{"id":45740,"date":"2023-12-25T01:08:04","date_gmt":"2023-07-17T01:41:30","guid":{"rendered":"https:\/\/www.silicloud.com\/zh\/blog\/45740-2\/"},"modified":"2024-05-04T15:15:14","modified_gmt":"2024-05-04T07:15:14","slug":"45740-2","status":"publish","type":"post","link":"https:\/\/www.silicloud.com\/zh\/blog\/45740-2\/","title":{"rendered":""},"content":{"rendered":"<p>\u672c\u8a18\u4e8b\u306f Rust Advent Calendar 2020 \u306e\uff16\u65e5\u76ee\u306e\u8a18\u4e8b\u3067\u3059\u3002<br \/>\n\u524d\u65e5\u306e\u8a18\u4e8b\u306fcipepser\u69d8\u306b\u3088\u308b\u300eRust\u306e\u53ef\u5909\u9577\u5f15\u6570\u95a2\u6570\u3068HList\u306e\u8a71\u300f\u3067\u3059\u3002<br \/>\n\u7fcc\u65e5\u306e\u8a18\u4e8b\u306ftermoshtt\u69d8\u306b\u3088\u308b\u300eRust \u306e Foreign Function Interface (FFI)\u300f\u3067\u3059\u3002<\/p>\n<h2>\u5909\u66f4\u5c65\u6b74<\/h2>\n<div>\n<div class=\"post-table\">\u5909\u66f4ID\u5e74\u6708\u65e5\u5185\u5bb912020\/12\/05\u521d\u7248\u4f5c\u621022020\/12\/06<a href=\"https:\/\/qiita.com\/namn1125\">namn1125<\/a>\u69d8\u306e\u30b3\u30e1\u30f3\u30c8\u3092\u53d7\u3051\u3066\u3001\u6bd4\u8f03\u7d50\u679c\u3092 <code>cargo run --release<\/code> \u3057\u305f\u3082\u306e\u306b\u4fee\u6b6332020\/12\/06\u30ec\u30a4\u30e4\u30fc\u304c\u53d7\u3051\u308b\u30c7\u30fc\u30bf\u578b\u3092 Array2 \u306b\u7d71\u4e00\u3057\u305f\u3068\u304d\u306e\u8a13\u7df4\u901f\u5ea6\u3092\u6bd4\u8f03\u7d50\u679c\u306b\u8ffd\u8a18<\/div>\n<\/div>\n<h1>\u672c\u8a18\u4e8b\u306e\u307e\u3068\u3081<\/h1>\n<p>\u300e\u30bc\u30ed\u304b\u3089\u4f5c\u308bDeep Learning\u2015Python\u3067\u5b66\u3076\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u7406\u8ad6\u3068\u5b9f\u88c5\u300f\uff08\u4ee5\u4e0b\u3001\u53c2\u8003\u66f8\u3068\u547c\u3076\u3053\u3068\u306b\u3057\u307e\u3059\uff09\u3092\u8aad\u307f\u30015 \u7ae0\u307e\u3067\u306e\u5185\u5bb9\u3092 Rust \u3067\u5b9f\u88c5\u3057\u3066\u307f\u307e\u3057\u305f\u3002<\/p>\n<p>MNIST \u306e\u624b\u66f8\u304d\u6587\u5b57\u3092\u8a8d\u8b58\u3059\u308b\u4e09\u5c64\u30d1\u30fc\u30bb\u30d7\u30c8\u30ed\u30f3\u30e2\u30c7\u30eb\u3092\u5b9f\u88c5\u3059\u308b\u4e0a\u3067\u3001\u7b46\u8005\uff08\u79c1\uff09\u304c\u63a1\u7528\u3057\u305f\u65b9\u91dd\u3068\u3001ndarray \u3092\u4f7f\u3046\u4e0a\u3067\u53c2\u8003\u306b\u306a\u308a\u305d\u3046\u306a\u30dd\u30a4\u30f3\u30c8\u3092\u30ea\u30b9\u30c8\u30a2\u30c3\u30d7\u3057\u307e\u3059\u3002<\/p>\n<h1>\u30ea\u30dd\u30b8\u30c8\u30ea\u306b\u3064\u3044\u3066<\/h1>\n<p>\u3053\u3061\u3089\u306e\u30ea\u30dd\u30b8\u30c8\u30ea\u306b Rust \u3067\u5b9f\u88c5\u3057\u305f\u30b3\u30fc\u30c9\u304c\u683c\u7d0d\u3055\u308c\u3066\u304a\u308a\u307e\u3059\u3002<\/p>\n<p>\u3053\u3061\u3089\u306e Oreilly Japan \u793e\u306e GitHub \u30ea\u30dd\u30b8\u30c8\u30ea\u306b\u5143\u306e Python \u30b3\u30fc\u30c9\u304c\u683c\u7d0d\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\u3053\u3061\u3089\u306e\u30ea\u30dd\u30b8\u30c8\u30ea\u306f\u4e0a\u8a18 Oreilly Japan \u793e\u30ea\u30dd\u30b8\u30c8\u30ea\u3092 fork \u3057\u305f\u3082\u306e\u3067\u3042\u308a\u3001\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u7528\u306b\u4fee\u6b63\u3057\u305f\u30b3\u30fc\u30c9\u3068 Keras in TensorFlow \u3067\u306e\u5b9f\u88c5\u304c\u683c\u7d0d\u3055\u308c\u3066\u304a\u308a\u307e\u3059\u3002<\/p>\n<h1>\u672c\u7de8\u306e\u76ee\u6b21<\/h1>\n<ol>\n<li style=\"list-style-type: none;\">\n<ol>\u52d5\u6a5f<\/ol>\n<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<ol>\n<li style=\"list-style-type: none;\">\n<ol>\u5b9f\u88c5\u306e\u57fa\u672c\u65b9\u91dd<\/ol>\n<\/li>\n<\/ol>\n<p>\u30e2\u30c7\u30eb\u306e\u5b9f\u88c5<\/p>\n<p>\u30c7\u30fc\u30bf\u578b<br \/>\n\u30ec\u30a4\u30e4\u30fc\u30e2\u30c7\u30eb<br \/>\n\u9806\u4f1d\u64ad\u3068\u8aa4\u5dee\u9006\u4f1d\u64ad<\/p>\n<p>\u30dd\u30a4\u30f3\u30c8<\/p>\n<p>\u5b9f\u88c5\u304c\u5fc5\u8981\u306a\u95a2\u6570<br \/>\nArray\u3068ArrayView\u306e\u5909\u63db<br \/>\nArray2\u3068ArrayD\u306e\u5909\u63db<br \/>\nArray\u304b\u3089\u7121\u4f5c\u70ba\u306b\u8981\u7d20\u3092\u62bd\u51fa\u3059\u308b\u65b9\u6cd5<\/p>\n<p>\u30d9\u30f3\u30c1\u30de\u30fc\u30af<br \/>\n\u304a\u308f\u308a\u306b<\/p>\n<p>\u53c2\u8003<\/p>\n<p>\u5148\u99c6\u8005\u306e\u65b9\u3005<br \/>\nRust\u3067\u4f7f\u3048\u308b\u6df1\u5c64\u5b66\u7fd2\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af<\/p>\n<h1>\u52d5\u6a5f<\/h1>\n<p>\u3053\u308c\u307e\u3067 Keras in TensorFlow \u3092\u4f7f\u7528\u3057\u3066\u6df1\u5c64\u5b66\u7fd2\u3067\u904a\u3076\u5074\u3067\u3057\u305f\u304c\u3001\u5185\u90e8\u306e\u52d5\u4f5c\uff08\u4f8b\u3048\u3070\u8aa4\u5dee\u9006\u4f1d\u64ad\uff09\u306e\u5b9f\u88c5\u65b9\u6cd5\u306b\u3064\u3044\u3066\u306f\u304b\u3089\u3063\u304d\u3057\u3067\u3057\u305f\u3002<br \/>\n\u3061\u3087\u3046\u3069 Rust \u3092\u52c9\u5f37\u3057\u3066\u3044\u308b\u3068\u3053\u308d\u3067\u3042\u308a\u3001\u6df1\u5c64\u5b66\u7fd2\u30e2\u30c7\u30eb\u306e\u5b9f\u88c5\u304c Rust \u306e\u5b66\u7fd2\u306b\u304a\u3044\u3066\u826f\u3044\u984c\u6750\u306b\u306a\u308b\u3068\u601d\u3044\u3001\u5b9f\u88c5\u306b\u30c1\u30e3\u30ec\u30f3\u30b8\u3057\u307e\u3057\u305f\u3002<\/p>\n<h1>\u5b9f\u88c5\u306e\u57fa\u672c\u65b9\u91dd<\/h1>\n<p>\u4eca\u56de\u306e\u5b9f\u88c5\u306e\u57fa\u672c\u65b9\u91dd\u306f\u6b21\u306e\u901a\u308a\u3067\u3059\u3002<\/p>\n<ol>\n<li style=\"list-style-type: none;\">\n<ol>\u3067\u304d\u308b\u3060\u3051 Rust-native \u306a\u30b3\u30fc\u30c9\u306b\u3059\u308b\u3002<\/ol>\n<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<ol>\n<li style=\"list-style-type: none;\">\n<ol>\u4eca\u56de\u3001\u4ed6\u306e\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u8a00\u8a9e\u3067\u4f5c\u6210\u3055\u308c\u305f\u30e9\u30a4\u30d6\u30e9\u30ea\u306b\u5bfe\u3059\u308b Rust bindings \u3092\u4f5c\u6210\u3057\u307e\u305b\u3093\u3067\u3057\u305f\u3002\u307e\u305f\u3001Rust bindings \u306e\u4f7f\u7528\u3092\u3067\u304d\u308b\u3060\u3051\u63a7\u3048\u307e\u3057\u305f\u3002\u4f8b\u3048\u3070 NumPy \u306e Rust