30b9\u30c8\u7528\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306e\u4e88\u6e2c\u3092\u884c\u308f\u305b\u3001\u4e88\u6e2c\u7d50\u679c\u3068\u30c6\u30b9\u30c8\u7528\u6559\u5e2b\u30e9\u30d9\u30eb\u3092\u7167\u5408\u3057\u307e\u3059\u3002
    \u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3067\u306f\u30a8\u30dd\u30c3\u30af\u56de\u6570\u30925\u3068\u8a2d\u5b9a\u3057\u3066\u3044\u307e\u3059\u3002
    model.fit\u95a2\u6570\u3092\u4f7f\u3044\u3001\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3092\u958b\u59cb\u3057\u307e\u3059\u3002\u4e0b\u8a18\u306e\u3088\u3046\u306b\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u904e\u7a0b\u306e\u640d\u5931\u5024(loss)\u3068\u6b63\u89e3\u7387(accuracy)\u304c\u968f\u6642\u3001\u8868\u793a\u3055\u308c\u307e\u3059\u3002<\/p>\n\n\n
    \n>>>model.fit(x_train, y_train, epochs=5)\nEpoch 1\/5\n1875\/1875 [==============================] - 4s 2ms\/step - loss: 0.2965 - accuracy: 0.9138\nEpoch 2\/5\n1875\/1875 [==============================] - 3s 2ms\/step - loss: 0.1459 - accuracy: 0.9571\nEpoch 3\/5\n1875\/1875 [==============================] - 4s 2ms\/step - loss: 0.1082 - accuracy: 0.9674\nEpoch 4\/5\n1875\/1875 [==============================] - 4s 2ms\/step - loss: 0.0881 - accuracy: 0.9724\nEpoch 5\/5\n1875\/1875 [==============================] - 4s 2ms\/step - loss: 0.0754 - accuracy: 0.9768\n&lt;tensorflow.python.keras.callbacks.History object at 0x7f03d0b977b8>\n<\/pre><\/div>\n\n\n
    \u6b63\u89e3\u7387\u306e\u8a55\u4fa1<\/h3>\n\n\n\n
    \u5f15\u304d\u7d9a\u304d\u30e2\u30c7\u30eb\u306e\u691c\u8a3c\u3092\u884c\u3044\u307e\u3059\u30025\u56de\u306e\u30a8\u30dd\u30c3\u30af\u3067\u30e2\u30c7\u30eb\u306e\u6b63\u89e3\u7387\u304c\u793a\u3055\u308c\u307e\u3057\u305f\u304c\u3001\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3067\u5229\u7528\u3057\u306a\u304b\u3063\u305f\u30c6\u30b9\u30c8\u7528\u30c7\u30fc\u30bf\u3068\u30c6\u30b9\u30c8\u7528\u6559\u5e2b\u30e9\u30d9\u30eb\u30c7\u30fc\u30bf\u3092\u4f7f\u3044\u3001\u8aa4\u5dee\u3068\u6b63\u89e3\u7387\u3092\u6c42\u3081\u307e\u3059\u3002\uff08\u3053\u306e\u30e2\u30c7\u30eb\u306e\u8a55\u4fa1\u3092\u884c\u3044\u307e\u3059\uff09
    3\u756a\u76ee\u306e\u30d1\u30e9\u30fc\u30e1\u30fc\u30bf verbose=2 \u306f\u30d7\u30ed\u30b0\u30ec\u30b9\u30d0\u30fc\u8868\u793a\u3092\u3057\u306a\u3044\u6307\u5b9a\u3067\u3059\u3002\uff08\u30a8\u30dd\u30c3\u30af\u3054\u3068\u306e\u30ed\u30b0\u306f\u3042\u308a\uff09<\/p>\n\n\n
    \n>>>model.evaluate(x_test, y_test, verbose=2)\n313\/313 - 1s - loss: 0.0737 - accuracy: 0.9774\n[0.07369402796030045, 0.977400004863739]\n<\/pre><\/div>\n\n\n
    \u691c\u8a3c\u7d50\u679c\u306f\u4e0a\u8a18\u306e\u3088\u3046\u306b\u8aa4\u5dee\u304c\u7d04 0.07<\/strong> \u3001\u6b63\u89e3\u7387\u306f\u7d04 0.98<\/strong> \u3068\u8868\u793a\u3055\u308c\u307e\u3057\u305f\u3002
    5\u56de\u306e\u5b66\u7fd2\u3067\u9ad8\u3044\u6b63\u89e3\u7387\u304c\u51fa\u305f\u3053\u3068\u3067\u3001\u3053\u3053\u3067\u5b66\u7fd2\u306f\u7d42\u4e86\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
    \u30c6\u30b9\u30c8\u7528\u753b\u50cf\u3067\u63a8\u8ad6<\/h3>\n\n\n\n
