{"id":45698,"date":"2023-09-11T05:54:53","date_gmt":"2024-03-09T22:57:41","guid":{"rendered":"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/"},"modified":"2024-05-03T23:09:45","modified_gmt":"2024-05-03T15:09:45","slug":"45698-2","status":"publish","type":"post","link":"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/","title":{"rendered":""},"content":{"rendered":"<p>\u672c\u8a18\u4e8b\u306fRust Advent Calendar 2017\u306e12\u670821\u65e5\u306e\u8a18\u4e8b\u3067\u3059\u3002<\/p>\n<p>Rust\u3067\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3057\u305f\u304f\u3066Rust\u3067\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3067\u304d\u308b\u3088\u3046\u306b\u3057\u305f\u306e\u3067\u7d39\u4ecb\u3057\u307e\u3059\u3002<\/p>\n<h1>\u306f\u3058\u3081\u306b<\/h1>\n<p>\u6628\u4eca\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u304c\u52c3\u8208\u3057\u9ad8\u6a5f\u80fd\u5316\u304c\u9032\u3080\u4e00\u65b9\u3067\u3001Theano\u306e\u3088\u3046\u306b\u4e00\u6642\u4ee3\u3092\u7bc9\u3044\u305f\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306e\u958b\u767a\u505c\u6b62\u304c\u30a2\u30ca\u30a6\u30f3\u30b9\u3055\u308c\u308b\u306a\u30691\u3001\u79fb\u308a\u5909\u308f\u308a\u304c\u6fc0\u3057\u3044\u72b6\u6cc1\u3067\u3059\u3002<br \/>\n\u591a\u304f\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306fPython\u3067\u958b\u767a\u3001\u3042\u308b\u3044\u306f\u516c\u5f0f\u306bPython API\u306e\u63d0\u4f9b\u3092\u3057\u3066\u304a\u308a\u3001\u6a5f\u68b0\u5b66\u7fd2\u30fb\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30e9\u30a4\u30d6\u30e9\u30ea\u306fPython\u3092\u4e2d\u5fc3\u306b\u5229\u7528\u3067\u304d\u308b\u3088\u3046\u306b\u306a\u3063\u3066\u3044\u308b\u306e\u3067\u3059\u304c\u3001\u8af8\u3005\u306e\u4e8b\u60c5\u3067Python\u3067\u3082\u306a\u304fC++\u3067\u3082\u306a\u304fRust\u3067\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u304c\u3057\u305f\u3044\u3068\u3044\u3046\u30cb\u30fc\u30ba\u304c\u3042\u308b\u304b\u3068\u601d\u3044\u307e\u3059\u3002<\/p>\n<h2>\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0 in Rust<\/h2>\n<p>\u79c1\u306e\u77e5\u308b\u9650\u308a\u3067\u306fRust\u754c\u9688\u306e\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u4e8b\u60c5\u306f\u4ee5\u4e0b\u306e\u3068\u304a\u308a\u3067\u3059\u3002<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">2015\u5e74 \u30d9\u30eb\u30ea\u30f3\u306e\u30b9\u30bf\u30fc\u30c8\u30a2\u30c3\u30d7 Autumn (autumunai) \u304c Leaf \u3092\u958b\u767a<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">2016\u5e74 TensorFlow\u304c\u516c\u5f0f\u306eRust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u3092\u63d0\u4f9b\u958b\u59cb<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u540c\u5e74Leaf\u306e\u958b\u767a\u304c\u505c\u6b62<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">2017\u5e74 PyTorch\u306eRust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0 (\u975e\u516c\u5f0f?) , DyNet\u306eRust\u306e\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0 (\u975e\u516c\u5f0f) \u306a\u3069\u3001\u500b\u4eba\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3068\u3057\u3066Rust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u304c\u958b\u767a\u3055\u308c\u308b<\/ul>\n<p>Leaf\u306f\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306e\u30b3\u30a2\u306e\u90e8\u5206\u307e\u3067Rust\u3067\u66f8\u304b\u308c\u3066\u3044\u305f\u306e\u3067\u3059\u304c\u3001\u5f53\u6642\u306fRust\u3067CPU\/GPU\u3067\u9ad8\u901f\u306b\u884c\u5217\u6f14\u7b97\u3067\u304d\u308b\u30e9\u30a4\u30d6\u30e9\u30ea\u304c\u306a\u304f2\u3001\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30b3\u30a2\u306e\u90e8\u5206(\u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9, \u5fae\u5206\u8a08\u7b97\u306e\u8a18\u8ff0, backward\/update)\u81ea\u4f53\u3088\u308a\u3082\u30d0\u30c3\u30af\u30a8\u30f3\u30c9\u306e\u884c\u5217\u6f14\u7b97\u306e\u30e9\u30a4\u30d6\u30e9\u30ea\u5229\u7528\u30fb\u958b\u767a\u306b\u96e3\u5100\u3057\u3066\u3044\u305f\u3088\u3046\u3067\u3059\u30023 4<br \/>\n\u305d\u306e\u3088\u3046\u306a\u72b6\u6cc1\u3067TensorFlow\u306eRust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u304c\u516c\u958b\u3055\u308c\u3001Autumn\u306e\u958b\u767a\u8005\u304cLeaf\u306e\u958b\u767a\u3092\u505c\u6b62\u3057\u305f\u306e\u3067\u3059\u304c5\u3001TensorFlow\u306eRust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u306f\u5b66\u7fd2\u6e08\u307f\u306e\u30e2\u30c7\u30eb\u3092load\u3057\u3066inference\u306b\u4f7f\u7528\u3067\u304d\u308b\u3060\u3051\u3068\u3044\u3046\u3082\u306e\u3067\u3001\u5b9f\u969b\u306b\u306fPython\u306a\u3069\u4ed6\u306e\u8a00\u8a9e\u3067\u30e2\u30c7\u30eb\u306e\u958b\u767a\u30fb\u5b66\u7fd2\u30fb\u4fdd\u5b58\u3059\u308b\u5fc5\u8981\u304c\u3042\u308a\u3001Leaf\u306e\u958b\u767a\u306e\u4e2d\u6b62\u306f\u975e\u5e38\u306b\u6b8b\u5ff5\u3067\u3057\u305f\u30026<br \/>\n\u305d\u3053\u3067\u500b\u4eba\u7684\u306bDyNet\u306eRust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u3092\u8a66\u4f5c\u3057\u3066\u3042\u308b\u7a0b\u5ea6\u52d5\u304f\u72b6\u6cc1\u3068\u306a\u3063\u305f\u306e\u3067\u3059\u304c7\u3001\u672c\u683c\u7684\u306b\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u3092\u4f5c\u308b\u306b\u306f\u307e\u305aFFI\u3067\u95a2\u6570\u547c\u3073\u51fa\u3057\u3092\u3059\u308b\u305f\u3081\u306eC API\u306e\u958b\u767a\u304c\u5fc5\u8981\u30678\u3001\u305d\u308c\u306b\u306f\u30b3\u30a2\u306e\u4ed5\u69d8\u30fb\u5b9f\u88c5\u306e\u7406\u89e3\u304b\u3089\u59cb\u3081\u3066\u81a8\u5927\u306a\u4f5c\u696d\u6642\u9593\u3092\u8981\u3059\u308b\u305f\u3081\u3001\u73fe\u5728\u306f\u65b9\u91dd\u3092\u691c\u8a0e\u4e2d\u3067\u3059\u30029<\/p>\n<p>\u305d\u306e\u3088\u3046\u306a\u306a\u304bprimitiv\u3068\u3044\u3046C++\u3067\u958b\u767a\u3055\u308c\u305f\u65b0\u8208\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u304c\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u3068\u3057\u3066\u516c\u958b\u3055\u308c\u307e\u3057\u305f\u3002<br \/>\n\u500b\u4eba\u7684\u306a\u610f\u898b\u3067\u3059\u304cprimitiv\u306f\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u304c\u975e\u5e38\u306b\u8a18\u8ff0\u3057\u3084\u3059\u304f\u3001\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306e\u7279\u5fb4\u304c\u81ea\u5206\u306e\u30cb\u30fc\u30ba\u306b\u30de\u30c3\u30c1\u3057\u3066\u304a\u308a\u3001\u8208\u5473\u3092\u6301\u3061\u307e\u3057\u305f\u3002<br \/>\n\u958b\u767a\u306e\u4e2d\u5fc3\u3068\u306a\u3063\u3066\u3044\u308b @odashi_t \u3055\u3093\u3068\u540c\u3058\u7814\u7a76\u30b3\u30df\u30e5\u30cb\u30c6\u30a3\u30fb\u5927\u5b66\u9662\u306b\u6240\u5c5e\u3057\u3066\u3044\u3066\u30b3\u30df\u30e5\u30cb\u30b1\u30fc\u30b7\u30e7\u30f3\u304c\u53d6\u308a\u3084\u3059\u3044\u72b6\u6cc1\u3067\u3042\u3063\u305f\u305f\u3081\u3001primitiv\u306e\u958b\u767a\u306b\u52a0\u308f\u308b\u3053\u3068\u306b\u306a\u308a\u307e\u3057\u305f\u3002<br \/>\n\u73fe\u5728\u306f\u79c1\u306e\u62c5\u5f53\u3068\u3057\u3066C API\u3068Rust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u306e\u958b\u767a\u3092\u9032\u3081\u3066\u3044\u307e\u3059\u3002<\/p>\n<h2>primitiv\u3068\u306f<\/h2>\n<p>primitiv\u306f\u3082\u3068\u3082\u3068 NICT (\u60c5\u5831\u901a\u4fe1\u7814\u7a76\u6a5f\u69cb) \u3067\u958b\u767a\u3055\u308c\u3001\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u3068\u3057\u3066\u516c\u958b\u3055\u308c\u3066\u4ee5\u964d\u306fNAIST\u306e\u97f3\u58f0\u30fb\u81ea\u7136\u8a00\u8a9e\u51e6\u7406\u306e\u7814\u7a76\u30b0\u30eb\u30fc\u30d7\u3092\u4e2d\u5fc3\u306b\u958b\u767a\u304c\u9032\u3081\u3089\u308c\u3066\u3044\u307e\u3059\u300210<\/p>\n<p>primitiv\u306e\u7279\u5fb4\u306f\u3001<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u884c\u5217\u6f14\u7b97\u306e\u30d0\u30c3\u30af\u30a8\u30f3\u30c9\u306e\u67d4\u8edf\u306a\u5207\u66ff<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">Define by