binding \u3067\u3042\u308b rust-numpy \u306f\u4f7f\u7528\u3057\u3066\u3044\u307e\u305b\u3093\u3002<\/ol>\n<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<ol>\n<li style=\"list-style-type: none;\">\n<ol>unsafe \u306a\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u306f\u967d\u306b\u306f\u884c\u308f\u306a\u3044\u3002<\/ol>\n<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<ol>\u7b46\u8005\u304c\u307e\u3060 Rust \u306b\u6163\u308c\u3066\u3044\u306a\u3044\u306e\u3067\u3001unsafe \u306a\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u3092\u884c\u308f\u306a\u3044\u3053\u3068\u3068\u3057\u307e\u3057\u305f\u3002\u4f7f\u7528\u3059\u308b\u65e2\u5b58\u306e\u30af\u30ec\u30fc\u30c8\u5185\u3067 unsafe \u306a\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u304c\u3055\u308c\u3066\u3044\u308b\u3082\u306e\u306f\u8a31\u5bb9\u3057\u307e\u3057\u305f\u3002<\/ol>\n<h1>\u30e2\u30c7\u30eb\u306e\u5b9f\u88c5<\/h1>\n<h2>\u30c7\u30fc\u30bf\u578b<\/h2>\n<p>\u4eca\u56de\u3001\u7279\u5fb4\u91cf\u3068\u30e9\u30d9\u30eb\u306e\u30c7\u30fc\u30bf\u578b\u3068\u3057\u3066 ndarray \u3067\u5b9f\u88c5\u3055\u308c\u3066\u3044\u308b\u578b\uff08Array \u7cfb\uff09\u3092\u63a1\u7528\u3057\u307e\u3057\u305f\u3002ndarray \u306e\u5229\u70b9\u3068\u3057\u3066\u6b21\u304c\u6319\u3052\u3089\u308c\u307e\u3059\u3002<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u591a\u6b21\u5143\u914d\u5217\u304c\u5b9f\u88c5\u3055\u308c\u3066\u3044\u308b\u3002<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\u4ee3\u8868\u7684\u306a\u914d\u5217\u6f14\u7b97\u304c\u5b9f\u88c5\u3055\u308c\u3066\u3044\u308b\u3002<\/ul>\n<p>\u4ed6\u306e\u5019\u88dc\u3068\u3057\u3066\u914d\u5217\u3068 Vec \u304c\u8003\u3048\u3089\u308c\u307e\u3057\u305f\u304c\u3001\u4ee5\u4e0b\u306b\u793a\u3059\u8ab2\u984c\u304c\u3042\u308a\u307e\u3057\u305f\u3002\u307e\u305a\u914d\u5217\u306b\u3064\u3044\u3066\u3001<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u30b3\u30f3\u30d1\u30a4\u30eb\u6642\u306b\u30b5\u30a4\u30ba\u304c\u78ba\u5b9a\u3057\u3066\u3044\u308b\u5fc5\u8981\u304c\u3042\u308b\u3002<\/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\">\u591a\u6b21\u5143\u914d\u5217\u306e\u5b9f\u88c5\u304c\u9762\u5012\u3002<\/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\">\u57fa\u672c\u7684\u306a\u914d\u5217\u9593\u306e\u6f14\u7b97\u3092\u81ea\u524d\u3067\u5b9f\u88c5\u3059\u308b\u5fc5\u8981\u304c\u3042\u308b\u3002<\/ul>\n<\/li>\n<\/ul>\n<p>numpy.dot \u3068\u540c\u3058\u51e6\u7406\u306a\u3069\u3002<\/p>\n<p>\u95a2\u6570\u304c\u914d\u5217\u3092\u623b\u308a\u5024\u306b\u3059\u308b\u3053\u3068\u304c\u96e3\u3057\u3044\u3002<\/p>\n<p>\u3053\u308c\u306f Box \u3067\u30e9\u30c3\u30d7\u3059\u308b\u3001\u53c2\u7167\u3092\u5f15\u6570\u306b\u3068\u308b\u3001\u7b49\u306e\u65b9\u6cd5\u3067\u89e3\u6c7a\u3067\u304d\u308b\u3068\u601d\u308f\u308c\u307e\u3059\u3002<\/p>\n<p>\u6b21\u306b Vec \u306b\u3064\u3044\u3066\u3001<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u591a\u6b21\u5143\u914d\u5217\u306e\u5b9f\u88c5\u304c\u9762\u5012\u3002<\/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\">\u57fa\u672c\u7684\u306a\u914d\u5217\u9593\u306e\u6f14\u7b97\u3092\u81ea\u524d\u3067\u5b9f\u88c5\u3059\u308b\u5fc5\u8981\u304c\u3042\u308b\u3002<\/ul>\n<\/li>\n<\/ul>\n<p>\u3082\u3057\u304b\u3057\u305f\u3089\u65e2\u5b58\u306e\u30af\u30ec\u30fc\u30c8\u3067\u5b9f\u88c5\u304c\u3042\u308b\u304b\u3082\u3057\u308c\u307e\u305b\u3093\u3002<\/p>\n<p>\u81ea\u524d\u306e\u5b9f\u88c5\u3067\u306f ndarray \u3088\u308a\u304b\u306a\u308a\u51e6\u7406\u901f\u5ea6\u304c\u9045\u304b\u3063\u305f\u3002<\/p>\n<p>\u5b9f\u88c5\u6b21\u7b2c\u3067\u306f\u9ad8\u901f\u306b\u306a\u308b\u3068\u601d\u3044\u307e\u3059\u3002<\/p>\n<p>\u3053\u308c\u3089\u306e\u8ab2\u984c\u3092\u89e3\u6c7a\u3059\u308b\u624b\u9593\u3068 ndarray \u306e\u5229\u4fbf\u6027\u3092\u9451\u307f\u3066\u3001ndarray \u3092\u63a1\u7528\u3057\u307e\u3057\u305f\u3002<\/p>\n<h2>\u30ec\u30a4\u30e4\u30fc\u30e2\u30c7\u30eb<\/h2>\n<div><img decoding=\"async\" class=\"post-images\" title=\"\" src=\"https:\/\/cdn.silicloud.com\/blog-img\/blog\/img\/657d60e437434c4406cfc4b3\/24-0.png\" alt=\"\u4e09\u5c64\u30d1\u30fc\u30bb\u30d7\u30c8\u30ed\u30f3\u30e2\u30c7\u30eb\" \/><\/div>\n<p>\u5b9f\u88c5\u3057\u305f\u30ec\u30a4\u30e4\u30fc\uff08\u69cb\u9020\u4f53\uff09\u3092\u4ee5\u4e0b\u306b\u6319\u3052\u307e\u3059\u3002<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">Affine \u5909\u63db ${\\mathbf A}(\\vec{x}) = {\\mathbf W}\\vec{x} + \\vec{b}$<\/ul>\n<\/li>\n<\/ul>\n<p>${\\mathbf W}$\uff1a\u91cd\u307f\u3001$\\vec{b}$\uff1a\u30d0\u30a4\u30a2\u30b9<\/p>\n<p>\u6d3b\u6027\u5316\u95a2\u6570 $\\phi(x)$<\/p>\n<p>ReLU\uff1a$\\phi(x)=\\left\\{\\begin{array}{cc} x &amp; (x &gt; 0) \\\\ 0 &amp; (x \\leq 0) \\end{array}\\right.$<br \/>\nSigmoid\uff1a$\\phi(x)=1\/(1 + e^{-x})$<\/p>\n<p>\u4eca\u56de\u306e\u30e2\u30c7\u30eb\u3067\u306f\u4f7f\u7528\u3057\u3066\u3044\u307e\u305b\u3093\u3002<\/p>\n<p>Softmax\uff1a$\\vec{x}=(x_1,x_2,\\dots,x_n)\\rightarrow \\phi(x_k\\lvert\\vec{x}) = e^{x_k}\/\\sum_j e^{x_j}$<\/p>\n<p>\u640d\u5931\u95a2\u6570<\/p>\n<p>Cross-entropy loss\uff1a$E(\\vec{y}, \\vec{t}) = -\\sum_k t_k\\log y_k$<\/p>\n<p>$\\vec{y},\\vec{t}$ \u306f\u305d\u308c\u305e\u308c\u51fa\u529b\u5c64\u306e\u51fa\u529b\u3001\u30e9\u30d9\u30eb\u3092\u8868\u3057\u307e\u3059\u3002<\/p>\n<p>\u7c21\u5358\u306e\u305f\u3081\u3001softmax \u30ec\u30a4\u30e4\u30fc\u3068 cross entropy error \u30ec\u30a4\u30e4\u30fc\u3092\u7d50\u5408\u3057\u305f\u30ec\u30a4\u30e4\u30fc\u3092\u63a1\u7528\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\u5404\u30ec\u30a4\u30e4\u30fc\u306b\u5171\u901a\u306e\u95a2\u6570\u306f\u30c8\u30ec\u30a4\u30c8\u3068\u3057\u3066\u3001\u6b21\u306e\u3088\u3046\u306a\u5f62\u5f0f\u3067\u5b9f\u88c5\u3057\u307e\u3057\u305f\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"k\">pub<\/span> <span class=\"k\">trait<\/span> <span class=\"n\">LayerBase<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">T<\/span><span