    \u5b66\u7fd2\u304c\u7d42\u308f\u3063\u305f\u306e\u3067\u3001\u6700\u5f8c\u306b\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u6e08\u30e2\u30c7\u30eb\u3092\u4f7f\u3063\u3066\u63a8\u8ad6\u51e6\u7406\u3057\u307e\u3059\u304c\u3001\u6570\u5024\u3067\u306f\u306a\u304f\u78ba\u7387\u3092\u51fa\u529b\u3057\u305f\u3044\u5834\u5408\u306b\u306f\u524d\u6bb5\u3067\u4f7f\u3063\u305f\u30bd\u30d5\u30c8\u30de\u30c3\u30af\u30b9\u95a2\u6570\u3092\u30e2\u30c7\u30eb\u5168\u4f53\u306b\u5bfe\u3057\u3066\u6307\u5b9a\u3057\u307e\u3059\u3002\u30c6\u30b9\u30c8\u7528\u753b\u50cf\uff15\u3064\u306b\u3064\u3044\u3066\u63a8\u8ad6\u3057\u307e\u3059\u3002<\/p>\n\n\n
    \n>>>probability_model = tf.keras.Sequential([\n\u3000\u3000model,\n\u3000\u3000tf.keras.layers.Softmax()\n])\n>>>probability_model(x_test[:5])\n<\/pre><\/div>\n\n\n
    \u4ee5\u4e0b\u306e\u3088\u3046\u306a\u7d50\u679c\u306b\u306a\u308a\u307e\u3057\u305f\u3002<\/p>\n\n\n
    \n&lt;tf.Tensor: shape=(5, 10), dtype=float32, numpy=\n\u3000\u3000\u3000array([[7.41418518e-08, 7.91325783e-10, 3.75941590e-06, 2.61097477e-04,\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000 9.48091952e-11, 3.16857694e-07, 4.86965676e-11, 9.99731362e-01,\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000 4.52538757e-07, 2.88244155e-06],\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000[1.33913982e-11, 2.29599886e-04, 9.99751031e-01, 1.92872030e-05,\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000 1.66144136e-15, 1.58800315e-08, 2.13381668e-08, 9.98626540e-16,\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000 3.95744095e-08, 4.06508526e-14],\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000[4.50110349e-07, 9.99431074e-01, 3.84905434e-05, 1.16055037e-06,\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000 3.32167328e-05, 5.31233536e-06, 1.03985847e-04, 3.38415528e-04,\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000 4.75658890e-05, 1.63871476e-07],\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000[9.99950290e-01, 7.11293871e-12, 3.84013947e-05, 2.12064151e-07,\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000 1.02987372e-07, 4.20437999e-07, 2.38990742e-06, 2.65470180e-06,\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000 1.09450021e-07, 5.40031533e-06],\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000[7.92320952e-06, 2.07563033e-09, 8.12820872e-06, 2.62512536e-08,\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000 9.88838971e-01, 1.17090956e-07, 1.72804573e-06, 2.94550318e-05,\n\u3000\u3000\u3000\u3000\u3000\u3000\u3000 5.51807261e-06, 1.11081181e-02]], dtype=float32)>\n<\/pre><\/div>\n\n\n
    \u914d\u5217\u306e\u4e2d\u8eab\u3092\u78ba\u8a8d\u3057\u307e\u3059\u3002
    \uff11\u3064\u76ee\u306e\u30c6\u30b9\u30c8\u753b\u50cf\u306b\u5bfe\u3059\u308b\u63a8\u8ad6\u7d50\u679c\u306f\u3001\u6570\u5b57\u306e\uff17\u3067\u3042\u308b\u78ba\u7387\u304c\u7d0499%<\/strong>\u3068\u6700\u3082\u9ad8\u304f\u306a\u308a\u307e\u3057\u305f\u3002\uff081\u756a\u76ee\u306e\u914d\u5217\u306e\u4e26\u3073\u9806\u306b\u6570\u5b57\u306e0,1,2,3,4,5,6,7,8,9\u3067\u3042\u308b\u78ba\u7387\u3092\u793a\u3057\u3066\u304a\u308a\u30019.99731362e-01 \u3068\u4e00\u756a\u9ad8\u3044\u78ba\u7387\u3092\u793a\u3057\u3066\u3044\u308b\u306e\u304c\u3001\u6570\u5b57\u306e\uff17\u306b\u5bfe\u5fdc\u3057\u305f\u30ce\u30fc\u30c9\u3067\u3042\u308b\u304b\u3089\uff09