Run\u65b9\u5f0f\u306e\u52d5\u7684\u306a\u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9<\/ul>\n<p>\u304c\u6319\u3052\u3089\u308c\u3001\u3056\u3063\u304f\u308a\u3068\u8a00\u3046\u3068DyNet\u306e\u30b9\u30d4\u30f3\u30aa\u30d5\u306e\u3088\u3046\u306a\u4f4d\u7f6e\u4ed8\u3051\u3067\u3059\u3002<\/p>\n<p>\u3053\u308c\u307e\u3067Chainer\u3084DyNet\u3092\u4f7f\u3063\u3066\u304d\u305f\u79c1\u306e\u611f\u60f3\u3068\u3057\u3066\u3001\u30a4\u30f3\u30bf\u30fc\u30d5\u30a7\u30fc\u30b9\u3068\u3057\u3066\u306f\u5404\u30e2\u30b8\u30e5\u30fc\u30eb\u306e\u69cb\u6210\u306fChainer\u306b\u8fd1\u304f\u3001\u6570\u5f0f\u3092Theano\u306e\u3088\u3046\u306b\u76f4\u611f\u7684\u306b\u8a18\u8ff0\u3067\u304d\u308b\u3068\u3053\u308d\u304c\u512a\u308c\u305f\u70b9\u3060\u3068\u601d\u3044\u307e\u3059\u3002<br \/>\n\u884c\u5217\u6f14\u7b97\u306e\u30d0\u30c3\u30af\u30a8\u30f3\u30c9\u306f\u3001\u4f9d\u5b58\u30e9\u30a4\u30d6\u30e9\u30ea\u306a\u3057\u306e\u30ca\u30a4\u30fc\u30d6\u306aCPU\u3067\u306e\u6f14\u7b97\u3001Eigen\u3092\u4f7f\u3063\u305fCPU\u3067\u306e\u9ad8\u901f\u306a\u6f14\u7b97\u3001NVIDIA\u306eCUDA\u3092\u4f7f\u3063\u305fGPU\u3067\u306e\u6f14\u7b97\u3001OpenCL\u3092\u4f7f\u3063\u305fAMD\u3084Intel\u306eGPU\u3067\u306e\u6f14\u7b97\u304c\u5207\u66ff\u53ef\u80fd\u3067\u3059\u3002<br \/>\n\u8a08\u7b97\u30b0\u30e9\u30d5\u306fDefine by Run\u306b\u3088\u3063\u3066\u69cb\u7bc9\u3055\u308c\u307e\u3059\u304c\u3001\u5b9f\u969b\u306eforward\u8a08\u7b97\u306f\u660e\u793a\u7684\u306b\u8a08\u7b97\u7d50\u679c\u3092\u5f97\u308b\u64cd\u4f5c\u3092\u547c\u3073\u51fa\u3059\u307e\u3067\u306f\u884c\u308f\u308c\u307e\u305b\u3093\u3002<br \/>\n\u3053\u306e\u9045\u5ef6\u8a55\u4fa1\u65b9\u5f0f\u306f\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u5074\u3067\u8a08\u7b97\u304c\u6700\u9069\u5316\u3057\u3084\u3059\u304f\u306a\u308b\u3068\u3044\u3046\u5229\u70b9\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n<h1>\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5<\/h1>\n<p>primitiv Rust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5\u306b\u3064\u3044\u3066\u8aac\u660e\u3057\u307e\u3059\u3002<br \/>\n\u306a\u304a\u4e0b\u8a18\u624b\u9806\u304a\u3088\u3073\u30b3\u30fc\u30c9\u306f\u672c\u8a18\u4e8b\u57f7\u7b46\u6642\u70b9\u306e\u3082\u306e\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"c\"># primitiv core\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb<\/span>\r\n<span class=\"nv\">$ <\/span>git clone https:\/\/github.com\/primitiv\/primitiv\/\r\n<span class=\"nv\">$ <\/span><span class=\"nb\">cd <\/span>primitiv\r\n<span class=\"nv\">$ <\/span>git checkout dc022e3fd4c343f7b46f2d04698f940211bca773\r\n<span class=\"nv\">$ <\/span><span class=\"nb\">mkdir <\/span>build\r\n<span class=\"nv\">$ <\/span><span class=\"nb\">cd <\/span>build\r\n<span class=\"nv\">$ <\/span>cmake .. <span class=\"nt\">-DPRIMITIV_BUILD_C_API<\/span><span class=\"o\">=<\/span>ON\r\n<span class=\"nv\">$ <\/span>make\r\n<span class=\"nv\">$ <\/span>make <span class=\"nb\">install<\/span>\r\n\r\n<span class=\"c\"># primitiv-rust\u306e\u30d3\u30eb\u30c9<\/span>\r\n<span class=\"nv\">$ <\/span><span class=\"nb\">cd<\/span> ..\/..\/\r\n<span class=\"nv\">$ <\/span>git clone https:\/\/github.com\/primitiv\/primitiv-rust\/\r\n<span class=\"nv\">$ <\/span>git checkout 0cea052615172e71695d20c80e480fc91b78bc2c\r\n<span class=\"nv\">$ <\/span><span class=\"nb\">cd <\/span>primitiv-rust\r\n<span class=\"nv\">$ <\/span>cargo build\r\n<\/code><\/pre>\n<h1>\u5b9f\u88c5\u4f8b<\/h1>\n<p>MNIST\u3092\u4f8b\u306b\u5b9f\u88c5\u65b9\u6cd5\u3092\u7c21\u5358\u306b\u8aac\u660e\u3057\u307e\u3059\u3002<\/p>\n<h2>\u30e2\u30b8\u30e5\u30fc\u30eb\u306eimport\u3001\u30c7\u30fc\u30bf\u306e\u8aad\u307f\u8fbc\u307f<\/h2>\n<p>\u4eca\u56de\u306e\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3067\u4f7f\u7528\u3059\u308b\u30e2\u30b8\u30e5\u30fc\u30eb\u3068\u30c7\u30fc\u30bf\u3092\u8aad\u307f\u8fbc\u307f\u307e\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"c\">\/\/ \u30e2\u30b8\u30e5\u30fc\u30eb\u306eimport<\/span>\r\n<span class=\"k\">use<\/span> <span class=\"nn\">primitiv<\/span><span class=\"p\">::<\/span><span class=\"n\">device<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">use<\/span> <span class=\"nn\">primitiv<\/span><span class=\"p\">::<\/span><span class=\"n\">Graph<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">use<\/span> <span class=\"nn\">primitiv<\/span><span class=\"p\">::<\/span><span class=\"n\">Optimizer<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">use<\/span> <span class=\"nn\">primitiv<\/span><span class=\"p\">::<\/span><span class=\"n\">Parameter<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">use<\/span> <span class=\"nn\">primitiv<\/span><span class=\"p\">::<\/span><span class=\"n\">Shape<\/span><span class=\"p\">;<\/span>\r\n\r\n<span class=\"k\">use<\/span> <span class=\"nn\">primitiv<\/span><span class=\"p\">::<\/span><span class=\"n\">devices<\/span> <span class=\"k\">as<\/span> <span class=\"n\">D<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">use<\/span> <span class=\"nn\">primitiv<\/span><span class=\"p\">::<\/span><span class=\"n\">functions<\/span> <span class=\"k\">as<\/span> <span class=\"n\">F<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">use<\/span> <span class=\"nn\">primitiv<\/span><span class=\"p\">::<\/span><span class=\"n\">initializers<\/span> <span class=\"k\">as<\/span> <span class=\"n\">I<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">use<\/span> <span class=\"nn\">primitiv<\/span><span class=\"p\">::<\/span><span class=\"n\">optimizers<\/span> <span class=\"k\">as<\/span> <span class=\"n\">O<\/span><span class=\"p\">;<\/span>\r\n\r\n<span class=\"c\">\/\/ \u30b5\u30f3\u30d7\u30eb\u306b\u4f7f\u7528\u3059\u308b\u5b9a\u6570\u306e\u5b9a\u7fa9<\/span>\r\n<span class=\"k\">const<\/span> <span class=\"n\">NUM_TRAIN_SAMPLES<\/span><span class=\"p\">:<\/span> <span class=\"nb\">u32<\/span> <span class=\"o\">=<\/span> <span class=\"mi\">60000<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">const<\/span> <span class=\"n\">NUM_TEST_SAMPLES<\/span><span class=\"p\">:<\/span> <span class=\"nb\">u32<\/span> <span class=\"o\">=<\/span> <span class=\"mi\">10000<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">const<\/span> <span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">:<\/span> <span class=\"nb\">u32<\/span> <span class=\"o\">=<\/span> <span class=\"mi\">28<\/span> <span class=\"o\">*<\/span> <span class=\"mi\">28<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">const<\/span> <span class=\"n\">NUM_HIDDEN_UNITS<\/span><span class=\"p\">:<\/span> <span class=\"nb\">u32<\/span> <span class=\"o\">=<\/span> <span class=\"mi\">800<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">const<\/span> <span class=\"n\">NUM_OUTPUT_UNITS<\/span><span class=\"p\">:<\/span> <span class=\"nb\">u32<\/span> <span class=\"o\">=<\/span> <span class=\"mi\">10<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">const<\/span> <span class=\"n\">BATCH_SIZE<\/span><span class=\"p\">:<\/span> <span class=\"nb\">u32<\/span> <span class=\"o\">=<\/span> <span class=\"mi\">200<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">const<\/span> <span class=\"n\">NUM_TRAIN_BATCHES<\/span><span class=\"p\">:<\/span> <span class=\"nb\">u32<\/span> <span class=\"o\">=<\/span> <span class=\"n\">NUM_TRAIN_SAMPLES<\/span> <span class=\"o\">\/<\/span> <span class=\"n\">BATCH_SIZE<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">const<\/span> <span class=\"n\">NUM_TEST_BATCHES<\/span><span class=\"p\">:<\/span> <span class=\"nb\">u32<\/span> <span class=\"o\">=<\/span> <span class=\"n\">NUM_TEST_SAMPLES<\/span> <span class=\"o\">\/<\/span> <span class=\"n\">BATCH_SIZE<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">const<\/span> <span class=\"n\">MAX_EPOCH<\/span><span class=\"p\">:<\/span> <span class=\"nb\">u32<\/span> <span class=\"o\">=<\/span> <span class=\"mi\">100<\/span><span class=\"p\">;<\/span>\r\n\r\n<span class=\"o\">...