class=\"o\">&gt;<\/span> <span class=\"p\">{<\/span>\r\n    <span class=\"c\">\/\/ \u9806\u4f1d\u64ad<\/span>\r\n    <span class=\"k\">fn<\/span> <span class=\"nf\">forward<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"k\">self<\/span><span class=\"p\">,<\/span> <span class=\"n\">x<\/span><span class=\"p\">:<\/span> <span class=\"o\">&amp;<\/span><span class=\"n\">ArrayD<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">T<\/span><span class=\"o\">&gt;<\/span><span class=\"p\">)<\/span> <span class=\"k\">-&gt;<\/span> <span class=\"n\">ArrayD<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">T<\/span><span class=\"o\">&gt;<\/span><span class=\"p\">;<\/span>\r\n    <span class=\"c\">\/\/ \u9006\u4f1d\u64ad<\/span>\r\n    <span class=\"k\">fn<\/span> <span class=\"nf\">backward<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"k\">self<\/span><span class=\"p\">,<\/span> <span class=\"n\">x<\/span><span class=\"p\">:<\/span> <span class=\"o\">&amp;<\/span><span class=\"n\">ArrayD<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">T<\/span><span class=\"o\">&gt;<\/span><span class=\"p\">,<\/span> <span class=\"n\">dx<\/span><span class=\"p\">:<\/span> <span class=\"o\">&amp;<\/span><span class=\"n\">ArrayD<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">T<\/span><span class=\"o\">&gt;<\/span><span class=\"p\">)<\/span> <span class=\"k\">-&gt;<\/span> <span class=\"n\">ArrayD<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">T<\/span><span class=\"o\">&gt;<\/span><span class=\"p\">;<\/span>\r\n    <span class=\"c\">\/\/ \u30ec\u30a4\u30e4\u30fc\u306e\u91cd\u307f\u306e\u66f4\u65b0\u203b<\/span>\r\n    <span class=\"k\">fn<\/span> <span class=\"nf\">update<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">mut<\/span> <span class=\"k\">self<\/span><span class=\"p\">,<\/span> <span class=\"n\">lr<\/span><span class=\"p\">:<\/span> <span class=\"n\">T<\/span><span class=\"p\">);<\/span>\r\n<span class=\"p\">}<\/span>\r\n<\/code><\/pre>\n<p>\u203b\u4eca\u56de\u5b9f\u88c5\u3057\u305f\u6d3b\u6027\u5316\u95a2\u6570\u30ec\u30a4\u30e4\u30fc\u3084\u640d\u5931\u95a2\u6570\u30ec\u30a4\u30e4\u30fc\u306b\u306f\u5fc5\u8981\u3042\u308a\u307e\u305b\u3093\u304c\u3001\u4fbf\u5b9c\u4e0a\u5171\u901a\u95a2\u6570\u3068\u3057\u3066\u5b9f\u88c5\u3057\u307e\u3057\u305f\u3002<\/p>\n<h2>\u9806\u4f1d\u64ad\u3068\u8aa4\u5dee\u9006\u4f1d\u64ad<\/h2>\n<p>\u6df1\u5c64\u5b66\u7fd2\u306b\u304a\u3051\u308b\u51fa\u529b\u306e\u8a08\u7b97\u3068\u30e2\u30c7\u30eb\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u66f4\u65b0\u624b\u9806\u306b\u3064\u3044\u3066\u306f\u3001\u591a\u304f\u306e\u65b9\u304c\u8a18\u4e8b\u306b\u3055\u308c\u3066\u3044\u307e\u3059\uff08Qiita \u5185\u3067\u3042\u308c\u3070\u3053\u3061\u3089\u306e\u8a18\u4e8b\u3084\u3053\u3061\u3089\u306e\u8a18\u4e8b\u304c\u53c2\u8003\u306b\u306a\u308b\u3068\u601d\u3044\u307e\u3059\uff09\u3002\u3053\u3053\u3067\u306f\u8efd\u304f\u89e6\u308c\u308b\u306b\u7559\u3081\u307e\u3059\u3002<\/p>\n<h3>\u9806\u4f1d\u64ad<\/h3>\n<p>$k$ \u5c64\u76ee\u306e\u96a0\u308c\u5c64\u306e\u30b5\u30a4\u30ba\u3092 $L_k$ \u3068\u3057\u307e\u3059\u3002\u6b21\u306e\u4e8c\u3064\u306e\u624b\u9806\u3092\u7e70\u308a\u8fd4\u3059\u3053\u3068\u3067\u3001\u5165\u529b $\\vec{y}_1$ \u306b\u5bfe\u3057\u3066\u4e88\u6e2c\u5024 $\\vec{y}_N$ \u3092\u5f97\u307e\u3059\u3002<\/p>\n<ol>\n<li style=\"list-style-type: none;\">\n<ol>Affine \u5909\u63db\u30ec\u30a4\u30e4\u30fc ${\\mathbf A}_k$ \u3092\u901a\u3059\uff1a$\\vec{v}_{k} = {\\mathbf A}_k(\\vec{y}_{k}) = {\\mathbf W}_k\\vec{y}_{k} + \\vec{b}_k$<\/ol>\n<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<ol>\u6d3b\u6027\u5316\u95a2\u6570\u30ec\u30a4\u30e4\u30fc ${\\mathbf \\Phi}_k$ \u3092\u901a\u3059\uff1a$\\vec{y}_{k+1} = {\\mathbf \\Phi}_k(\\vec{v}_k)$<\/ol>\n<h3>\u8aa4\u5dee\u9006\u4f1d\u64ad<\/h3>\n<p>\u30d1\u30fc\u30bb\u30d7\u30c8\u30ed\u30f3\u30e2\u30c7\u30eb\u3067\u66f4\u65b0\u3055\u308c\u308b\u306e\u306f Affine\u5909\u63db\u30ec\u30a4\u30e4\u30fc\u306e\u91cd\u307f ${\\mathbf W}_k = (w_{ij}^{(k)})$ \u3068\u30d0\u30a4\u30a2\u30b9 $\\vec{b}_k = (b_1^{(k)},b_2^{(k)},\\dots,b_{L_k}^{(k)})$ \u3067\u3059\u3002\u3053\u308c\u3089\u306f\u6b21\u5f0f\u306e\u3088\u3046\u306b\u640d\u5931\u95a2\u6570\u306e\u504f\u5fae\u5206\u3092\u7528\u3044\u3066\u66f4\u65b0\u3055\u308c\u307e\u3059\u3002<br \/>\n$$<br \/>\nw_{ij}^{(k)}\\rightarrow w_{ij}^{(k)} &#8211; \\epsilon\\frac{\\partial E}{\\partial w_{ij}^{(k)}},\\ b_j^{(k)} \\rightarrow b_j^{(k)} &#8211; \\epsilon\\frac{\\partial E}{\\partial b_j^{(k)}}.<br \/>\n$$<br \/>\n\u3053\u3053\u3067 $\\epsilon\\ (&gt; 0)$ \u306f\u5b66\u7fd2\u7387\u3067\u3059\u3002<br \/>\n\u8aa4\u5dee\u9006\u4f1d\u64ad\u3067\u3059\u304b\u3089\u3001\u51fa\u529b\u5c64\u306b\u8fd1\u3044\u5074\u304b\u3089\u5165\u529b\u5c64\u306b\u8fd1\u3044\u5074\u3078\u3068\u8aa4\u5dee\u304c\u4f1d\u64ad\u3057\u3001\u5404\u5c64\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\u304c\u66f4\u65b0\u3055\u308c\u307e\u3059\u3002<br \/>\n\u5404\u5c64\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u66f4\u65b0\u3082\u307e\u305f\u5c64\u3054\u3068\u306b\u72ec\u7acb\u306b\u5b9f\u884c\u3067\u304d\u308b\u305f\u3081\u3001update \u95a2\u6570\u3092\u5171\u901a\u95a2\u6570\u3068\u3057\u3066\u30c8\u30ec\u30a4\u30c8\u306b\u5b9f\u88c5\u3057\u307e\u3057\u305f\u3002<\/p>\n<p>\u3055\u3066\u3001\u7b2c $N-1$ \u5c64\u76ee\u306b\u5bfe\u5fdc\u3059\u308b Affine \u5909\u63db\u30ec\u30a4\u30e4\u30fc\u306b\u5bfe\u3059\u308b\u8aa4\u5dee\u306e\u504f\u5fae\u5206\u3092\u8a08\u7b97\u3057\u3066\u307f\u307e\u3059\u3002$E = E(\\vec{y}_N, \\vec{t})$ \u306f\u4e00\u898b ${\\mathbf W}_{N-1},\\vec{b}_{N-1}$ \u306b\u4f9d\u3089\u306a\u3044\u3088\u3046\u306b\u898b\u3048\u307e\u3059\u304c\u3001<br \/>\n$$<br \/>\n\\vec{y}_N = {\\mathbf \\Phi}_{N-1}(\\vec{v}_{N-1}) = {\\mathbf \\Phi}_{N-1}({\\mathbf W}_{N-1}\\vec{y}_{N-1} + \\vec{b}_{N-1})<br \/>\n$$<br \/>\n\u3068\u8868\u73fe\u3067\u304d\u308b\u3053\u3068\u3092\u8003\u616e\u3059\u308b\u3068\u3001$E(\\vec{y}_N, \\vec{t})$ \u304c ${\\mathbf W}_{N-1},\\vec{b}_{N-1}$ \u306b\u4f9d\u5b58\u3059\u308b\u3053\u3068\u304c\u5206\u304b\u308a\u307e\u3059\u3002\u3088\u3063\u3066\u504f\u5fae\u5206\u306e\u9023\u9396\u5f8b\u3088\u308a<br \/>\n$$<br \/>\n\\begin{array}{rcl}<br \/>\n\\frac{\\partial E}{\\partial w_{ij}^{(N-1)}} &amp;=&amp; \\frac{\\partial E}{\\partial \\vec{y}_N}\\cdot\\frac{\\partial \\vec{y}_{N}}{\\partial \\vec{v}_{N-1}}\\cdot\\frac{\\partial \\vec{v}_{N-1}}{\\partial w_{ij}^{(N-1)}} = \\sum_l \\sum_m \\frac{\\partial E}{\\partial y_l^{(N)}}\\frac{\\partial y_l^{(N)}}{\\partial v_m^{(N-1)}}\\frac{\\partial v_m^{(N-1)}}{\\partial w_{ij}^{(N-1)}},\\\\<br \/>\n\\frac{\\partial E}{\\partial b_j^{(N-1)}} &amp;=&amp; \\frac{\\partial E}{\\partial \\vec{y}_N}\\cdot\\frac{\\partial \\vec{y}_{N}}{\\partial \\vec{v}_{N-1}}\\cdot\\frac{\\partial \\vec{v}_{N-1}}{\\partial b_j^{(N-1)}} = \\sum_l \\sum_m \\frac{\\partial E}{\\partial y_l^{(N)}}\\frac{\\partial y_l^{(N)}}{\\partial v_m^{(N-1)}}\\frac{\\partial v_m^{(N-1)}}{\\partial b_j^{(N-1)}}.<br \/>\n\\end{array}<br \/>\n$$<br \/>\n\u4e0a\u5f0f\u306e\u3046\u3061 $\\partial E\/\\partial \\vec{y}_N$ \u306f\u640d\u5931\u95a2\u6570\u306e\u504f\u5fae\u5206\u3001$\\partial \\vec{y}_{N}\/\\partial \\vec{v}_{N-1}$ \u306f\u6d3b\u6027\u5316\u95a2\u6570\u306e\u504f\u5fae\u5206\u3001$\\partial v_m^{(N-1)}\/\\partial w_{ij}^{(N-1)}$ \u304a\u3088\u3073 $\\partial v_m^{(N-1)}\/\\partial b_j^{(N-1)}$ \u306f Affine \u5909\u63db\u306e\u504f\u5fae\u5206\u3067\u3042\u308a\u3001\u305d\u308c\u305e\u308c\u72ec\u7acb\u306b\u8a08\u7b97\u3067\u304d\u307e\u3059\u3002\u3057\u305f\u304c\u3063\u3066\u3001\u305d\u308c\u305e\u308c\u306e\u30ec\u30a4\u30e4\u30fc\u306b\u504f\u5fae\u5206\u64cd\u4f5c\u3092\u6301\u305f\u305b\u308b\u3088\u3046\u306b\u8aa4\u5dee\u9006\u4f1d\u64ad\u306e\u8a08\u7b97\u3092\u5b9f\u88c5\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002\u3053\u306e\u3088\u3046\u306a\u80cc\u666f\u304c\u3042\u3063\u3066\u3001backward \u95a2\u6570\u3092\u5171\u901a\u306e\u95a2\u6570\u3068\u3057\u3066\u30c8\u30ec\u30a4\u30c8\u306b\u5b9f\u88c5\u3057\u307e\u3057\u305f\u3002<\/p>\n<p>\u306a\u304a\u3001\u6b8b\u308a\u306e\u30ec\u30a4\u30e4\u30fc\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\u306b\u5bfe\u3059\u308b\u8aa4\u5dee\u306e\u504f\u5fae\u5206\u3082\u540c\u69d8\u306b\u8a08\u7b97\u3067\u304d\u307e\u3059\u3002\u8a73\u3057\u304f\u306f\u3053\u3061\u3089\u306e\u8a18\u4e8b\u3092\u3054\u53c2\u7167\u304f\u3060\u3055\u3044\u3002<\/p>\n<h1>\u30dd\u30a4\u30f3\u30c8<\/h1>\n<h2>\u5b9f\u88c5\u304c\u5fc5\u8981\u306a\u95a2\u6570<\/h2>\n<p>\u4fbf\u5b9c\u4e0a\u3001\u6b21\u306e\u95a2\u6570\u3092 MathFunc \u30c8\u30ec\u30a4\u30c8\u3068\u3057\u3066 ndarray::{Array1, Array2, ArrayD} \u306b\u5b9f\u88c5\u3057\u307e\u3057\u305f\u3002ArrayBase \u306b\u5b9f\u88c5\u3059\u308b\u3060\u3051\u3067\u826f\u304b\u3063\u305f\u304b\u3082\u3057\u308c\u307e\u305b\u3093\u304c\u3001\u3061\u3087\u3063\u3068\u3046\u307e\u304f\u884c\u304d\u307e\u305b\u3093\u3067\u3057\u305f\u3002\u3053\u306e\u8fba\u306f\u5c06\u6765\u7684\u306b\u4fee\u6b63\u3057\u305f\u3044\u3068\u3053\u308d\u3067\u3059\u3002<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u6307\u6570\u95a2\u6570\uff1a$x \\rightarrow \\exp(x)$<\/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\">\u5bfe\u6570\u95a2\u6570\uff1a$x \\rightarrow \\log_t(x)$<\/ul>\n<\/li>\n<\/ul>\n<p>\u81ea\u7136\u5bfe\u6570\u3092\u5e95\u3068\u3059\u308b\u5bfe\u6570\u95a2\u6570\uff1a$x \\rightarrow \\log_e(x)$<\/p>\n<pre class=\"post-pre\"><code><span class=\"k\">pub<\/span> <span class=\"k\">trait<\/span> <span class=\"n\">MathFunc<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">T<\/span><span class=\"o\">&gt;<\/span><span class=\"p\">{<\/span>\r\n    <span class=\"k\">fn<\/span> <span class=\"nf\">exp<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">self<\/span><span class=\"p\">)<\/span> <span class=\"k\">-&gt;<\/span> <span class=\"n\">Self<\/span><span class=\"p\">;<\/span>\r\n    <span class=\"k\">fn<\/span> <span class=\"k\">log<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">self<\/span><span class=\"p\">,<\/span> <span class=\"n\">e<\/span><span class=\"p\">:<\/span> <span class=\"n\">T<\/span><span class=\"p\">)<\/span> <span class=\"k\">-&gt;<\/span> <span class=\"n\">Self<\/span><span class=\"p\">;<\/span>\r\n    <span class=\"k\">fn<\/span> <span class=\"nf\">log_natural<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">self<\/span><span class=\"p\">)<\/span> <span class=\"k\">-&gt;<\/span> <span class=\"n\">Self<\/span><span class=\"p\">;<\/span>\r\n<span class=\"p\">}<\/span>\r\n<\/code><\/pre>\n<p>\u4eca\u56de\u3001\u30e9\u30d9\u30eb\u306e\u5f62\u3068\u3057\u3066 one-hot \u8868\u73fe\u3092\u63a1\u7528\u3057\u3066\u3044\u308b\u305f\u3081\u3001\u4e88\u6e2c\u7d50\u679c\u3092\u6bd4\u8f03\u3059\u308b\u969b\u306b argmax \u304c\u5fc5\u8981\u306b\u306a\u308a\u307e\u3059\u3002argmax \u306f ndarray-stats \u306e QuantileExt \u30c8\u30ec\u30a4\u30c8\u306b\u5b9f\u88c5\u3055\u308c\u3066\u3044\u308b\u306e\u3067\u3001\u305d\u308c\u3092\u5229\u7528\u3057\u307e\u3057\u305f\u3002<\/p>\n<h2>Array\u3068ArrayView\u306e\u5909\u63db<\/h2>\n<p>\u4eca\u56de\u3001\u5f15\u6570\u306b ArrayD \u306a\u3069\u306e Array&lt;A, D&gt; = ArrayBase&lt;OwnedRepr<a>, D&gt; \u7cfb\u306e\u578b\u3092\u3068\u308b\u95a2\u6570\u3092\u5b9f\u88c5\u3057\u305f\u306e\u3067\u3059\u304c\u3001ndarray \u306e slice \u7cfb\u306e\u95a2\u6570\u304c ArrayView&lt;&#8216;a, A, D&gt; = ArrayBase&lt;ViewRepr&lt;&amp;&#8217;a A&gt;, D&gt; \u3092\u8fd4\u3059\u306e\u3067\u3001\u305d\u306e\u9593\u306e\u5909\u63db\u64cd\u4f5c\u304c\u5fc5\u8981\u306b\u306a\u308b\u5834\u9762\u304c\u4f55\u5ea6\u304b\u3042\u308a\u307e\u3057\u305f\u3002<br \/>\nArrayView \u7cfb\u304b\u3089 Array \u7cfb\u306b\u5909\u63db\u3059\u308b\u95a2\u6570\u304c ndarray \u30af\u30ec\u30fc\u30c8\u5185\u306b\u306f\u7121\u3044\u3088\u3046\u3067\u3001\u4ee3\u66ff\u624b\u6bb5\u3092\u898b\u3064\u3051\u308b\u306e\u306b\u82e6\u52b4\u3057\u307e\u3057\u305f\u3002<br \/>\n\u4f8b\u3048\u3070\u6b21\u306e\u3088\u3046\u306b\u53c2\u7167\u306b\u5b9a\u6570\u3092\u4e57\u3059\u308b\u3053\u3068\u3067 ArrayView \u3092 Array \u306b\u5909\u63db\u3067\u304d\u307e\u3059\u3002<\/a><\/p>\n<pre class=\"post-pre\"><code><span class=\"k\">let<\/span> <span class=\"n\">a<\/span><span class=\"p\">:<\/span> <span class=\"n\">Array1<\/span><span class=\"o\">&lt;<\/span><span class=\"nb\">f32<\/span><span class=\"o\">&gt;<\/span> <span class=\"o\">=<\/span> <span class=\"nd\">arr!