    \u4ee5\u4e0b\u3001\u540c\u69d8\u306b\u898b\u3066\u3044\u304d\u307e\u3059\u3002
    2\u3064\u76ee\u306e\u30c6\u30b9\u30c8\u753b\u50cf\u306e\u63a8\u8ad6\u7d50\u679c\u306f\u6570\u5b57\u306e2\u3067\u3042\u308b\u78ba\u7387\u304c\u7d0499\uff05<\/strong>\u3001
    3\u3064\u76ee\u306e\u30c6\u30b9\u30c8\u753b\u50cf\u306e\u63a8\u8ad6\u7d50\u679c\u306f\u6570\u5b57\u306e1\u3067\u3042\u308b\u78ba\u7387\u304c\u7d0499\uff05<\/strong>\u3001
    4\u3064\u76ee\u306e\u30c6\u30b9\u30c8\u753b\u50cf\u306e\u63a8\u8ad6\u7d50\u679c\u306f\u6570\u5b57\u306e0\u3067\u3042\u308b\u78ba\u7387\u304c\u7d0499\uff05<\/strong>\u3001
    5\u3064\u76ee\u306e\u30c6\u30b9\u30c8\u753b\u50cf\u306e\u63a8\u8ad6\u7d50\u679c\u306f\u6570\u5b57\u306e4\u3067\u3042\u308b\u78ba\u7387\u304c\u7d0499\uff05<\/strong>
    \u3068\u7b97\u51fa\u3055\u308c\u307e\u3057\u305f\u3002<\/p>\n\n\n
    \n
    \"MNIST\"<\/figure>\n<\/div>\n\n\n
    \u30c6\u30b9\u30c8\u7528\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306e\u753b\u50cf\u30d5\u30a1\u30a4\u30eb\u306e\u4e2d\u8eab\u306e\u4e00\u90e8\u306f\u4e0b\u8a18\u306e\u3088\u3046\u306a\u3063\u3066\u3044\u307e\u3059\u3002
    <\/p>\n\n\n\n
    \u5de6\u4e0a\u7aef\u304b\u3089\u53f3\u65b9\u5411\u306b\uff11\u3064\u76ee\u304b\u30895\u3064\u76ee\u306e\u753b\u50cf\u3068\u63a8\u8ad6\u7d50\u679c\u3092\u7167\u3089\u3057\u5408\u308f\u305b\u308b\u3068\u3001\uff17\u3001\uff12\u3001\uff11\u3001\uff10\u3001\uff14\u3067\u5408\u81f4\u3057\u3066\u3044\u308b\u3053\u3068\u304c\u308f\u304b\u308a\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
    \u516c\u5f0f\u30ac\u30a4\u30c9\u306e\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u306b\u6cbf\u3063\u3066\u3001\u624b\u66f8\u304d\u6587\u5b57\u8a8d\u8b58\u306e\u30e2\u30c7\u30eb\u69cb\u7bc9\u3001\u5b66\u7fd2\u3001\u63a8\u8ad6\u3068\u3044\u3046\u4e00\u9023\u306e\u6d41\u308c\u3092\u898b\u3066\u304d\u307e\u3057\u305f\u3002\u4eca\u56de\u306f\u3053\u3053\u3067\u7d42\u4e86\u3067\u3059\u3002
    \u30b3\u30f3\u30c6\u30ca\u5185\u90e8\u3067\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3092\u5229\u7528\u3059\u308b\u3053\u3068\u3067\u3001\u307b\u3068\u3093\u3069\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u3059\u308b\u3053\u3068\u306a\u304f\u3001\u95a2\u6570\u306e\u547c\u3073\u51fa\u3057\u3068\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u8a2d\u5b9a\u306b\u3088\u308a\u3001\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3067\u304d\u308b<\/strong>\u3053\u3068\u304c\u78ba\u8a8d\u3067\u304d\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
    \u6b21\u56de\u4e88\u544a \u7b2c4\u56de\uff1aChainer\u3092\u4f7f\u3063\u305f\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0<\/h2>\n\n\n\n
    \u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3068\u3057\u3066\u6709\u540d\u306aChainer\uff08\u30c1\u30a7\u30a4\u30ca\u30fc\uff09\u306e\u5229\u7528\u65b9\u6cd5\u3092\u7d39\u4ecb\u3057\u307e\u3059\u3002<\/p>\n\n\n
    \n
    HPC\u30b5\u30fc\u30d3\u30b9 SX-Aurora TSUBASA \u30af\u30e9\u30a6\u30c9<\/div>
    \n
    NEC\u306e\u30b9\u30fc\u30d1\u30fc\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u30fc\u300cSX-Aurora TSUBASA\u300d\u3092\u30af\u30e9\u30a6\u30c9\u74b0\u5883\u3067\u3054\u5229\u7528\u3067\u304d\u308b\u696d\u754c\u968f\u4e00\u306e\u30b5\u30fc\u30d3\u30b9\u3067\u3059\u3002<\/p>\n\n\n\n