<\/span>\r\n\r\n<span class=\"k\">fn<\/span> <span class=\"nf\">main<\/span><span class=\"p\">()<\/span> <span class=\"p\">{<\/span>\r\n    <span class=\"c\">\/\/ \u30c7\u30fc\u30bf\u306e\u8aad\u307f\u8fbc\u307f<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"n\">train_inputs<\/span> <span class=\"o\">=<\/span> <span class=\"nf\">load_images<\/span><span class=\"p\">(<\/span><span class=\"s\">\"data\/train-images-idx3-ubyte\"<\/span><span class=\"p\">,<\/span> <span class=\"n\">NUM_TRAIN_SAMPLES<\/span><span class=\"p\">);<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"n\">train_labels<\/span> <span class=\"o\">=<\/span> <span class=\"nf\">load_labels<\/span><span class=\"p\">(<\/span><span class=\"s\">\"data\/train-labels-idx1-ubyte\"<\/span><span class=\"p\">,<\/span> <span class=\"n\">NUM_TRAIN_SAMPLES<\/span><span class=\"p\">);<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"n\">test_inputs<\/span> <span class=\"o\">=<\/span> <span class=\"nf\">load_images<\/span><span class=\"p\">(<\/span><span class=\"s\">\"data\/t10k-images-idx3-ubyte\"<\/span><span class=\"p\">,<\/span> <span class=\"n\">NUM_TEST_SAMPLES<\/span><span class=\"p\">);<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"n\">test_labels<\/span> <span class=\"o\">=<\/span> <span class=\"nf\">load_labels<\/span><span class=\"p\">(<\/span><span class=\"s\">\"data\/t10k-labels-idx1-ubyte\"<\/span><span class=\"p\">,<\/span> <span class=\"n\">NUM_TEST_SAMPLES<\/span><span class=\"p\">);<\/span>\r\n<\/code><\/pre>\n<h2>\u30c7\u30d0\u30a4\u30b9\u306e\u767b\u9332<\/h2>\n<p>primitiv\u3067\u306f\u306f\u3058\u3081\u306bDevice\u306e\u30aa\u30d6\u30b8\u30a7\u30af\u30c8\u3092\u751f\u6210\u3057\u3001device::set_default()\u306b\u3088\u3063\u3066\u30c7\u30d5\u30a9\u30eb\u30c8\u306e\u30c7\u30d0\u30a4\u30b9\u3068\u3057\u3066\u767b\u9332\u3059\u308b\u3053\u3068\u3067\u3001\u4ee5\u964d\u306e\u8a08\u7b97\u3092\u30c7\u30d5\u30a9\u30eb\u30c8\u30c7\u30d0\u30a4\u30b9\u4e0a\u3067\u884c\u3044\u307e\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code>    <span class=\"c\">\/\/ \u30c7\u30d5\u30a9\u30eb\u30c8\u30c7\u30d0\u30a4\u30b9\u306e\u767b\u9332<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">dev<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">D<\/span><span class=\"p\">::<\/span><span class=\"nn\">Naive<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">();<\/span>\r\n    <span class=\"nn\">device<\/span><span class=\"p\">::<\/span><span class=\"nf\">set_default<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">dev<\/span><span class=\"p\">);<\/span>\r\n<\/code><\/pre>\n<p>\u5f8c\u8ff0\u3057\u307e\u3059\u304c\u3001\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u521d\u671f\u5316\u3084\u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u5165\u529b\u30ce\u30fc\u30c9\u306e\u6307\u5b9a\u306b\u660e\u793a\u7684\u306b\u30c7\u30d0\u30a4\u30b9\u3092\u6307\u5b9a\u3059\u308b\u3053\u3068\u3067\u3001\u30c7\u30d5\u30a9\u30eb\u30c8\u4ee5\u5916\u306e\u30c7\u30d0\u30a4\u30b9\u4e0a\u3067\u6f14\u7b97\u304c\u3067\u304d\u3001\u8907\u6570\u306e\u30c7\u30d0\u30a4\u30b9\u3092\u4f7f\u3063\u305f\u51e6\u7406\u3082\u53ef\u80fd\u3067\u3059\u3002<\/p>\n<h2>\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u5b9a\u7fa9\u30fb\u521d\u671f\u5316\u3001Optimizer\u3078\u306e\u767b\u9332<\/h2>\n<p>\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u521d\u671f\u5316\u306b\u306f\u660e\u793a\u7684\u306bInitializer\u306e\u30aa\u30d6\u30b8\u30a7\u30af\u30c8\u3092\u6307\u5b9a\u3059\u308b\u3053\u3068\u3067\u884c\u3044\u307e\u3059\u300211<\/p>\n<pre class=\"post-pre\"><code>    <span class=\"c\">\/\/ \u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u5b9a\u7fa9\u30fb\u521d\u671f\u5316<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">pw1<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">Parameter<\/span><span class=\"p\">::<\/span><span class=\"nf\">from_initializer<\/span><span class=\"p\">([<\/span><span class=\"n\">NUM_HIDDEN_UNITS<\/span><span class=\"p\">,<\/span> <span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">],<\/span> <span class=\"o\">&amp;<\/span><span class=\"nn\">I<\/span><span class=\"p\">::<\/span><span class=\"nn\">XavierUniform<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">(<\/span><span class=\"mf\">1.0<\/span><span class=\"p\">));<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">pb1<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">Parameter<\/span><span class=\"p\">::<\/span><span class=\"nf\">from_initializer<\/span><span class=\"p\">([<\/span><span class=\"n\">NUM_HIDDEN_UNITS<\/span><span class=\"p\">],<\/span> <span class=\"o\">&amp;<\/span><span class=\"nn\">I<\/span><span class=\"p\">::<\/span><span class=\"nn\">Constant<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">(<\/span><span class=\"mf\">0.0<\/span><span class=\"p\">));<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">pw2<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">Parameter<\/span><span class=\"p\">::<\/span><span class=\"nf\">from_initializer<\/span><span class=\"p\">([<\/span><span class=\"n\">NUM_OUTPUT_UNITS<\/span><span class=\"p\">,<\/span> <span class=\"n\">NUM_HIDDEN_UNITS<\/span><span class=\"p\">],<\/span> <span class=\"o\">&amp;<\/span><span class=\"nn\">I<\/span><span class=\"p\">::<\/span><span class=\"nn\">XavierUniform<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">(<\/span><span class=\"mf\">1.0<\/span><span class=\"p\">));<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">pb2<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">Parameter<\/span><span class=\"p\">::<\/span><span class=\"nf\">from_initializer<\/span><span class=\"p\">([<\/span><span class=\"n\">NUM_OUTPUT_UNITS<\/span><span class=\"p\">],<\/span> <span class=\"o\">&amp;<\/span><span class=\"nn\">I<\/span><span class=\"p\">::<\/span><span class=\"nn\">Constant<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">(<\/span><span class=\"mf\">0.0<\/span><span class=\"p\">));<\/span>\r\n\r\n    <span class=\"c\">\/\/ Optimizer\u3078\u306e\u767b\u9332<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">optimizer<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">O<\/span><span class=\"p\">::<\/span><span class=\"nn\">SGD<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">(<\/span><span class=\"mf\">0.5<\/span><span class=\"p\">);<\/span>\r\n    <span class=\"n\">optimizer<\/span><span class=\"nf\">.add_parameter<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">pw1<\/span><span class=\"p\">);<\/span>\r\n    <span class=\"n\">optimizer<\/span><span class=\"nf\">.add_parameter<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">pb1<\/span><span class=\"p\">);<\/span>\r\n    <span class=\"n\">optimizer<\/span><span class=\"nf\">.add_parameter<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">pw2<\/span><span class=\"p\">);<\/span>\r\n    <span class=\"n\">optimizer<\/span><span class=\"nf\">.add_parameter<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">pb2<\/span><span class=\"p\">);<\/span>\r\n<\/code><\/pre>\n<h2>\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9<\/h2>\n<p>primitiv\u3067\u306fGraph\u30aa\u30d6\u30b8\u30a7\u30af\u30c8\u3092\u751f\u6210\u3057\u3001\u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9\u306b\u4f7f\u7528\u3057\u307e\u3059\u3002<br \/>\nGraph\u30aa\u30d6\u30b8\u30a7\u30af\u30c8\u306f\u8a08\u7b97\u30b0\u30e9\u30d5\u3092\u69cb\u7bc9\u3059\u308b\u305f\u3081\u306e\u95a2\u6570\u3092\u547c\u3073\u51fa\u3059\u969b\u306b\u5f15\u6570\u3068\u3057\u3066\u6307\u5b9a\u3057\u307e\u3059\u304c\u3001Graph::set_default()\u306b\u3088\u3063\u3066\u30c7\u30d5\u30a9\u30eb\u30c8\u306e\u30b0\u30e9\u30d5\u3068\u3057\u3066\u767b\u9332\u3057\u3066\u304a\u304f\u3053\u3068\u3067\u3001\u95a2\u6570\u547c\u3073\u51fa\u3057\u306e\u969b\u306f\u7701\u7565\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u300212<\/p>\n<pre class=\"post-pre\"><code>    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">g<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">Graph<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">();<\/span>\r\n    <span class=\"nn\">Graph<\/span><span class=\"p\">::<\/span><span class=\"nf\">set_default<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">g<\/span><span class=\"p\">);<\/span>\r\n<\/code><\/pre>\n<p>\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9\u306ffunctions (\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u4e2d\u3067\u306fF\u3068\u3044\u3046\u30a8\u30a4\u30ea\u30a2\u30b9)\u306b\u5b9a\u7fa9\u3057\u3066\u3044\u308b\u95a2\u6570\u3092\u4f7f\u3063\u3066\u884c\u3044\u307e\u3059\u3002<br \/>\n\u307e\u305a\u3001functions::input()\u3001functions::parameter() \u3092\u7528\u3044\u3066\u5165\u529b\u30c7\u30fc\u30bf\u3001\u30d1\u30e9\u30e1\u30fc\u30bf\u304b\u3089\u305d\u308c\u305e\u308cNode \u30aa\u30d6\u30b8\u30a7\u30af\u30c8\u3092\u751f\u6210\u3057\u307e\u3059\u3002<br \/>\nNode\u306f\u8a08\u7b97\u7d50\u679c\u3092\u793a\u3059\u30aa\u30d6\u30b8\u30a7\u30af\u30c8\u3067\u3001Node\u540c\u58eb\u3067\u76f4\u63a5\u7b97\u8853\u6f14\u7b97\u3092\u884c\u3063\u305f\u308a\u3001functions\u30e2\u30b8\u30e5\u30fc\u30eb\u3067\u5b9a\u7fa9\u3055\u308c\u3066\u3044\u308b\u95a2\u6570\u306b\u5f15\u6570\u3068\u3057\u3066\u6e21\u3057\u305f\u308a\u3059\u308b\u3053\u3068\u3067\u4efb\u610f\u306e\u6f14\u7b97\u3092\u8868\u73fe\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002<br \/>\nNode\u306b\u5bfe\u3059\u308b\u6f14\u7b97\u306f\u6307\u5b9a\u3057\u305fGraph\u306b\u81ea\u52d5\u7684\u306b\u8a18\u9332\u3055\u308c\u308b\u305f\u3081\u3001\u6570\u5f0f\u306b\u6cbf\u3063\u305f\u6f14\u7b97\u3092\u9069\u7528\u3057\u3066\u3044\u304f\u3060\u3051\u3067\u8a08\u7b97\u30b0\u30e9\u30d5\u3092\u69cb\u7bc9\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code>    <span class=\"c\">\/\/ \u5165\u529b\u306e\u30b9\u30e9\u30a4\u30b9\u3092\u53d7\u3051\u53d6\u3063\u3066`Node`\u3092\u8fd4\u3059\u30af\u30ed\u30fc\u30b8\u30e3\u3092\u5b9a\u7fa9<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">make_graph<\/span> <span class=\"o\">=<\/span> <span class=\"p\">|<\/span><span class=\"n\">inputs<\/span><span class=\"p\">:<\/span> <span class=\"o\">&amp;<\/span><span class=\"p\">[<\/span><span class=\"nb\">f32<\/span><span class=\"p\">],<\/span> <span class=\"n\">train<\/span><span class=\"p\">|<\/span> <span class=\"p\">{<\/span>\r\n        <span class=\"k\">let<\/span> <span class=\"n\">x<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">input<\/span><span class=\"p\">(([<\/span><span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">],<\/span> <span class=\"n\">BATCH_SIZE<\/span><span class=\"p\">),<\/span> <span class=\"o\">&amp;<\/span><span class=\"n\">inputs<\/span><span class=\"p\">);<\/span>\r\n        <span class=\"k\">let<\/span> <span class=\"n\">w1<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">parameter<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">pw1<\/span><span class=\"p\">);<\/span>\r\n        <span class=\"k\">let<\/span> <span class=\"n\">b1<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">parameter<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">pb1<\/span><span class=\"p\">);<\/span>\r\n        <span class=\"k\">let<\/span> <span class=\"n\">h<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">relu<\/span><span class=\"p\">(<\/span><span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">matmul<\/span><span class=\"p\">(<\/span><span class=\"n\">w1<\/span><span class=\"p\">,<\/span> <span class=\"n\">x<\/span><span class=\"p\">)<\/span> <span class=\"o\">+<\/span> <span class=\"n\">b1<\/span><span class=\"p\">);<\/span>\r\n        <span class=\"k\">let<\/span> <span class=\"n\">h<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">dropout<\/span><span class=\"p\">(<\/span><span class=\"n\">h<\/span><span class=\"p\">,<\/span> <span class=\"mf\">0.5<\/span><span class=\"p\">,<\/span> <span class=\"n\">train<\/span><span class=\"p\">);<\/span>\r\n        <span class=\"k\">let<\/span> <span class=\"n\">w2<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">parameter<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">pw2<\/span><span class=\"p\">);<\/span>\r\n        <span class=\"k\">let<\/span> <span class=\"n\">b2<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">parameter<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">pb2<\/span><span class=\"p\">);<\/span>\r\n        <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">matmul<\/span><span class=\"p\">(<\/span><span class=\"n\">w2<\/span><span class=\"p\">,<\/span> <span class=\"n\">h<\/span><span class=\"p\">)<\/span> <span class=\"o\">+<\/span> <span class=\"n\">b2<\/span>\r\n    <span class=\"p\">};<\/span>\r\n<\/code><\/pre>\n<h2>\u5b66\u7fd2\u30eb\u30fc\u30d7 (forward, backward, update) \u306e\u5b9f\u884c<\/h2>\n<p>\u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u521d\u671f\u5316\u3001\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9\u3001forward\u6f14\u7b97\u3001backward\u6f14\u7b97\u3001\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u66f4\u65b0\u306f\u6b21\u306e\u3088\u3046\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"n\">g<\/span><span class=\"nf\">.clear<\/span><span class=\"p\">();<\/span>  <span class=\"c\">\/\/ \u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u521d\u671f\u5316<\/span>\r\n\r\n<span class=\"k\">let<\/span> <span class=\"n\">y<\/span> <span class=\"o\">=<\/span> <span class=\"nf\">make_graph<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"n\">inputs<\/span><span class=\"p\">,<\/span> <span class=\"k\">true<\/span><span class=\"p\">);<\/span>  <span class=\"c\">\/\/ \u30b0\u30e9\u30d5\u306e\u69cb\u7bc9 (\u30c7\u30d5\u30a9\u30eb\u30c8\u306e`Graph`\u306b`Node`\u3092\u8ffd\u52a0)<\/span>\r\n<span class=\"k\">let<\/span> <span class=\"n\">loss<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">softmax_cross_entropy<\/span><span class=\"p\">(<\/span><span class=\"n\">y<\/span><span class=\"p\">,<\/span> <span class=\"o\">&amp;<\/span><span class=\"n\">labels<\/span><span class=\"p\">,<\/span> <span class=\"mi\">0<\/span><span class=\"p\">);<\/span>  <span class=\"c\">\/\/ \u30ed\u30b9\u306e`Node`\u3092\u8ffd\u52a0<\/span>\r\n<span class=\"k\">let<\/span> <span class=\"n\">avg_loss<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nn\">batch<\/span><span class=\"p\">::<\/span><span class=\"nf\">mean<\/span><span class=\"p\">(<\/span><span class=\"n\">loss<\/span><span class=\"p\">);<\/span>\r\n\r\n<span class=\"k\">let<\/span> <span class=\"n\">loss_val<\/span> <span class=\"o\">=<\/span> <span class=\"n\">avg_loss<\/span><span class=\"nf\">.to_float<\/span><span class=\"p\">();<\/span>  <span class=\"c\">\/\/ \u8a08\u7b97\u7d50\u679c\u3092\u53d6\u5f97<\/span>\r\n<span class=\"nd\">println!<\/span><span class=\"p\">(<\/span><span class=\"s\">\"  loss: {}\"<\/span><span class=\"p\">,<\/span> <span class=\"n\">loss_val<\/span><span class=\"p\">);<\/span>\r\n\r\n<span class=\"n\">optimizer<\/span><span class=\"nf\">.reset_gradients<\/span><span class=\"p\">();<\/span>  <span class=\"c\">\/\/ \u52fe\u914d\u306e\u30ea\u30bb\u30c3\u30c8<\/span>\r\n<span class=\"n\">avg_loss<\/span><span class=\"nf\">.backward<\/span><span class=\"p\">();<\/span>  <span class=\"c\">\/\/ backward\u6f14\u7b97<\/span>\r\n<span class=\"n\">optimizer<\/span><span class=\"nf\">.update<\/span><span class=\"p\">();<\/span>  <span class=\"c\">\/\/ \u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u66f4\u65b0<\/span>\r\n<\/code><\/pre>\n<p>Define by Run\u65b9\u5f0f\u306eChainer\u3084PyTorch\u3068\u5927\u304d\u304f\u7570\u306a\u308b\u70b9\u306f\u3001Chainer\u3084PyTorch\u304c\u8a08\u7b97\u30b0\u30e9\u30d5\u4e0a\u306e\u30ce\u30fc\u30c9\u306e\u8ffd\u52a0\u306e\u969b\u306bforward\u306e\u8a08\u7b97\u7d50\u679c\u3092\u5373\u6642\u306b\u6f14\u7b97\u3059\u308b\u306e\u306b\u5bfe\u3057\u3001primitiv\u3084DyNet\u3067\u306fforward\u306e\u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9\u6642\u306b\u5024\u306e\u8a08\u7b97\u3092\u884c\u308f\u305a\u3001\u5177\u4f53\u7684\u306a\u8a08\u7b97\u7d50\u679c\u304c\u5fc5\u8981\u306b\u306a\u3063\u305f\u3068\u304d (\u4e0a\u8a18\u306e\u4f8b\u3067\u306fto_float()\u306e\u547c\u3073\u51fa\u3057\u6642) \u306b\u3001\u521d\u3081\u3066\u5024\u306e\u8a08\u7b97\u3092\u884c\u3046\u3068\u3044\u3046\u9045\u5ef6\u8a55\u4fa1\u3092\u63a1\u7528\u3057\u3066\u3044\u308b\u3068\u3053\u308d\u3067\u3059\u3002<\/p>\n<p>\u5b66\u7fd2\u30eb\u30fc\u30d7\u5168\u4f53\u306f\u4e0b\u8a18\u306e\u3088\u3046\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code>    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">rng<\/span> <span class=\"o\">=<\/span> <span class=\"nf\">thread_rng<\/span><span class=\"p\">();<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">ids<\/span><span class=\"p\">:<\/span> <span class=\"nb\">Vec<\/span><span class=\"o\">&lt;<\/span><span class=\"nb\">usize<\/span><span class=\"o\">&gt;<\/span> <span class=\"o\">=<\/span> <span class=\"p\">(<\/span><span class=\"mi\">0u<\/span><span class=\"n\">size<\/span><span class=\"o\">..