<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"p\">[<\/span><span class=\"mf\">1.0<\/span><span class=\"p\">,<\/span> <span class=\"mf\">2.0<\/span><span class=\"p\">,<\/span> <span class=\"mf\">3.0<\/span><span class=\"p\">]);<\/span>\r\n<span class=\"k\">let<\/span> <span class=\"n\">b<\/span><span class=\"p\">:<\/span> <span class=\"n\">Array1<\/span><span class=\"o\">&lt;<\/span><span class=\"nb\">f32<\/span><span class=\"o\">&gt;<\/span> <span class=\"o\">=<\/span> <span class=\"mf\">1.0<\/span> <span class=\"o\">*<\/span> <span class=\"o\">&amp;<\/span><span class=\"n\">a<\/span><span class=\"nf\">.slice<\/span><span class=\"p\">(<\/span><span class=\"nd\">s!<\/span><span class=\"p\">[<\/span><span class=\"o\">..<\/span><span class=\"mi\">2<\/span><span class=\"p\">]);<\/span>\r\n<\/code><\/pre>\n<h2>Array2\u3068ArrayD\u306e\u5909\u63db<\/h2>\n<p>\u4eca\u56de\u3001\u4e09\u5c64\u30d1\u30fc\u30bb\u30d7\u30c8\u30ed\u30f3\u30e2\u30c7\u30eb\u3067\uff12\u6b21\u5143 Affine \u5909\u63db\u3092\u63a1\u7528\u3057\u307e\u3057\u305f\u3002ndarray::Array2 \u306b\u306f\u5185\u7a4d\u95a2\u6570 dot \u304c\u5b9f\u88c5\u3055\u308c\u3066\u3044\u307e\u3059\u304c\u3001ArrayD \u306b\u306f\u5b9f\u88c5\u3055\u308c\u3066\u3044\u307e\u305b\u3093\uff08\u3082\u3057\u304b\u3057\u305f\u3089\u30c6\u30f3\u30bd\u30eb\u7a4d\u304c\u3042\u308b\u304b\u3082\u3057\u308c\u307e\u305b\u3093\u304c\u3001\u3061\u3087\u3063\u3068\u898b\u3064\u3051\u3089\u308c\u3066\u3044\u307e\u305b\u3093\uff09\u3002<br \/>\n\u6d3b\u6027\u5316\u95a2\u6570\u306a\u3069\u306f\u3067\u304d\u308b\u3060\u3051\u591a\u6b21\u5143\u914d\u5217\u306b\u5bfe\u5fdc\u3055\u305b\u305f\u3044\u3068\u3044\u3046\u3053\u3068\u3082\u3042\u308a\u3001Array2 \u3068 ArrayD \u3092\u4e92\u3044\u306b\u5909\u63db\u3059\u308b\u5fc5\u8981\u304c\u751f\u3058\u307e\u3057\u305f\u3002<br \/>\nndarray \u30af\u30ec\u30fc\u30c8\u5185\u306b\u5b9f\u88c5\u3055\u308c\u3066\u3044\u308b\u6b21\u306e\u95a2\u6570\u304c\u524d\u8ff0\u306e\u5909\u63db\u64cd\u4f5c\u3092\u5b9f\u73fe\u3057\u307e\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"k\">pub<\/span> <span class=\"k\">fn<\/span> <span class=\"n\">into_dimensionality<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">D2<\/span><span class=\"o\">&gt;<\/span><span class=\"p\">(<\/span><span class=\"k\">self<\/span><span class=\"p\">)<\/span> <span class=\"k\">-&gt;<\/span> <span class=\"n\">Result<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">ArrayBase<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">S<\/span><span class=\"p\">,<\/span> <span class=\"n\">D2<\/span><span class=\"o\">&gt;<\/span><span class=\"p\">,<\/span> <span class=\"n\">ShapeError<\/span><span class=\"o\">&gt;<\/span>\r\n<span class=\"k\">where<\/span>\r\n    <span class=\"n\">D2<\/span><span class=\"p\">:<\/span> <span class=\"n\">Dimension<\/span><span class=\"p\">,<\/span> \r\n<\/code><\/pre>\n<h2>Array\u304b\u3089\u7121\u4f5c\u70ba\u306b\u8981\u7d20\u3092\u62bd\u51fa\u3059\u308b\u65b9\u6cd5<\/h2>\n<p>\u6df1\u5c64\u5b66\u7fd2\u3067\u306f\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u304b\u3089\u7121\u4f5c\u70ba\u306b\u62bd\u51fa\u3057\u305f\u30b5\u30d6\u30bb\u30c3\u30c8\uff08\u30d0\u30c3\u30c1\uff09\u3092\u7528\u3044\u3066\u8a13\u7df4\u3059\u308b\u3053\u3068\u304c\u3088\u304f\u3042\u308a\u307e\u3059\u3002Array \u306b\u5bfe\u3057\u3066\u7121\u4f5c\u70ba\u62bd\u51fa\u3059\u308b\u65b9\u6cd5\u3092\u4e8c\u3064\u7d39\u4ecb\u3057\u307e\u3059\u3002<\/p>\n<h3>\u65b9\u6cd5\uff11<\/h3>\n<p>\u4e00\u3064\u76ee\u306f rand \u30af\u30ec\u30fc\u30c8\u3092\u7528\u3044\u3066\u30a4\u30f3\u30c7\u30c3\u30af\u30b9\u914d\u5217\u3092\u751f\u6210\u3057\u3001Array::select \u3092\u7528\u3044\u308b\u65b9\u6cd5\u3067\u3059\u3002\u4eca\u56de\u306f\u3053\u3061\u3089\u3092\u4f7f\u7528\u3057\u307e\u3057\u305f\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"c\">\/\/ ArrayBase::select<\/span>\r\n<span class=\"k\">pub<\/span> <span class=\"k\">fn<\/span> <span class=\"nf\">select<\/span><span class=\"p\">(<\/span><span class=\"o\">&amp;<\/span><span class=\"k\">self<\/span><span class=\"p\">,<\/span> <span class=\"n\">axis<\/span><span class=\"p\">:<\/span> <span class=\"n\">Axis<\/span><span class=\"p\">,<\/span> <span class=\"n\">indices<\/span><span class=\"p\">:<\/span> <span class=\"o\">&amp;<\/span><span class=\"p\">[<\/span><span class=\"n\">Ix<\/span><span class=\"p\">])<\/span> <span class=\"k\">-&gt;<\/span> <span class=\"n\">Array<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">A<\/span><span class=\"p\">,<\/span> <span class=\"n\">D<\/span><span class=\"o\">&gt;<\/span>\r\n<span class=\"k\">where<\/span>\r\n    <span class=\"n\">A<\/span><span class=\"p\">:<\/span> <span class=\"nb\">Copy<\/span><span class=\"p\">,<\/span>\r\n    <span class=\"n\">S<\/span><span class=\"p\">:<\/span> <span class=\"n\">Data<\/span><span class=\"p\">,<\/span>\r\n    <span class=\"n\">D<\/span><span class=\"p\">:<\/span> <span class=\"n\">RemoveAxis<\/span><span class=\"p\">,<\/span> \r\n<\/code><\/pre>\n<p>\u4f8b\u3048\u3070\u6b21\u306e\u3088\u3046\u306b\u7528\u3044\u307e\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"k\">use<\/span> <span class=\"nn\">rand<\/span><span