    \u4e16\u754c\u30c8\u30c3\u30d7\u30af\u30e9\u30b9\u306e\u30b9\u30da\u30c3\u30af\u3067\u5927\u898f\u6a21\u30c7\u30fc\u30bf\u306e\u9ad8\u901f\u51e6\u7406\u3092\u5b9f\u73fe\u3059\u308b\u30d9\u30af\u30c8\u30eb\u578b\u30b9\u30fc\u30d1\u30fc\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u30fc\u3092\u3001\u6708\u984d\u5b9a\u984d\u6599\u91d1\u306e\u30af\u30e9\u30a6\u30c9\u30b5\u30fc\u30d3\u30b9\u3068\u3057\u3066\u5229\u7528\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n
    \"\"<\/a><\/figure>\n<\/div><\/div>\n<\/div><\/div>
      
              
            \u3010\u9023\u8f09\u4f01\u753b\u3011GPU\u30b3\u30f3\u30c6\u30ca\u6d3b\u7528 \u3010\u51686\u56de\u3011<\/th>      <\/tr>    <\/thead>    
          
            GPU\u30b3\u30f3\u30c6\u30ca\u3068\u306f\u4f55\u304b\uff1f\u4f55\u304c\u4fbf\u5229\u306a\u306e\u304b\uff1f\u3010\u7b2c1\u56de\uff1aGPU\u30b3\u30f3\u30c6\u30ca\u3067\u901f\u653b\u74b0\u5883\u69cb\u7bc9\u3011<\/a>        <\/td>      <\/tr>      
            TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2460\u3010\u7b2c2\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u6e96\u5099\u7de8\u301c\u3011<\/a>        <\/td>      <\/tr>      
            TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011        <\/td>      <\/tr>      
            Chainer\u3092\u4f7f\u3063\u305f\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3010\u7b2c4\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u6a5f\u68b0\u5b66\u7fd2\u3059\u308b\u3011<\/a>        <\/td>      <\/tr>      
            PyTorch\u3067\u6a5f\u68b0\u5b66\u7fd2\u3010\u7b2c5\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u30c6\u30f3\u30bd\u30eb\u306e\u57fa\u672c\u3092\u7406\u89e3\u3059\u308b\u3011<\/a>        <\/td>      <\/tr>    <\/tbody>  <\/table><\/div>","protected":false},"excerpt":{"rendered":"\u3010\u9023\u8f09\u4f01\u753b\u3011GPU\u30b3\u30f3\u30c6\u30ca\u6d3b\u7528 \u3010\u51686\u56de\u3011 GPU\u30b3\u30f3\u30c6\u30ca\u3068\u306f\u4f55\u304b\uff1f\u4f55\u304c\u4fbf\u5229\u306a\u306e\u304b\uff1f\u3010\u7b2c1\u56de\uff1aGPU\u30b3\u30f3\u30c6\u30ca\u3067\u901f\u653b\u74b0\u5883\u69cb\u7bc9\u3011 TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2460\u3010\u7b2c2\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u6e96\u5099\u7de8\u301c\u3011 TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf […]","protected":false},"author":10,"featured_media":3822,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_coblocks_attr":"","_coblocks_dimensions":"","_coblocks_responsive_height":"","_coblocks_accordion_ie_support":"","footnotes":""},"categories":[145],"tags":[67,100,34,56],"pickup":[],"sitedisplay":[],"sitetarget":[],"class_list":["post-3821","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-hpc","tag-67","tag-hpc","tag-gpu","tag-56"],"yoast_head":"\nTensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011 - \u30ab\u30b4\u30e4\u306e\u30b5\u30fc\u30d0\u30fc\u7814\u7a76\u5ba4<\/title>\n<meta