<\/span><span class=\"n\">NUM_TRAIN_SAMPLES<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">)<\/span><span class=\"nf\">.collect<\/span><span class=\"p\">();<\/span>\r\n\r\n    <span class=\"k\">for<\/span> <span class=\"n\">epoch<\/span> <span class=\"n\">in<\/span> <span class=\"mi\">0<\/span><span class=\"o\">..<\/span><span class=\"n\">MAX_EPOCH<\/span> <span class=\"p\">{<\/span>\r\n        <span class=\"n\">rng<\/span><span class=\"nf\">.shuffle<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">ids<\/span><span class=\"p\">);<\/span>  <span class=\"c\">\/\/ \u30c7\u30fc\u30bf\u306e\u30a4\u30f3\u30c7\u30c3\u30af\u30b9\u306e\u30b7\u30e3\u30c3\u30d5\u30eb<\/span>\r\n\r\n        <span class=\"k\">for<\/span> <span class=\"n\">batch<\/span> <span class=\"n\">in<\/span> <span class=\"mi\">0<\/span><span class=\"o\">..<\/span><span class=\"n\">NUM_TRAIN_BATCHES<\/span> <span class=\"p\">{<\/span>\r\n            <span class=\"nd\">print!<\/span><span class=\"p\">(<\/span><span class=\"s\">\"<\/span><span class=\"se\">\\r<\/span><span class=\"s\">Training... {} \/ {}\"<\/span><span class=\"p\">,<\/span> <span class=\"n\">batch<\/span> <span class=\"o\">+<\/span> <span class=\"mi\">1<\/span><span class=\"p\">,<\/span> <span class=\"n\">NUM_TRAIN_BATCHES<\/span><span class=\"p\">);<\/span>\r\n            <span class=\"c\">\/\/ \u30df\u30cb\u30d0\u30c3\u30c1\u306e\u30c7\u30fc\u30bf\u306e\u53d6\u308a\u51fa\u3057<\/span>\r\n            <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">inputs<\/span><span class=\"p\">:<\/span> <span class=\"nb\">Vec<\/span><span class=\"o\">&lt;<\/span><span class=\"nb\">f32<\/span><span class=\"o\">&gt;<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">Vec<\/span><span class=\"p\">::<\/span><span class=\"nf\">with_capacity<\/span><span class=\"p\">((<\/span><span class=\"n\">BATCH_SIZE<\/span> <span class=\"o\">*<\/span> <span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">)<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">);<\/span>\r\n            <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">labels<\/span><span class=\"p\">:<\/span> <span class=\"nb\">Vec<\/span><span class=\"o\">&lt;<\/span><span class=\"nb\">u32<\/span><span class=\"o\">&gt;<\/span> <span class=\"o\">=<\/span> <span class=\"nd\">vec!<\/span><span class=\"p\">[<\/span><span class=\"mi\">0<\/span><span class=\"p\">;<\/span> <span class=\"n\">BATCH_SIZE<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">];<\/span>\r\n            <span class=\"k\">for<\/span> <span class=\"n\">i<\/span> <span class=\"n\">in<\/span> <span class=\"mi\">0<\/span><span class=\"o\">..<\/span><span class=\"n\">BATCH_SIZE<\/span> <span class=\"p\">{<\/span>\r\n                <span class=\"k\">let<\/span> <span class=\"n\">id<\/span> <span class=\"o\">=<\/span> <span class=\"n\">ids<\/span><span class=\"p\">[(<\/span><span class=\"n\">i<\/span> <span class=\"o\">+<\/span> <span class=\"n\">batch<\/span> <span class=\"o\">*<\/span> <span class=\"n\">BATCH_SIZE<\/span><span class=\"p\">)<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">];<\/span>\r\n                <span class=\"k\">let<\/span> <span class=\"n\">from<\/span> <span class=\"o\">=<\/span> <span class=\"n\">id<\/span> <span class=\"o\">*<\/span> <span class=\"n\">NUM_INPUT_UNITS<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">;<\/span>\r\n                <span class=\"k\">let<\/span> <span class=\"n\">to<\/span> <span class=\"o\">=<\/span> <span class=\"p\">(<\/span><span class=\"n\">id<\/span> <span class=\"o\">+<\/span> <span class=\"mi\">1<\/span><span class=\"p\">)<\/span> <span class=\"o\">*<\/span> <span class=\"n\">NUM_INPUT_UNITS<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">;<\/span>\r\n                <span class=\"n\">inputs<\/span><span class=\"nf\">.extend_from_slice<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"n\">train_inputs<\/span><span class=\"p\">[<\/span><span class=\"n\">from<\/span><span class=\"o\">..<\/span><span class=\"n\">to<\/span><span class=\"p\">]);<\/span>\r\n                <span class=\"n\">labels<\/span><span class=\"p\">[<\/span><span class=\"n\">i<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">]<\/span> <span class=\"o\">=<\/span> <span class=\"n\">train_labels<\/span><span class=\"p\">[<\/span><span class=\"n\">id<\/span><span class=\"p\">]<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">u32<\/span><span class=\"p\">;<\/span>\r\n            <span class=\"p\">}<\/span>\r\n\r\n            <span class=\"c\">\/\/ \u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u521d\u671f\u5316<\/span>\r\n            <span class=\"n\">g<\/span><span class=\"nf\">.clear<\/span><span class=\"p\">();<\/span>\r\n\r\n            <span class=\"c\">\/\/ \u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9<\/span>\r\n            <span class=\"k\">let<\/span> <span class=\"n\">y<\/span> <span class=\"o\">=<\/span> <span class=\"nf\">make_graph<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"n\">inputs<\/span><span class=\"p\">,<\/span> <span class=\"k\">true<\/span><span class=\"p\">);<\/span>\r\n            <span class=\"k\">let<\/span> <span class=\"n\">loss<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">softmax_cross_entropy<\/span><span class=\"p\">(<\/span><span class=\"n\">y<\/span><span class=\"p\">,<\/span> <span class=\"o\">&amp;<\/span><span class=\"n\">labels<\/span><span class=\"p\">,<\/span> <span class=\"mi\">0<\/span><span class=\"p\">);<\/span>\r\n            <span class=\"k\">let<\/span> <span class=\"n\">avg_loss<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nn\">batch<\/span><span class=\"p\">::<\/span><span class=\"nf\">mean<\/span><span class=\"p\">(<\/span><span class=\"n\">loss<\/span><span class=\"p\">);<\/span>\r\n\r\n            <span class=\"c\">\/\/ \u52fe\u914d\u306e\u521d\u671f\u5316, backward, \u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u66f4\u65b0 \u203b\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u6642\u306e\u307f<\/span>\r\n            <span class=\"n\">optimizer<\/span><span class=\"nf\">.reset_gradients<\/span><span class=\"p\">();<\/span>\r\n            <span class=\"n\">avg_loss<\/span><span class=\"nf\">.backward<\/span><span class=\"p\">();<\/span>\r\n            <span class=\"n\">optimizer<\/span><span class=\"nf\">.update<\/span><span class=\"p\">();<\/span>\r\n        <span class=\"p\">}<\/span>\r\n\r\n        <span class=\"nd\">println!<\/span><span class=\"p\">();<\/span>\r\n\r\n        <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">match_<\/span> <span class=\"o\">=<\/span> <span class=\"mi\">0<\/span><span class=\"p\">;<\/span>\r\n\r\n        <span class=\"k\">for<\/span> <span class=\"n\">batch<\/span> <span class=\"n\">in<\/span> <span class=\"mi\">0<\/span><span class=\"o\">..<\/span><span class=\"n\">NUM_TEST_BATCHES<\/span> <span class=\"p\">{<\/span>\r\n            <span class=\"nd\">print!