class=\"p\">::<\/span><span class=\"nn\">prelude<\/span><span class=\"p\">::<\/span><span class=\"o\">*<\/span><span class=\"p\">;<\/span>\r\n\r\n<span class=\"k\">const<\/span> <span class=\"n\">BATCH_SIZE<\/span><span class=\"p\">:<\/span> <span class=\"nb\">usize<\/span> <span class=\"o\">=<\/span> <span class=\"mi\">100<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">const<\/span> <span class=\"n\">NBR_TRAIN_IMAGES<\/span><span class=\"p\">:<\/span> <span class=\"nb\">usize<\/span> <span class=\"o\">=<\/span> <span class=\"mi\">60000<\/span><span class=\"p\">;<\/span>\r\n<span class=\"k\">fn<\/span> <span class=\"nf\">main<\/span><span class=\"p\">()<\/span> <span class=\"p\">{<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"n\">data_set<\/span><span class=\"p\">:<\/span> <span class=\"n\">Array2<\/span><span class=\"o\">&lt;<\/span><span class=\"nb\">f32<\/span><span class=\"o\">&gt;<\/span> <span class=\"o\">=<\/span> <span class=\"o\">...<\/span><span class=\"p\">;<\/span>\r\n    <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\">indices<\/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=\"nd\">vec!<\/span><span class=\"p\">[<\/span><span class=\"mi\">0u<\/span><span class=\"n\">size<\/span><span class=\"p\">;<\/span> <span class=\"n\">BATCH_SIZE<\/span><span class=\"p\">];<\/span>\r\n    <span class=\"k\">for<\/span> <span class=\"n\">jj<\/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=\"n\">indices<\/span><span class=\"p\">[<\/span><span class=\"n\">jj<\/span><span class=\"p\">]<\/span> <span class=\"o\">=<\/span> <span class=\"n\">rng<\/span><span class=\"nf\">.gen_range<\/span><span class=\"p\">(<\/span><span class=\"mi\">0<\/span><span class=\"p\">,<\/span> <span class=\"n\">NBR_TRAIN_IMAGES<\/span><span class=\"p\">);<\/span>\r\n    <span class=\"p\">}<\/span>\r\n    <span class=\"k\">let<\/span> <span class=\"n\">batch<\/span><span class=\"p\">:<\/span> <span class=\"n\">Array2<\/span><span class=\"o\">&lt;<\/span><span class=\"nb\">f32<\/span><span class=\"o\">&gt;<\/span> <span class=\"o\">=<\/span> <span class=\"n\">data_set<\/span><span class=\"nf\">.select<\/span><span class=\"p\">(<\/span><span class=\"nf\">Axis<\/span><span class=\"p\">(<\/span><span class=\"mi\">0<\/span><span class=\"p\">),<\/span> <span class=\"o\">&amp;<\/span><span class=\"n\">indices<\/span><span class=\"p\">);<\/span>\r\n<span class=\"p\">}<\/span>\r\n<\/code><\/pre>\n<h3>\u65b9\u6cd5\uff12<\/h3>\n<p>\u4e8c\u3064\u76ee\u306f ndarray-rand \u306e sample_axis \u3092\u4f7f\u3046\u65b9\u6cd5\u3067\u3059\u3002\u3053\u3061\u3089\u306f ndarray \u30ea\u30dd\u30b8\u30c8\u30ea\u5185\u3067\u7ba1\u7406\u3055\u308c\u3066\u3044\u308b\u30af\u30ec\u30fc\u30c8\u3067\u3059\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"k\">pub<\/span> <span class=\"k\">fn<\/span> <span class=\"nf\">sample_axis<\/span><span class=\"p\">(<\/span>\r\n    <span class=\"o\">&amp;<\/span><span class=\"k\">self<\/span><span class=\"p\">,<\/span>\r\n    <span class=\"n\">axis<\/span><span class=\"p\">:<\/span> <span class=\"n\">Axis<\/span><span class=\"p\">,<\/span>\r\n    <span class=\"n\">n_samples<\/span><span class=\"p\">:<\/span> <span class=\"nb\">usize<\/span><span class=\"p\">,<\/span>\r\n    <span class=\"n\">strategy<\/span><span class=\"p\">:<\/span> <span class=\"n\">SamplingStrategy<\/span>\r\n<span class=\"p\">)<\/span> <span class=\"k\">-&gt;<\/span> <span class=\"n\">Array<\/span><span class=\"o\">&lt;<\/span><span class=\"n\">A<\/span><span class=\"p\">,<\/span> <span class=\"n\">D<\/span><span class=\"o\">&gt;<\/span>\r\n<span class=\"k\">where<\/span>\r\n    <span class=\"n\">A<\/span><span class=\"p\">:<\/span> <span class=\"nb\">Copy<\/span><span class=\"p\">,<\/span>\r\n    <span class=\"n\">D<\/span><span class=\"p\">:<\/span> <span class=\"n\">RemoveAxis<\/span><span class=\"p\">,<\/span> \r\n<\/code><\/pre>\n<p>\u3053\u3061\u3089\u306e\u4f7f\u7528\u4f8b\u306f Docs.rs \u3084 ndarray-rand \u30ea\u30dd\u30b8\u30c8\u30ea\u306e tests\/tests.rs \u306b\u63b2\u8f09\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n<h1>\u30d9\u30f3\u30c1\u30de\u30fc\u30af<\/h1>\n<p>Rust \u3067\u4e09\u5c64\u30d1\u30fc\u30bb\u30d7\u30c8\u30ed\u30f3\u30e2\u30c7\u30eb\u3092\u5b9f\u88c5\u3057\u307e\u3057\u305f\u304c\u3001\u5143\u306e\u30b3\u30fc\u30c9\u3067\u306e\u5b9f\u88c5\u3068\u905c\u8272\u306a\u3044\u6027\u80fd\u3092\u6709\u3057\u3066\u307b\u3057\u3044\u3068\u3053\u308d\u3067\u3059\u3002\u305d\u3053\u3067\u4ee5\u4e0b\u306e\u30e2\u30c7\u30eb\u3092\u3001\u8a13\u7df4\u6642\u9593\u3068\u78ba\u5ea6\uff08accuracy\uff09\u306e\u89b3\u70b9\u3067\u6bd4\u8f03\u3057\u307e\u3057\u305f\u3002<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">Rust \u3067\u30c7\u30fc\u30bf\u69cb\u9020\u306b ndarray \u3092\u63a1\u7528\u3057\u3066\u5b9f\u88c5\u3057\u305f\u3082\u306e<\/ul>\n<\/li>\n<\/ul>\n<p>\u672c\u30ea\u30dd\u30b8\u30c8\u30ea\u306e ch05 \u306b\u5165\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>src\/main.rs \u3067 rs_deep::dlfs01::ch05::train_neural_net::main(); \u4ee5\u5916\u3092\u30b3\u30e1\u30f3\u30c8\u30a2\u30a6\u30c8\u3057\u305f\u72b6\u614b\u3067\u3001\u30ea\u30dd\u30b8\u30c8\u30ea\u306e\u30d9\u30fc\u30b9\u30c7\u30a3\u30ec\u30af\u30c8\u30ea\u5185\u3067 cargo run &#8211;release \u3092\u5b9f\u884c\u3059\u308c\u3070\u52d5\u304d\u307e\u3059\u3002<\/p>\n<p>\u5143\u306e Python \u30b3\u30fc\u30c9\u3067\u306e\u5b9f\u88c5<\/p>\n<p>\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u7528\u30b9\u30af\u30ea\u30d7\u30c8\u306f\u3053\u3061\u3089\u3067\u3059\u3002<\/p>\n<p>Keras in TensorFlow \u3092\u7528\u3044\u3066\u5b9f\u88c5\u3057\u305f\u3082\u306e<\/p>\n<p>\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u7528\u30b9\u30af\u30ea\u30d7\u30c8\u306f\u3053\u3061\u3089\u3067\u3059\u3002<\/p>\n<h2>\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306b\u7528\u3044\u305f\u74b0\u5883<\/h2>\n<div>\n<div