name=\"description\" content=\"AI\u521d\u5b66\u8005\u306b\u3068\u3063\u3066\u30cf\u30fc\u30c9\u30eb\u306e\u3072\u3068\u3064\u306b\u306a\u3063\u3066\u3044\u308bGPU\u74b0\u5883\u69cb\u7bc9\u3092\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306b\u884c\u3046\u305f\u3081\u306b\u3001\u30b3\u30f3\u30c6\u30ca\u3092\u6709\u52b9\u6d3b\u7528\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u306b\u9023\u8f09\u3092\u30b9\u30bf\u30fc\u30c8\u3057\u307e\u3057\u305f\u3002\u524d\u56de\u306f\u30c7\u30a3\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/\" \/>\n<meta property=\"og:locale\" content=\"ja_JP\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011 - \u30ab\u30b4\u30e4\u306e\u30b5\u30fc\u30d0\u30fc\u7814\u7a76\u5ba4\" \/>\n<meta property=\"og:description\" content=\"AI\u521d\u5b66\u8005\u306b\u3068\u3063\u3066\u30cf\u30fc\u30c9\u30eb\u306e\u3072\u3068\u3064\u306b\u306a\u3063\u3066\u3044\u308bGPU\u74b0\u5883\u69cb\u7bc9\u3092\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306b\u884c\u3046\u305f\u3081\u306b\u3001\u30b3\u30f3\u30c6\u30ca\u3092\u6709\u52b9\u6d3b\u7528\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u306b\u9023\u8f09\u3092\u30b9\u30bf\u30fc\u30c8\u3057\u307e\u3057\u305f\u3002\u524d\u56de\u306f\u30c7\u30a3\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/\" \/>\n<meta property=\"og:site_name\" content=\"\u30ab\u30b4\u30e4\u306e\u30b5\u30fc\u30d0\u30fc\u7814\u7a76\u5ba4\" \/>\n<meta property=\"article:published_time\" content=\"2020-09-30T01:12:08+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2023-07-18T08:47:04+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.kagoya.jp\/howto\/wp-content\/uploads\/gpu202009b_cacth.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1200\" \/>\n\t<meta property=\"og:image:height\" content=\"630\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"\u9db4\u5ca1 \u8b19\u543e\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u57f7\u7b46\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"\u9db4\u5ca1 \u8b19\u543e\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u63a8\u5b9a\u8aad\u307f\u53d6\u308a\u6642\u9593\" \/>\n\t<meta name=\"twitter:data2\" content=\"31\u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/\"},\"author\":{\"name\":\"\u9db4\u5ca1 \u8b19\u543e\",\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/#\\\/schema\\\/person\\\/df4e644df3b6b3d2d54b27fb8623fb9d\"},\"headline\":\"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011\",\"datePublished\":\"2020-09-30T01:12:08+00:00\",\"dateModified\":\"2023-07-18T08:47:04+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/\"},\"wordCount\":260,\"image\":{\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/wp-content\\\/uploads\\\/gpu202009b_cacth.png\",\"keywords\":[\"\u5b66\u8853\",\"HPC\",\"GPU\",\"\u30b3\u30f3\u30c6\u30ca\"],\"articleSection\":[\"AI\u30fbIoT\u30fbHPC\"],\"inLanguage\":\"ja\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/\",\"url\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/\",\"name\":\"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011 - \u30ab\u30b4\u30e4\u306e\u30b5\u30fc\u30d0\u30fc\u7814\u7a76\u5ba4\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/wp-content\\\/uploads\\\/gpu202009b_cacth.png\",\"datePublished\":\"2020-09-30T01:12:08+00:00\",\"dateModified\":\"2023-07-18T08:47:04+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/#\\\/schema\\\/person\\\/df4e644df3b6b3d2d54b27fb8623fb9d\"},\"description\":\"AI\u521d\u5b66\u8005\u306b\u3068\u3063\u3066\u30cf\u30fc\u30c9\u30eb\u306e\u3072\u3068\u3064\u306b\u306a\u3063\u3066\u3044\u308bGPU\u74b0\u5883\u69cb\u7bc9\u3092\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306b\u884c\u3046\u305f\u3081\u306b\u3001\u30b3\u30f3\u30c6\u30ca\u3092\u6709\u52b9\u6d3b\u7528\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u306b\u9023\u8f09\u3092\u30b9\u30bf\u30fc\u30c8\u3057\u307e\u3057\u305f\u3002\u524d\u56de\u306f\u30c7\u30a3\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/#breadcrumb\"},\"inLanguage\":\"ja\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"ja\",\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/#primaryimage\",\"url\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/wp-content\\\/uploads\\\/gpu202009b_cacth.png\",\"contentUrl\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/wp-content\\\/uploads\\\/gpu202009b_cacth.png\",\"width\":1200,\"height\":630,\"caption\":\"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/engineer\\\/hpc\\\/gpu-container3\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u30db\u30fc\u30e0\",\"item\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/#website\",\"url\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/\",\"name\":\"\u30ab\u30b4\u30e4\u306e\u30b5\u30fc\u30d0\u30fc\u7814\u7a76\u5ba4\",\"description\":\"\u30c7\u30fc\u30bf\u30bb\u30f3\u30bf\u30fc\u904b\u55b6\u4f1a\u793e\u304c\u304a\u5c4a\u3051\u3059\u308b\u30bd\u30ea\u30e5\u30fc\u30b7\u30e7\u30f3\u30e1\u30c7\u30a3\u30a2\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"ja\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/#\\\/schema\\\/person\\\/df4e644df3b6b3d2d54b27fb8623fb9d\",\"name\":\"\u9db4\u5ca1 \u8b19\u543e\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"ja\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/3aa257cb88c81f1f1b20b69ff2c2449d9943b4413ede75b7abcb3d7e1cc78850?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/3aa257cb88c81f1f1b20b69ff2c2449d9943b4413ede75b7abcb3d7e1cc78850?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/3aa257cb88c81f1f1b20b69ff2c2449d9943b4413ede75b7abcb3d7e1cc78850?s=96&d=mm&r=g\",\"caption\":\"\u9db4\u5ca1 \u8b19\u543e\"},\"url\":\"https:\\\/\\\/www.kagoya.jp\\\/howto\\\/author\\\/tsuruoka\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011 - \u30ab\u30b4\u30e4\u306e\u30b5\u30fc\u30d0\u30fc\u7814\u7a76\u5ba4","description":"AI\u521d\u5b66\u8005\u306b\u3068\u3063\u3066\u30cf\u30fc\u30c9\u30eb\u306e\u3072\u3068\u3064\u306b\u306a\u3063\u3066\u3044\u308bGPU\u74b0\u5883\u69cb\u7bc9\u3092\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306b\u884c\u3046\u305f\u3081\u306b\u3001\u30b3\u30f3\u30c6\u30ca\u3092\u6709\u52b9\u6d3b\u7528\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u306b\u9023\u8f09\u3092\u30b9\u30bf\u30fc\u30c8\u3057\u307e\u3057\u305f\u3002\u524d\u56de\u306f\u30c7\u30a3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/","og_locale":"ja_JP","og_type":"article","og_title":"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011 - \u30ab\u30b4\u30e4\u306e\u30b5\u30fc\u30d0\u30fc\u7814\u7a76\u5ba4","og_description":"AI\u521d\u5b66\u8005\u306b\u3068\u3063\u3066\u30cf\u30fc\u30c9\u30eb\u306e\u3072\u3068\u3064\u306b\u306a\u3063\u3066\u3044\u308bGPU\u74b0\u5883\u69cb\u7bc9\u3092\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306b\u884c\u3046\u305f\u3081\u306b\u3001\u30b3\u30f3\u30c6\u30ca\u3092\u6709\u52b9\u6d3b\u7528\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u306b\u9023\u8f09\u3092\u30b9\u30bf\u30fc\u30c8\u3057\u307e\u3057\u305f\u3002\u524d\u56de\u306f\u30c7\u30a3","og_url":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/","og_site_name":"\u30ab\u30b4\u30e4\u306e\u30b5\u30fc\u30d0\u30fc\u7814\u7a76\u5ba4","article_published_time":"2020-09-30T01:12:08+00:00","article_modified_time":"2023-07-18T08:47:04+00:00","og_image":[{"width":1200,"height":630,"url":"https:\/\/www.kagoya.jp\/howto\/wp-content\/uploads\/gpu202009b_cacth.png","type":"image\/png"}],"author":"\u9db4\u5ca1 \u8b19\u543e","twitter_card":"summary_large_image","twitter_misc":{"\u57f7\u7b46\u8005":"\u9db4\u5ca1 \u8b19\u543e","\u63a8\u5b9a\u8aad\u307f\u53d6\u308a\u6642\u9593":"31\u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/#article","isPartOf":{"@id":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/"},"author":{"name":"\u9db4\u5ca1 \u8b19\u543e","@id":"https:\/\/www.kagoya.jp\/howto\/#\/schema\/person\/df4e644df3b6b3d2d54b27fb8623fb9d"},"headline":"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011","datePublished":"2020-09-30T01:12:08+00:00","dateModified":"2023-07-18T08:47:04+00:00","mainEntityOfPage":{"@id":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/"},"wordCount":260,"image":{"@id":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/#primaryimage"},"thumbnailUrl":"https:\/\/www.kagoya.jp\/howto\/wp-content\/uploads\/gpu202009b_cacth.png","keywords":["\u5b66\u8853","HPC","GPU","\u30b3\u30f3\u30c6\u30ca"],"articleSection":["AI\u30fbIoT\u30fbHPC"],"inLanguage":"ja"},{"@type":"WebPage","@id":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/","url":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/","name":"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011 - \u30ab\u30b4\u30e4\u306e\u30b5\u30fc\u30d0\u30fc\u7814\u7a76\u5ba4","isPartOf":{"@id":"https:\/\/www.kagoya.jp\/howto\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/#primaryimage"},"image":{"@id":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/#primaryimage"},"thumbnailUrl":"https:\/\/www.kagoya.jp\/howto\/wp-content\/uploads\/gpu202009b_cacth.png","datePublished":"2020-09-30T01:12:08+00:00","dateModified":"2023-07-18T08:47:04+00:00","author":{"@id":"https:\/\/www.kagoya.jp\/howto\/#\/schema\/person\/df4e644df3b6b3d2d54b27fb8623fb9d"},"description":"AI\u521d\u5b66\u8005\u306b\u3068\u3063\u3066\u30cf\u30fc\u30c9\u30eb\u306e\u3072\u3068\u3064\u306b\u306a\u3063\u3066\u3044\u308bGPU\u74b0\u5883\u69cb\u7bc9\u3092\u3067\u304d\u308b\u3060\u3051\u52b9\u7387\u7684\u306b\u884c\u3046\u305f\u3081\u306b\u3001\u30b3\u30f3\u30c6\u30ca\u3092\u6709\u52b9\u6d3b\u7528\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u306b\u9023\u8f09\u3092\u30b9\u30bf\u30fc\u30c8\u3057\u307e\u3057\u305f\u3002\u524d\u56de\u306f\u30c7\u30a3","breadcrumb":{"