<\/span><span class=\"p\">(<\/span><span class=\"s\">\"<\/span><span class=\"se\">\\r<\/span><span class=\"s\">Testing... {} \/ {}\"<\/span><span class=\"p\">,<\/span> <span class=\"n\">batch<\/span> <span class=\"o\">+<\/span> <span class=\"mi\">1<\/span><span class=\"p\">,<\/span> <span class=\"n\">NUM_TEST_BATCHES<\/span><span class=\"p\">);<\/span>\r\n            <span class=\"c\">\/\/ \u8a55\u4fa1\u30c7\u30fc\u30bf\u306e\u53d6\u308a\u51fa\u3057<\/span>\r\n            <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">inputs<\/span><span class=\"p\">:<\/span> <span class=\"nb\">Vec<\/span><span class=\"o\">&lt;<\/span><span class=\"nb\">f32<\/span><span class=\"o\">&gt;<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">Vec<\/span><span class=\"p\">::<\/span><span class=\"nf\">with_capacity<\/span><span class=\"p\">((<\/span><span class=\"n\">BATCH_SIZE<\/span> <span class=\"o\">*<\/span> <span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">)<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">);<\/span>\r\n            <span class=\"k\">let<\/span> <span class=\"n\">from<\/span> <span class=\"o\">=<\/span> <span class=\"p\">(<\/span><span class=\"n\">batch<\/span> <span class=\"o\">*<\/span> <span class=\"n\">BATCH_SIZE<\/span> <span class=\"o\">*<\/span> <span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">)<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">;<\/span>\r\n            <span class=\"k\">let<\/span> <span class=\"n\">to<\/span> <span class=\"o\">=<\/span> <span class=\"p\">((<\/span><span class=\"n\">batch<\/span> <span class=\"o\">+<\/span> <span class=\"mi\">1<\/span><span class=\"p\">)<\/span> <span class=\"o\">*<\/span> <span class=\"n\">BATCH_SIZE<\/span> <span class=\"o\">*<\/span> <span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">)<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">;<\/span>\r\n            <span class=\"n\">inputs<\/span><span class=\"nf\">.extend_from_slice<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"n\">test_inputs<\/span><span class=\"p\">[<\/span><span class=\"n\">from<\/span><span class=\"o\">..<\/span><span class=\"n\">to<\/span><span class=\"p\">]);<\/span>\r\n\r\n            <span class=\"c\">\/\/ \u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u521d\u671f\u5316<\/span>\r\n            <span class=\"n\">g<\/span><span class=\"nf\">.clear<\/span><span class=\"p\">();<\/span>\r\n\r\n            <span class=\"c\">\/\/ \u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9<\/span>\r\n            <span class=\"k\">let<\/span> <span class=\"n\">y<\/span> <span class=\"o\">=<\/span> <span class=\"nf\">make_graph<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"n\">inputs<\/span><span class=\"p\">,<\/span> <span class=\"k\">false<\/span><span class=\"p\">);<\/span>\r\n\r\n            <span class=\"c\">\/\/ \u8a08\u7b97\u7d50\u679c\u306e\u7b97\u51fa<\/span>\r\n            <span class=\"k\">let<\/span> <span class=\"n\">y_val<\/span> <span class=\"o\">=<\/span> <span class=\"n\">y<\/span><span class=\"nf\">.to_vector<\/span><span class=\"p\">();<\/span>\r\n            <span class=\"c\">\/\/ \u8a55\u4fa1<\/span>\r\n            <span class=\"k\">for<\/span> <span class=\"n\">i<\/span> <span class=\"n\">in<\/span> <span class=\"mi\">0<\/span><span class=\"o\">..<\/span><span class=\"n\">BATCH_SIZE<\/span> <span class=\"p\">{<\/span>\r\n                <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">maxval<\/span> <span class=\"o\">=<\/span> <span class=\"o\">-<\/span><span class=\"mf\">1e10<\/span><span class=\"p\">;<\/span>\r\n                <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">argmax<\/span><span class=\"p\">:<\/span> <span class=\"nb\">i32<\/span> <span class=\"o\">=<\/span> <span class=\"o\">-<\/span><span class=\"mi\">1<\/span><span class=\"p\">;<\/span>\r\n                <span class=\"k\">for<\/span> <span class=\"n\">j<\/span> <span class=\"n\">in<\/span> <span class=\"mi\">0<\/span><span class=\"o\">..<\/span><span class=\"n\">NUM_OUTPUT_UNITS<\/span> <span class=\"p\">{<\/span>\r\n                    <span class=\"k\">let<\/span> <span class=\"n\">v<\/span> <span class=\"o\">=<\/span> <span class=\"n\">y_val<\/span><span class=\"p\">[(<\/span><span class=\"n\">j<\/span> <span class=\"o\">+<\/span> <span class=\"n\">i<\/span> <span class=\"o\">*<\/span> <span class=\"n\">NUM_OUTPUT_UNITS<\/span><span class=\"p\">)<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">];<\/span>\r\n                    <span class=\"k\">if<\/span> <span class=\"n\">v<\/span> <span class=\"o\">&gt;<\/span> <span class=\"n\">maxval<\/span> <span class=\"p\">{<\/span>\r\n                        <span class=\"n\">maxval<\/span> <span class=\"o\">=<\/span> <span class=\"n\">v<\/span><span class=\"p\">;<\/span>\r\n                        <span class=\"n\">argmax<\/span> <span class=\"o\">=<\/span> <span class=\"n\">j<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">i32<\/span><span class=\"p\">;<\/span>\r\n                    <span class=\"p\">}<\/span>\r\n                <span class=\"p\">}<\/span>\r\n                <span class=\"k\">if<\/span> <span class=\"n\">argmax<\/span> <span class=\"o\">==<\/span> <span class=\"n\">test_labels<\/span><span class=\"p\">[(<\/span><span class=\"n\">i<\/span> <span class=\"o\">+<\/span> <span class=\"n\">batch<\/span> <span class=\"o\">*<\/span> <span class=\"n\">BATCH_SIZE<\/span><span class=\"p\">)<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">]<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">i32<\/span> <span class=\"p\">{<\/span>\r\n                    <span class=\"n\">match_<\/span> <span class=\"o\">+=<\/span> <span class=\"mi\">1<\/span><span class=\"p\">;<\/span>\r\n                <span class=\"p\">}<\/span>\r\n            <span class=\"p\">}<\/span>\r\n        <span class=\"p\">}<\/span>\r\n\r\n        <span class=\"k\">let<\/span> <span class=\"n\">accuracy<\/span> <span class=\"o\">=<\/span> <span class=\"mf\">100.0<\/span> <span class=\"o\">*<\/span> <span class=\"n\">match_<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">f32<\/span> <span class=\"o\">\/<\/span> <span class=\"n\">NUM_TEST_SAMPLES<\/span> <span class=\"k\">as<\/span> <span class=\"nb\">f32<\/span><span class=\"p\">;<\/span>\r\n        <span class=\"nd\">println!<\/span><span class=\"p\">(<\/span><span class=\"s\">\"<\/span><span class=\"se\">\\n<\/span><span class=\"s\">epoch {}: accuracy: {:.2}%\"<\/span><span class=\"p\">,<\/span> <span class=\"n\">epoch<\/span><span class=\"p\">,<\/span> <span class=\"n\">accuracy<\/span><span class=\"p\">);<\/span>\r\n    <span class=\"p\">}<\/span>\r\n<\/code><\/pre>\n<p>\u30bd\u30fc\u30b9\u30b3\u30fc\u30c9\u5168\u4f53\u306f\u4e0b\u8a18\u306b\u7f6e\u3044\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>primitiv-rust\/mnist.rs &#8211; GitHub<\/p>\n<h2>GPU\u4e0a\u3067\u306e\u5b9f\u884c<\/h2>\n<p>CUDA\u3084OpenCL\u3092\u4f7f\u3063\u305fGPU\u4e0a\u3067\u306e\u6f14\u7b97\u306f\u3001\u4f7f\u7528\u3059\u308b\u30c7\u30d0\u30a4\u30b9\u3092\u5207\u308a\u66ff\u3048\u308b\u3060\u3051\u3067\u53ef\u80fd\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code>    <span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">dev<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">D<\/span><span class=\"p\">::<\/span><span class=\"nn\">CUDA<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">(<\/span><span class=\"mi\">0<\/span><span class=\"p\">);<\/span>\r\n    <span class=\"c\">\/\/ let mut dev = D::Naive::new();<\/span>\r\n    <span class=\"c\">\/\/ let mut dev = D::OpenCL::new(0, 1);<\/span>\r\n    <span class=\"c\">\/\/ let mut dev = D::Eigen::new();<\/span>\r\n    <span class=\"nn\">device<\/span><span class=\"p\">::<\/span><span class=\"nf\">set_default<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">dev<\/span><span class=\"p\">);<\/span>\r\n<\/code><\/pre>\n<h2>\u8907\u6570\u30c7\u30d0\u30a4\u30b9\u3092\u4f7f\u7528\u3057\u305f\u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9\u30fb\u6f14\u7b97<\/h2>\n<p>Parameter\u306e\u751f\u6210\u3084\u5165\u529b\u306b\u4f7f\u7528\u3059\u308bNode\u306e\u69cb\u7bc9\u6642\u306b\u660e\u793a\u7684\u306b\u30c7\u30d0\u30a4\u30b9\u3092\u6307\u5b9a\u3059\u308b\u3053\u3068\u3067\u3001\u8907\u6570\u306e\u30c7\u30d0\u30a4\u30b9\u3092\u4f7f\u7528\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">dev0<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">D<\/span><span class=\"p\">::<\/span><span class=\"nn\">CUDA<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">(<\/span><span class=\"mi\">0<\/span><span class=\"p\">);<\/span>\r\n<span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">dev1<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">D<\/span><span class=\"p\">::<\/span><span class=\"nn\">CUDA<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">(<\/span><span class=\"mi\">1<\/span><span class=\"p\">);<\/span>\r\n\r\n<span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">pw1<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">Parameter<\/span><span class=\"p\">::<\/span><span class=\"nf\">from_initializer_with_device<\/span><span class=\"p\">(<\/span>\r\n    <span class=\"p\">[<\/span><span class=\"n\">NUM_HIDDEN_UNITS<\/span><span class=\"p\">,<\/span> <span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">],<\/span>\r\n    <span class=\"o\">&amp;<\/span><span class=\"nn\">I<\/span><span class=\"p\">::<\/span><span class=\"nn\">XavierUniform<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">(<\/span><span class=\"mf\">1.0<\/span><span class=\"p\">),<\/span>\r\n    <span class=\"nf\">Some<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">dev0<\/span><span class=\"p\">),<\/span>\r\n<span class=\"p\">);<\/span>\r\n<span class=\"k\">let<\/span> <span class=\"k\">mut<\/span> <span class=\"n\">pw2<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">Parameter<\/span><span class=\"p\">::<\/span><span class=\"nf\">from_initializer_with_device<\/span><span class=\"p\">(<\/span>\r\n    <span class=\"p\">[<\/span><span class=\"n\">NUM_HIDDEN_UNITS<\/span><span class=\"p\">,<\/span> <span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">],<\/span>\r\n    <span class=\"o\">&amp;<\/span><span class=\"nn\">I<\/span><span class=\"p\">::<\/span><span class=\"nn\">XavierUniform<\/span><span class=\"p\">::<\/span><span class=\"nf\">new<\/span><span class=\"p\">(<\/span><span class=\"mf\">1.0<\/span><span class=\"p\">),<\/span>\r\n    <span class=\"nf\">Some<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">dev1<\/span><span class=\"p\">),<\/span>\r\n<span class=\"p\">);<\/span>\r\n\r\n<span class=\"o\">...<\/span>\r\n\r\n<span class=\"k\">let<\/span> <span class=\"n\">x1<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">input_with_device<\/span><span class=\"p\">(([<\/span><span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">],<\/span> <span class=\"n\">BATCH_SIZE<\/span><span class=\"p\">),<\/span> <span class=\"o\">&amp;<\/span><span class=\"n\">inputs1<\/span><span class=\"p\">,<\/span> <span class=\"nf\">Some<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">dev0<\/span><span class=\"p\">));<\/span>\r\n<span class=\"k\">let<\/span> <span class=\"n\">x2<\/span> <span class=\"o\">=<\/span> <span class=\"nn\">F<\/span><span class=\"p\">::<\/span><span class=\"nf\">input_with_device<\/span><span class=\"p\">(([<\/span><span class=\"n\">NUM_INPUT_UNITS<\/span><span class=\"p\">],<\/span> <span class=\"n\">BATCH_SIZE<\/span><span class=\"p\">),<\/span> <span class=\"o\">&amp;<\/span><span class=\"n\">inputs2<\/span><span class=\"p\">,<\/span> <span class=\"nf\">Some<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"n\">dev1<\/span><span class=\"p\">));<\/span>\r\n<\/code><\/pre>\n<h1>Future Work<\/h1>\n<p>primitiv\u306e\u30b3\u30a2\u306e\u6a5f\u80fd\u306e\u958b\u767a\u3068\u3057\u3066\u306f\u4e0b\u8a18\u3092\u4e88\u5b9a\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u30de\u30eb\u30c1\u30b9\u30ec\u30c3\u30c9\u3067\u306e\u6f14\u7b97\u306e\u5bfe\u5fdc<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">Define by Run + \u9045\u5ef6\u8a55\u4fa1 \u3092\u5229\u7528\u3057\u305f\u30df\u30cb\u30d0\u30c3\u30c1\u8a08\u7b97\u306e\u6700\u9069\u5316 (autobatch)<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u9ad8\u968e\u5fae\u5206<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">Convolution, Pooling\u6f14\u7b97\u306e\u5b9f\u88c5 13<\/ul>\n<p>Rust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u306fC API\u304cunstable\u306a\u305f\u3081\u958b\u767a\u304c\u9045\u308c\u3066\u3044\u307e\u3059\u304c\u3001\u3053\u308c\u304b\u3089\u9806\u6b21\u5bfe\u5fdc\u3092\u9032\u3081\u3066\u3044\u304d\u307e\u3059\u3002<br \/>\n(\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5\u3084\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u306e\u6574\u5099\u3001\u30c6\u30b9\u30c8\u30b9\u30af\u30ea\u30d7\u30c8\u306e\u4f5c\u6210\u7b49\u3044\u308d\u3044\u308d\u9593\u306b\u5408\u308f\u305a\u3001\u5927\u5909\u7533\u3057\u8a33\u306a\u3044\u3067\u3059\u304c\u8208\u5473\u3092\u6301\u3063\u305f\u4eba\u304c\u3059\u3050\u306b\u8a66\u305b\u308b\u72b6\u6cc1\u306b\u81f3\u3063\u3066\u3044\u307e\u305b\u3093\u3002)<\/p>\n<p>\u4ed6\u306e\u8a00\u8a9e\u306e\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u3068\u3057\u3066\u306fPython\u3067primitiv version 0.3\u306e\u6a5f\u80fd\u304c\u4f7f\u3048\u308b\u3088\u3046\u306b\u306a\u3063\u3066\u3044\u307e\u3059\u3002<br \/>\n\u3053\u308c\u304b\u3089Java\u306e\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u3092\u958b\u767a\u3057\u3001Kotlin\u3084Scala\u304b\u3089\u3082\u4f7f\u3048\u308b\u3088\u3046\u306b\u9032\u3081\u3066\u3044\u304f\u4e88\u5b9a\u3067\u3059\u3002<\/p>\n<h1>\u307e\u3068\u3081<\/h1>\n<p>\u672c\u8a18\u4e8b\u3067\u306f\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af primitiv \u306eRust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u306e\u6982\u8981\u3068\u3001Rust\u30d0\u30a4\u30f3\u30c7\u30a3\u30f3\u30b0\u3092\u4f7f\u3063\u305f\u30e2\u30c7\u30eb\u306e\u69cb\u7bc9\u30fb\u5b66\u7fd2\u65b9\u6cd5\u306b\u3064\u3044\u3066\u7d39\u4ecb\u3057\u307e\u3057\u305f\u3002<br \/>\nprimitiv\u306f\u7570\u306a\u308b\u30d7\u30e9\u30c3\u30c8\u30d5\u30a9\u30fc\u30e0\u306e\u30c7\u30d0\u30a4\u30b9\u3092\u4f7f\u3044\u5206\u3051\u3067\u304d\u308b\u9045\u5ef6\u8a55\u4fa1\u578b\u306eDefine by Run\u65b9\u5f0f\u306e\u8a08\u7b97\u30b0\u30e9\u30d5\u69cb\u7bc9\u30fb\u6f14\u7b97\u3092\u7279\u5fb4\u3068\u3059\u308b\u65b0\u8208\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3059\u3002<br \/>\n\u5f8c\u767a\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3059\u304c\u65e2\u5b58\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306e\u826f\u3044\u3068\u3053\u308d\u3092\u53d6\u308a\u5165\u308c\u3001\u6b63\u5f0f\u306a\u30ea\u30ea\u30fc\u30b9\u306b\u5411\u3051\u3066\u6a5f\u80fd\u306e\u62e1\u5145\u3092\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">primitiv\/primitiv: A Neural Network Toolkit. &#8211; GitHub<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">primitiv\/primitiv-python: Python binding of primitiv. &#8211; GitHub<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">primitiv\/primitiv-rust: Rust binding of primitiv. &#8211; GitHub<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">primitiv\/primitiv-java: Java binding of primitiv. &#8211; GitHub<\/ul>\n<p>primitiv developer\u30c1\u30fc\u30e0\u306f\u958b\u767a\u306b\u5354\u529b\u3057\u3066\u304f\u308c\u308b\u4eba\u3092\u52df\u96c6\u3057\u3066\u3044\u307e\u3059\u3002<br \/>\n\u8208\u5473\u306e\u3042\u308b\u65b9\u306f @odashi_t \u3055\u3093\u3001\u307e\u305f\u306f @chantera \u307e\u3067\u3054\u9023\u7d61\u304f\u3060\u3055\u3044\u3002<\/p>\n<h1>\u8ffd\u8a18<\/h1>\n<h2>2017\u5e7412\u670825\u65e5<\/h2>\n<p>\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5\u306b\u3064\u3044\u3066\u8a18\u8f09\u3057\u307e\u3057\u305f\u3002<br \/>\nhttps:\/\/qiita.com\/odashi_t\/items\/4b90efc20b09c30e58bf<\/p>\n<p>Leaf\u306e\u5f8c\u7d99 Juice \u306b\u3064\u3044\u3066\u8a18\u8f09\u3057\u307e\u3057\u305f\u3002<br \/>\n@fx-kirin \u3055\u3093\u3001\u60c5\u5831\u63d0\u4f9b\u3042\u308a\u304c\u3068\u3046\u3054\u3056\u3044\u307e\u3059\u3002<\/p>\n<p>@odashi_t \u3055\u3093\u304cprimitiv\u306e\u89e3\u8aac\u8a18\u4e8b\u3092\u66f8\u3044\u3066\u304f\u308c\u307e\u3057\u305f\u3002<br \/>\nprimitiv: \u65b0\u3057\u3044\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306e\u30e9\u30a4\u30d6\u30e9\u30ea<\/p>\n<hr \/>\n<div>\n<p>MILA and the future of Theano\u00a0\u21a9<\/p>\n<p>\u73fe\u5728\u3067\u306fCPU\u3067\u306e\u884c\u5217\u6f14\u7b97\u306e\u30e9\u30a4\u30d6\u30e9\u30ea\u3068\u3057\u3066rust-ndarray\u306a\u3069\u304c\u5229\u7528\u53ef\u80fd\u3002\u00a0\u21a9<\/p>\n<p>hobofan 771 days ago, \u8a18\u4e8b: Leaf: machine intelligence framework in Rust | Hacker News\u00a0\u21a9<\/p>\n<p>Collenchyma: CUDA, OpenCL and Native Machine Intelligence for Leaf : rust\u00a0\u21a9<\/p>\n<p>Tensorflow wins \u2013 Michael Hirn \u2013 Medium\u00a0\u21a9<\/p>\n<p>Leaf \u306e\u5f8c\u7d99\u3082\u3042\u308b\u3089\u3057\u3044\u3067\u3059\u3002 spearow\/juice &#8211; GitHub\u00a0\u21a9<\/p>\n<p>dynet-rs\/bilstmtagger.rs &#8211; GitHub\u00a0\u21a9<\/p>\n<p>rust-bindgen\u306eC++\u5bfe\u5fdc\u304c\u5b8c\u5168\u3067\u306f\u306a\u3044\u305f\u3081\u30fb\u00a0\u21a9<\/p>\n<p>https:\/\/github.com\/chantera\/dynet-rs\/issues\/1\u00a0\u21a9<\/p>\n<p>NAIST\u306f\u5948\u826f\u306e\u5c71\u5965\u306b\u4f4d\u7f6e\u3057\u3066\u3044\u3066\u5a2f\u697d\u304c\u306a\u3044\u306e\u3067\u3001\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3092\u4f5c\u308b\u3050\u3089\u3044\u3057\u304b\u3084\u308b\u3053\u3068\u304c\u306a\u3044\u3068\u3044\u3046\u4eba\u304c\u5272\u3068\u3044\u307e\u3059\u3002\u00a0\u21a9<\/p>\n<p>\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3067\u306f\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u521d\u671f\u5316\u306f\u5b66\u7fd2\u7d50\u679c\u306b\u5927\u304d\u304f\u5f71\u97ff\u3092\u53ca\u307c\u3059\u8981\u56e0\u3067\u3059\u304c\u3001\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u5074\u3067\u30c7\u30d5\u30a9\u30eb\u30c8\u306e\u521d\u671f\u5316\u65b9\u6cd5\u3092\u6c7a\u3081\u3066\u3057\u307e\u3046\u3068\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u521d\u671f\u5316\u304c\u958b\u767a\u8005\u306b\u8efd\u8996\u3055\u308c\u308b\u3060\u3051\u3067\u306a\u304f\u3001\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306e\u30bd\u30fc\u30b9\u30b3\u30fc\u30c9\u3092\u8aad\u3093\u3067\u4ed5\u69d8\u3092\u628a\u63e1\u3059\u308b\u5fc5\u8981\u304c\u3042\u308b\uff08\u6700\u60aa\u306e\u5834\u5408\u306f\u30c7\u30d5\u30a9\u30eb\u30c8\u306e\u521d\u671f\u5316\u65b9\u6cd5\u304c\u5909\u66f4\u3055\u308c\u3066\u3057\u307e\u3046\uff09\u305f\u3081\u3001\u500b\u4eba\u7684\u306b\u306f\u73fe\u72b6\u306eprimitiv\u306e\u3088\u3046\u306bInitializer\u3092\u6307\u5b9a\u3059\u308b\u30a4\u30f3\u30bf\u30fc\u30d5\u30a7\u30fc\u30b9\u304c\u826f\u3044\u3068\u601d\u3044\u307e\u3059\u3002\u00a0\u21a9<\/p>\n<p>\u95a2\u6570\u547c\u3073\u51fa\u3057\u306e\u969b\u306b\u660e\u793a\u7684\u306b\u30b0\u30e9\u30d5\u3092\u6e21\u3059\u3088\u3046\u306a\u72b6\u6cc1\u3068\u3057\u3066\u3001\u8907\u6570\u306e\u30b0\u30e9\u30d5\u3092\u4f7f\u3063\u305f\u30de\u30eb\u30c1\u30b9\u30ec\u30c3\u30c9\u3067\u306e\u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9\u30fb\u6f14\u7b97\u3092\u60f3\u5b9a\u3057\u3066\u3044\u307e\u3059\uff08\u30de\u30eb\u30c1\u30b9\u30ec\u30c3\u30c9\u3067\u306e\u8a08\u7b97\u30b0\u30e9\u30d5\u306e\u69cb\u7bc9\u30fb\u6f14\u7b97\u306e\u6a5f\u80fd\u306f\u73fe\u5728\u958b\u767a\u4e2d\u3067\u3059\uff09\u3002\u00a0\u21a9<\/p>\n<p>\u958b\u767a\u9663\u304c\u81ea\u7136\u8a00\u8a9e\u51e6\u7406\u95a2\u4fc2\u306e\u5c02\u9580\u306e\u4eba\u304c\u591a\u6570\u306e\u305f\u3081\u3001\u958b\u767a\u306e\u512a\u5148\u5ea6\u304c\u4f4e\u3044\u3067\u3059\u3002\u753b\u50cf\u95a2\u4fc2\u3067\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306b\u8a73\u3057\u3044\u958b\u767a\u8005\u30fb\u7814\u7a76\u8005\u304b\u3089\u306e\u3054\u610f\u898b\u30fb\u3054\u5354\u529b\u3092\u304a\u9858\u3044\u3057\u307e\u3059\u3002\u00a0\u21a9<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u672c\u8a18\u4e8b\u306fRust Advent Calendar 2017\u306e12\u670821\u65e5\u306e\u8a18\u4e8b\u3067\u3059\u3002 Rust\u3067\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb [&hellip;]<\/p>\n","protected":false},"author":6,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-45698","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"yoast_head":"<!-- This site is optimized with the Yoast SEO Premium plugin v21.5 (Yoast SEO v21.5) - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>- Blog - Silicon Cloud<\/title>\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.silicloud.com\/zh\/blog\/45698-2\/\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:description\" content=\"\u672c\u8a18\u4e8b\u306fRust Advent Calendar 2017\u306e12\u670821\u65e5\u306e\u8a18\u4e8b\u3067\u3059\u3002 Rust\u3067\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb [&hellip;]\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/\" \/>\n<meta property=\"og:site_name\" content=\"Blog - Silicon Cloud\" \/>\n<meta property=\"article:published_time\" content=\"2024-03-09T22:57:41+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-05-03T15:09:45+00:00\" \/>\n<meta name=\"author\" content=\"\u6587, \u7fd4\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"\u6587, \u7fd4\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/\",\"url\":\"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/\",\"name\":\"- Blog - Silicon Cloud\",\"isPartOf\":{\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#website\"},\"datePublished\":\"2024-03-09T22:57:41+00:00\",\"dateModified\":\"2024-05-03T15:09:45+00:00\",\"author\":{\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/64d5cc7727fffbff2f9a2a8da1de3e5c\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/\"]}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#website\",\"url\":\"https:\/\/www.silicloud.com\/zh\/blog\/\",\"name\":\"Blog - Silicon Cloud\",\"description\":\"\",\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/64d5cc7727fffbff2f9a2a8da1de3e5c\",\"name\":\"\u6587, \u7fd4\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/920c3d673e0bccacc98e5e6b7149bb3c22edd8d39cb753e5d7d7e471498118a1?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/920c3d673e0bccacc98e5e6b7149bb3c22edd8d39cb753e5d7d7e471498118a1?s=96&d=mm&r=g\",\"caption\":\"\u6587, \u7fd4\"},\"url\":\"https:\/\/www.silicloud.com\/zh\/blog\/author\/wenxiang\/\"},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/#local-main-organization-logo\",\"url\":\"\",\"contentUrl\":\"\",\"caption\":\"Blog - Silicon Cloud\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"- Blog - Silicon Cloud","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.silicloud.com\/zh\/blog\/45698-2\/","og_locale":"zh_CN","og_type":"article","og_description":"\u672c\u8a18\u4e8b\u306fRust Advent Calendar 2017\u306e12\u670821\u65e5\u306e\u8a18\u4e8b\u3067\u3059\u3002 Rust\u3067\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb [&hellip;]","og_url":"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/","og_site_name":"Blog - Silicon Cloud","article_published_time":"2024-03-09T22:57:41+00:00","article_modified_time":"2024-05-03T15:09:45+00:00","author":"\u6587, \u7fd4","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"\u6587, \u7fd4","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"6 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/","url":"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/","name":"- Blog - Silicon Cloud","isPartOf":{"@id":"https:\/\/www.silicloud.com\/zh\/blog\/#website"},"datePublished":"2024-03-09T22:57:41+00:00","dateModified":"2024-05-03T15:09:45+00:00","author":{"@id":"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/64d5cc7727fffbff2f9a2a8da1de3e5c"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/"]}]},{"@type":"WebSite","@id":"https:\/\/www.silicloud.com\/zh\/blog\/#website","url":"https:\/\/www.silicloud.com\/zh\/blog\/","name":"Blog - Silicon Cloud","description":"","inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/64d5cc7727fffbff2f9a2a8da1de3e5c","name":"\u6587, \u7fd4","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/920c3d673e0bccacc98e5e6b7149bb3c22edd8d39cb753e5d7d7e471498118a1?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/920c3d673e0bccacc98e5e6b7149bb3c22edd8d39cb753e5d7d7e471498118a1?s=96&d=mm&r=g","caption":"\u6587, \u7fd4"},"url":"https:\/\/www.silicloud.com\/zh\/blog\/author\/wenxiang\/"},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.silicloud.com\/zh\/blog\/45698-2\/#local-main-organization-logo","url":"","contentUrl":"","caption":"Blog - Silicon Cloud"}]}},"_links":{"self":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/45698","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/users\/6"}],"replies":[{"embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/comments?post=45698"}],"version-history":[{"count":2,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/45698\/revisions"}],"predecessor-version":[{"id":94664,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/45698\/revisions\/94664"}],"wp:attachment":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/media?parent=45698"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/categories?post=45698"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/tags?post=45698"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}