class=\"post-table\">\u5bfe\u8c61\u5024CPUIntel(R) Core(TM) i7-10510U @ 1.80 GHzRAM8.00 GBOSUbuntu 20.04 on WSL2\u5143\u306e\u30b3\u30fc\u30c9\u7528 PythonPython 3.7.9, NumPy 1.19.2Keras in TensorFlow \u3067\u306e\u5b9f\u88c5\u7528 PythonPython 3.7.9, NumPy 1.19.2, TensorFlow 2.3.0Rust \u7528rustup 1.23.0, rustc 1.48.0<\/div>\n<\/div>\n<h2>\u30d1\u30e9\u30e1\u30fc\u30bf<\/h2>\n<div>\n<div class=\"post-table\">\u30d1\u30e9\u30e1\u30fc\u30bf\u5024\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8<a href=\"http:\/\/yann.lecun.com\/exdb\/mnist\/\" target=\"_blank\" rel=\"nofollow noopener\">MNIST<\/a>\u5165\u529b\u5c64\u306e\u30b5\u30a4\u30ba $L_1$784 (=28 \u00d7 28)\u96a0\u308c\u5c64\u306e\u30b5\u30a4\u30ba $L_2$50\u51fa\u529b\u5c64\u306e\u30b5\u30a4\u30ba $L_3$10\u5b66\u7fd2\u7387 $\\epsilon$0.1\u8a13\u7df4\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306e\u657060,000\u30c6\u30b9\u30c8\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306e\u657010,000\u30d0\u30c3\u30c1\u30b5\u30a4\u30ba100\u30a4\u30c6\u30ec\u30fc\u30b7\u30e7\u30f3\u657010,000\u4e71\u6570\u306e\u8a2d\u5b9a\u306a\u3057<\/div>\n<\/div>\n<h2>\u7d50\u679c<\/h2>\n<p>\u6e2c\u5b9a\u7d50\u679c\u306e\u4e00\u4f8b\u3092\u4e0b\u8868\u306b\u793a\u3057\u307e\u3059\u3002\u3053\u3053\u3067<br \/>\n\u8a13\u7df4\u6642\u9593\uff1d\u300c\u30d0\u30c3\u30c1\u30b5\u30a4\u30ba\u5206\u306e\u30c7\u30fc\u30bf\u306e\u9078\u629e\uff0b\u9806\u4f1d\u64ad\uff0b\u9006\u4f1d\u64ad\uff0b\u30d1\u30e9\u30e1\u30fc\u30bf\u66f4\u65b0\u300d\u00d7\u300c\u30a4\u30c6\u30ec\u30fc\u30b7\u30e7\u30f3\u6570\u300d<br \/>\n\u3067\u3059\u3002<\/p>\n<div>\n<div class=\"post-table\">\u5bfe\u8c61CPU \u4f7f\u7528\u7387\u8a13\u7df4\u6642\u9593 (sec)\u78ba\u5ea6\u5143\u306e Python \u30b3\u30fc\u30c9\u223c80%\u223c10train=0.9809, test=0.9708Rust \u3067\u306e\u5b9f\u88c5\uff08Array2 \u3068 ArrayD \u306e\u6df7\u5408\uff09\u223c20%15.06 (debug \u3060\u3068 511)train=0.9437, test=0.934Rust \u3067\u306e\u5b9f\u88c5\uff08\u3059\u3079\u3066 Array2\uff09\u223c20%\u223c10train=0.9479, test=0.9422Keras in TensorFlow\u223c70%\u223c10train=0.9813, test=0.9709<\/div>\n<\/div>\n<p>Array2 \u3068 ArrayD \u306e\u6df7\u5408\u7248\u3067\u306f\u3001\u8a13\u7df4\u6642\u9593\u306f\u7d04 1.5 \u500d\u304b\u304b\u308a\u307e\u3057\u305f\u3002\u78ba\u5ea6\u306f0.035 \u7a0b\u5ea6\u52a3\u308a\u307e\u3059\u3002\u307e\u3060\u5b9f\u88c5\u306b\u6539\u5584\u70b9\u304c\u3042\u308a\u305d\u3046\u3067\u3059\u3002<br \/>\n\uff082020\/12\/06 \u8ffd\u8a18\uff09\u30ec\u30a4\u30e4\u30fc\u304c\u53d7\u3051\u308b\u30c7\u30fc\u30bf\u578b\u3092\u3059\u3079\u3066 Array2 \u306b\u7d71\u4e00\u3059\u308b\u3068\u3001\u8a13\u7df4\u6642\u9593\u304c Python \u3067\u306e\u5b9f\u88c5\u3068\u540c\u7b49\u306b\u306a\u308a\u307e\u3057\u305f\u3002<br \/>\n\u307e\u305f\u3001CPU \u4f7f\u7528\u7387\u304c Python \u306e\u30b3\u30fc\u30c9\u306b\u6bd4\u3079\u3066 1\/4 \u4f4d\u3057\u304b\u3042\u308a\u307e\u305b\u3093\u3002CPU \u3092\u6709\u52b9\u306b\u4f7f\u7528\u3059\u308b\u3068\u3044\u3046\u8ab2\u984c\u3082\u898b\u3048\u3066\u304d\u307e\u3057\u305f\u3002<\/p>\n<p>\u5143\u306e Python \u5b9f\u88c5\u304c Keras \u3067\u306e\u5b9f\u88c5\u3068\u540c\u7a0b\u5ea6\u306e\u8a13\u7df4\u6642\u9593\u3068\u7cbe\u5ea6\u304c\u51fa\u3066\u3044\u308b\u306e\u3082\u8208\u5473\u6df1\u3044\u3067\u3059\u3002<br \/>\n\u203bKeras \u3067\u306e\u5b9f\u88c5\u306f\u4ed6\u306e\u5b9f\u88c5\u3068\u5b8c\u5168\u306b\u540c\u4e00\u306e\u64cd\u4f5c\u304c\u884c\u308f\u308c\u3066\u3044\u308b\u308f\u3051\u3067\u306f\u306a\u3044\u306e\u3067\u3001\u305d\u306e\u8fba\u3067\u591a\u5c11\u306e\u5dee\u304c\u3064\u3044\u3066\u3044\u308b\u304b\u3082\u3057\u308c\u307e\u305b\u3093\u3002<\/p>\n<h1>\u304a\u308f\u308a\u306b<\/h1>\n<p>\u4eca\u56de\u306f\u4e09\u5c64\u30d1\u30fc\u30bb\u30d7\u30c8\u30ed\u30f3\u30e2\u30c7\u30eb\u3092 Rust \u3067\u5b9f\u88c5\u3057\u3001\u305d\u306e\u6027\u80fd\u3092\u5143\u306e\u30b3\u30fc\u30c9\u3084 Keras in TensorFlow \u3067\u306e\u5b9f\u88c5\u3068\u6bd4\u8f03\u3057\u307e\u3057\u305f\u3002<br \/>\n\u5143\u306e\u30b3\u30fc\u30c9\u306b\u6bd4\u3079\u3066\u9045\u3044\u306a\u3069\u3001\u4eca\u56de\u306e\u5b9f\u88c5\u306b\u306f\u6539\u5584\u70b9\u304c\u591a\u6570\u3042\u308a\u307e\u3059\u306e\u3067\u4fee\u6b63\u3092\u9032\u3081\u308b\u4e88\u5b9a\u3067\u3059\u3002<br \/>\n\u4f8b\u3048\u3070 Array2 \u3068 ArrayD \u306e\u9593\u306e\u5909\u63db\uff08into_dimensionality\uff09\u304c\u30d0\u30c3\u30c1\u3054\u3068\u306b\u5165\u308b\u306e\u3067\u3001Array2 \u306b\u7d71\u4e00\u3059\u308b\u3053\u3068\u3067\u901f\u304f\u306a\u308b\u304b\u3082\u3057\u308c\u307e\u305b\u3093\uff08\u30d5\u30ec\u30ad\u30b7\u30d3\u30ea\u30c6\u30a3\u306f\u4e0b\u304c\u308a\u307e\u3059\u304c\uff09\u3002<br \/>\n\uff082020\/12\/06 \u8ffd\u8a18\uff09\u5b9f\u969b\u3001into_dimensionality \u306b\u6642\u9593\u304c\u304b\u304b\u308b\u3088\u3046\u3067\u3059\u3002<br \/>\n\u672c\u8a18\u4e8b\u306e\u57f7\u7b46\u6642\u70b9\u3067\u306f\uff15\u7ae0\uff08\u4e09\u5c64\u30d1\u30fc\u30bb\u30d7\u30c8\u30ed\u30f3\u30e2\u30c7\u30eb\u306e\u5b9f\u88c5\uff09\u307e\u3067\u3057\u304b\u9032\u3081\u3089\u308c\u3066\u304a\u308a\u307e\u305b\u3093\u304c\u3001\u53c2\u8003\u66f8\u306e\u6b8b\u308a\u306e\u7ae0\u3082\u9032\u3081\u3066\u57fa\u672c\u7684\u306a\u30ec\u30a4\u30e4\u30fc\u306e\u5b9f\u88c5\u65b9\u6cd5\u3082\u8eab\u306b\u7740\u3051\u308b\u4e88\u5b9a\u3067\u3059\u3002<\/p>\n<p>\u307e\u305f\u3001\u53c2\u8003\u66f8\u306b\u306f\u4ee5\u4e0b\u306e\u7d9a\u7de8\u304c\u3042\u308a\u307e\u3059\u3002\u3053\u308c\u3089\u3082\u8aad\u3093\u3067 Rust \u3067\u5b9f\u88c5\u3057\u3066\u307f\u3088\u3046\u3068\u8003\u3048\u3066\u3044\u307e\u3059\u3002<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u300e\u30bc\u30ed\u304b\u3089\u4f5c\u308bDeep Learning\u2015\u81ea\u7136\u8a00\u8a9e\u51e6\u7406\u7de8\u300f<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\u300e\u30bc\u30ed\u304b\u3089\u4f5c\u308bDeep