@id":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/#breadcrumb"},"inLanguage":"ja","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/"]}]},{"@type":"ImageObject","inLanguage":"ja","@id":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/#primaryimage","url":"https:\/\/www.kagoya.jp\/howto\/wp-content\/uploads\/gpu202009b_cacth.png","contentUrl":"https:\/\/www.kagoya.jp\/howto\/wp-content\/uploads\/gpu202009b_cacth.png","width":1200,"height":630,"caption":"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0"},{"@type":"BreadcrumbList","@id":"https:\/\/www.kagoya.jp\/howto\/engineer\/hpc\/gpu-container3\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u30db\u30fc\u30e0","item":"https:\/\/www.kagoya.jp\/howto\/"},{"@type":"ListItem","position":2,"name":"TensorFlow\u3068Keras\u306b\u3088\u308b\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u2461\u3010\u7b2c3\u56de:GPU\u30b3\u30f3\u30c6\u30ca\u3067\u753b\u50cf\u89e3\u6790\u301c\u5b9f\u8df5\u7de8\u301c\u3011"}]},{"@type":"WebSite","@id":"https:\/\/www.kagoya.jp\/howto\/#website","url":"https:\/\/www.kagoya.jp\/howto\/","name":"\u30ab\u30b4\u30e4\u306e\u30b5\u30fc\u30d0\u30fc\u7814\u7a76\u5ba4","description":"\u30c7\u30fc\u30bf\u30bb\u30f3\u30bf\u30fc\u904b\u55b6\u4f1a\u793e\u304c\u304a\u5c4a\u3051\u3059\u308b\u30bd\u30ea\u30e5\u30fc\u30b7\u30e7\u30f3\u30e1\u30c7\u30a3\u30a2","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.kagoya.jp\/howto\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"ja"},{"@type":"Person","@id":"https:\/\/www.kagoya.jp\/howto\/#\/schema\/person\/df4e644df3b6b3d2d54b27fb8623fb9d","name":"\u9db4\u5ca1 \u8b19\u543e","image":{"@type":"ImageObject","inLanguage":"ja","@id":"https:\/\/secure.gravatar.com\/avatar\/3aa257cb88c81f1f1b20b69ff2c2449d9943b4413ede75b7abcb3d7e1cc78850?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/3aa257cb88c81f1f1b20b69ff2c2449d9943b4413ede75b7abcb3d7e1cc78850?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/3aa257cb88c81f1f1b20b69ff2c2449d9943b4413ede75b7abcb3d7e1cc78850?s=96&d=mm&r=g","caption":"\u9db4\u5ca1 \u8b19\u543e"},"url":"https:\/\/www.kagoya.jp\/howto\/author\/tsuruoka\/"}]}},"views":26206,"_links":{"self":[{"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/posts\/3821","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/users\/10"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/comments?post=3821"}],"version-history":[{"count":0,"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/posts\/3821\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/media\/3822"}],"wp:attachment":[{"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/media?parent=3821"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/categories?post=3821"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/tags?post=3821"},{"taxonomy":"pickup","embeddable":true,"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/pickup?post=3821"},{"taxonomy":"sitedisplay","embeddable":true,"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/sitedisplay?post=3821"},{"taxonomy":"sitetarget","embeddable":true,"href":"https:\/\/www.kagoya.jp\/howto\/wp-json\/wp\/v2\/sitetarget?post=3821"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}