Learning\u2015\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u7de8\u300f<\/ul>\n<h1>\u53c2\u8003<\/h1>\n<h2>\u5148\u99c6\u8005\u306e\u65b9\u3005<\/h2>\n<p>\u672c\u8a18\u4e8b\u3068\u540c\u7b49\u306e\u3053\u3068\u306f\u8ab0\u3082\u304c\u8003\u3048\u308b\u3060\u308d\u3046\u3068\u601d\u3063\u3066\u3044\u307e\u3057\u305f\u304c\u3001\u3084\u306f\u308a\u5148\u99c6\u8005\u304c\u3044\u3089\u3063\u3057\u3083\u3044\u307e\u3057\u305f\u3002\u305d\u306e\u3046\u3061 Qiita \u5185\u306b\u6295\u7a3f\u3055\u308c\u3066\u3044\u308b\u8a18\u4e8b\u3092\u7d39\u4ecb\u3044\u305f\u3057\u307e\u3059\u3002<\/p>\n<p>[WIP]Rust\u3067\u300c\u30bc\u30ed\u304b\u3089\u3064\u304f\u308bDeep Learning\u300d\uff08eielh\u6c0f\uff09<\/p>\n<p>\u30bc\u30ed\u304b\u3089\u4f5c\u308bDeepLearning by Rust\uff08\u7b2c\u4e09\u7ae0\u307e\u3067\uff09\uff08tkyk0317\u6c0f\uff09<\/p>\n<p>Rust\u3067DeepLearning\u5165\u9580\uff08ta_to_co\u6c0f\uff09<\/p>\n<p>ArrayView \u306e\u53c2\u7167\u3092\u5f15\u6570\u306b\u3068\u308b\u5f62\u5f0f\u3067\u95a2\u6570\u304c\u5b9f\u88c5\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n<h2>Rust\u3067\u4f7f\u3048\u308b\u6df1\u5c64\u5b66\u7fd2\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af<\/h2>\n<p>Rust \u3067\u4f7f\u3048\u308b\u6df1\u5c64\u5b66\u7fd2\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306b\u306f\u3001\u4f8b\u3048\u3070\u4ee5\u4e0b\u306e\u3082\u306e\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n<p>Leaf<\/p>\n<p>\u73fe\u5728\u306f\u958b\u767a\u304c\u6b62\u307e\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>juice<\/p>\n<p>\u4e0a\u8a18 Leaf \u306e\u5f8c\u7d99\u306e\u3088\u3046\u3067\u3059\u3002<\/p>\n<p>primitiv-rust<\/p>\n<p>primitiv \u306e Rust binding \u3067\u3059\u3002<\/p>\n<p>\u3053\u3061\u3089\u306e Qiita \u306e\u8a18\u4e8b\u306b\u3066\u7d39\u4ecb\u3055\u308c\u3066\u3044\u307e\u3059\u3002<br \/>\n\u3053\u3061\u3089\u3082\u6700\u8fd1\u306f\u958b\u767a\u304c\u9032\u3093\u3067\u3044\u306a\u3044\u3088\u3046\u3067\u3059\u3002<\/p>\n<p>Rust language bindings for TensorFlow<\/p>\n<p>\u516c\u5f0f\u306e Rust binding \u3067\u3059\u3002<br \/>\n\u30b3\u30a2\u306a\u90e8\u5206\u306f\u958b\u767a\u304c\u9032\u3093\u3067\u3044\u308b\u3088\u3046\u3067\u3059\u304c\u3001examples \u304c\u5145\u5b9f\u3057\u3066\u3044\u307e\u305b\u3093\u3002<br \/>\n\u307e\u305f\u3001examples \u3092\u52d5\u304b\u3059\u305f\u3081\u306e\u30e2\u30c7\u30eb\u306f Python \u3067\u30b3\u30f3\u30d1\u30a4\u30eb\u3055\u308c\u308b\u3088\u3046\u3067\u3059\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u672c\u8a18\u4e8b\u306f Rust Advent Calendar 2020 \u306e\uff16\u65e5\u76ee\u306e\u8a18\u4e8b\u3067\u3059\u3002 \u524d\u65e5\u306e\u8a18\u4e8b\u306fcipepse [&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-45740","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\/45740-2\/\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:description\" content=\"\u672c\u8a18\u4e8b\u306f Rust Advent Calendar 2020 \u306e\uff16\u65e5\u76ee\u306e\u8a18\u4e8b\u3067\u3059\u3002 \u524d\u65e5\u306e\u8a18\u4e8b\u306fcipepse [&hellip;]\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.silicloud.com\/zh\/blog\/45740-2\/\" \/>\n<meta property=\"og:site_name\" content=\"Blog - Silicon Cloud\" \/>\n<meta property=\"article:published_time\" content=\"2023-07-17T01:41:30+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-05-04T07:15:14+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/cdn.silicloud.com\/blog-img\/blog\/img\/657d60e437434c4406cfc4b3\/24-0.png\" \/>\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=\"5 \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\/45740-2\/\",\"url\":\"https:\/\/www.silicloud.com\/zh\/blog\/45740-2\/\",\"name\":\"- Blog - Silicon Cloud\",\"isPartOf\":{\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#website\"},\"datePublished\":\"2023-07-17T01:41:30+00:00\",\"dateModified\":\"2024-05-04T07:15:14+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\/45740-2\/\"]}]},{\"@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\/45740-2\/#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\/45740-2\/","og_locale":"zh_CN","og_type":"article","og_description":"\u672c\u8a18\u4e8b\u306f Rust Advent Calendar 2020 \u306e\uff16\u65e5\u76ee\u306e\u8a18\u4e8b\u3067\u3059\u3002 \u524d\u65e5\u306e\u8a18\u4e8b\u306fcipepse [&hellip;]","og_url":"https:\/\/www.silicloud.com\/zh\/blog\/45740-2\/","og_site_name":"Blog - Silicon Cloud","article_published_time":"2023-07-17T01:41:30+00:00","article_modified_time":"2024-05-04T07:15:14+00:00","og_image":[{"url":"https:\/\/cdn.silicloud.com\/blog-img\/blog\/img\/657d60e437434c4406cfc4b3\/24-0.png"}],"author":"\u6587, \u7fd4","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"\u6587, \u7fd4","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"5 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.silicloud.com\/zh\/blog\/45740-2\/","url":"https:\/\/www.silicloud.com\/zh\/blog\/45740-2\/","name":"- Blog - Silicon Cloud","isPartOf":{"@id":"https:\/\/www.silicloud.com\/zh\/blog\/#website"},"datePublished":"2023-07-17T01:41:30+00:00","dateModified":"2024-05-04T07:15:14+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\/45740-2\/"]}]},{"@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\/45740-2\/#local-main-organization-logo","url":"","contentUrl":"","caption":"Blog - Silicon Cloud"}]}},"_links":{"self":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/45740","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=45740"}],"version-history":[{"count":2,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/45740\/revisions"}],"predecessor-version":[{"id":99083,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/45740\/revisions\/99083"}],"wp:attachment":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/media?parent=45740"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/categories?post=45740"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/tags?post=45740"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}