Google Git
Sign in
apache / singa-site / 880502395b7c5f9727ef6c00e72bead0663f1fa3 / . / content / docs / 4.0.0_Viet / autograd / index.html
blob: 9e081e31ad6119671f2a1e88b3fbab76dd20ac29 [file] [log] [blame]
<!DOCTYPE html><html lang="en"><head><meta charSet="utf-8"/><meta http-equiv="X-UA-Compatible" content="IE=edge"/><title>Autograd · Apache SINGA</title><meta name="viewport" content="width=device-width"/><meta name="generator" content="Docusaurus"/><meta name="description" content="&lt;!--- Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to you under the Apache License, Version 2.0 (the &quot;License&quot;); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an &quot;AS IS&quot; BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --&gt;"/><meta name="docsearch:version" content="4.0.0_Viet"/><meta name="docsearch:language" content="en"/><meta property="og:title" content="Autograd · Apache SINGA"/><meta property="og:type" content="website"/><meta property="og:url" content="https://singa.apache.org/"/><meta property="og:description" content="&lt;!--- Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to you under the Apache License, Version 2.0 (the &quot;License&quot;); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an &quot;AS IS&quot; BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --&gt;"/><meta property="og:image" content="https://singa.apache.org/img/singa_twitter_banner.jpeg"/><meta name="twitter:card" content="summary"/><meta name="twitter:image" content="https://singa.apache.org/img/singa_twitter_banner.jpeg"/><link rel="shortcut icon" href="/img/favicon.ico"/><link rel="stylesheet" href="https://cdn.jsdelivr.net/docsearch.js/1/docsearch.min.css"/><link rel="stylesheet" href="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/atom-one-dark.min.css"/><link rel="alternate" type="application/atom+xml" href="https://singa.apache.org/blog/atom.xml" title="Apache SINGA Blog ATOM Feed"/><link rel="alternate" type="application/rss+xml" href="https://singa.apache.org/blog/feed.xml" title="Apache SINGA Blog RSS Feed"/><link rel="stylesheet" href="https://fonts.googleapis.com/css?family=Source+Sans+Pro:400,400i,700"/><link rel="stylesheet" href="https://fonts.googleapis.com/css2?family=Baloo+Paaji+2&amp;family=Source+Sans+Pro:wght@200;300&amp;display=swap"/><script type="text/javascript" src="https://buttons.github.io/buttons.js"></script><script src="https://unpkg.com/vanilla-back-to-top@7.1.14/dist/vanilla-back-to-top.min.js"></script><script>
document.addEventListener('DOMContentLoaded', function() {
addBackToTop(
{"zIndex":100}
)
});
</script><script src="/js/scrollSpy.js"></script><link rel="stylesheet" href="/css/main.css"/><script src="/js/codetabs.js"></script></head><body class="sideNavVisible separateOnPageNav"><div class="fixedHeaderContainer"><div class="headerWrapper wrapper"><header><a href="/"><img class="logo" src="/img/singa.png" alt="Apache SINGA"/></a><a href="/versions"><h3>4.0.0_Viet</h3></a><div class="navigationWrapper navigationSlider"><nav class="slidingNav"><ul class="nav-site nav-site-internal"><li class="siteNavGroupActive"><a href="/docs/4.0.0_Viet/installation" target="_self">Docs</a></li><li class=""><a href="/docs/4.0.0_Viet/source-repository" target="_self">Community</a></li><li class=""><a href="/blog/" target="_self">News</a></li><li class=""><a href="https://apache-singa.readthedocs.io/en/latest/" target="_self">API</a></li><li class="navSearchWrapper reactNavSearchWrapper"><input type="text" id="search_input_react" placeholder="Search" title="Search"/></li><li class=""><a href="https://github.com/apache/singa" target="_self">GitHub</a></li></ul></nav></div></header></div></div><div class="navPusher"><div class="docMainWrapper wrapper"><div class="docsNavContainer" id="docsNav"><nav class="toc"><div class="toggleNav"><section class="navWrapper wrapper"><div class="navBreadcrumb wrapper"><div class="navToggle" id="navToggler"><div class="hamburger-menu"><div class="line1"></div><div class="line2"></div><div class="line3"></div></div></div><h2><i>›</i><span>Guides</span></h2><div class="tocToggler" id="tocToggler"><i class="icon-toc"></i></div></div><div class="navGroups"><div class="navGroup"><h3 class="navGroupCategoryTitle">Getting Started</h3><ul class=""><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/installation">Cài đặt</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/software-stack">Software Stack</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/examples">Ví Dụ</a></li></ul></div><div class="navGroup"><h3 class="navGroupCategoryTitle">Guides</h3><ul class=""><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/device">Device</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/tensor">Tensor</a></li><li class="navListItem navListItemActive"><a class="navItem" href="/docs/4.0.0_Viet/autograd">Autograd</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/optimizer">Optimizer</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/graph">Model</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/onnx">ONNX</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/dist-train">Distributed Training</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/time-profiling">Time Profiling</a></li></ul></div><div class="navGroup"><h3 class="navGroupCategoryTitle">Development</h3><ul class=""><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/downloads">Tải SINGA</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/build">Cài đặt SINGA từ Nguồn (Source)</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/contribute-code">Tham gia viết code</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/contribute-docs">Tham gia chỉnh sửa Hướng Dẫn Sử Dụng</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/how-to-release">Chuẩn bị trước khi phát hành</a></li><li class="navListItem"><a class="navItem" href="/docs/4.0.0_Viet/git-workflow">Quy Trình Sử Dụng Git</a></li></ul></div></div></section></div><script>
var coll = document.getElementsByClassName('collapsible');
var checkActiveCategory = true;
for (var i = 0; i < coll.length; i++) {
var links = coll[i].nextElementSibling.getElementsByTagName('*');
if (checkActiveCategory){
for (var j = 0; j < links.length; j++) {
if (links[j].classList.contains('navListItemActive')){
coll[i].nextElementSibling.classList.toggle('hide');
coll[i].childNodes[1].classList.toggle('rotate');
checkActiveCategory = false;
break;
}
}
}
coll[i].addEventListener('click', function() {
var arrow = this.childNodes[1];
arrow.classList.toggle('rotate');
var content = this.nextElementSibling;
content.classList.toggle('hide');
});
}
document.addEventListener('DOMContentLoaded', function() {
createToggler('#navToggler', '#docsNav', 'docsSliderActive');
createToggler('#tocToggler', 'body', 'tocActive');
var headings = document.querySelector('.toc-headings');
headings && headings.addEventListener('click', function(event) {
var el = event.target;
while(el !== headings){
if (el.tagName === 'A') {
document.body.classList.remove('tocActive');
break;
} else{
el = el.parentNode;
}
}
}, false);
function createToggler(togglerSelector, targetSelector, className) {
var toggler = document.querySelector(togglerSelector);
var target = document.querySelector(targetSelector);
if (!toggler) {
return;
}
toggler.onclick = function(event) {
event.preventDefault();
target.classList.toggle(className);
};
}
});
</script></nav></div><div class="container mainContainer docsContainer"><div class="wrapper"><div class="post"><header class="postHeader"><a class="edit-page-link button" href="https://github.com/apache/singa-doc/blob/master/docs-site/docs/autograd.md" target="_blank" rel="noreferrer noopener">Edit</a><h1 id="__docusaurus" class="postHeaderTitle">Autograd</h1></header><article><div><span><!--- Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to you under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. -->
<p>Có hai cách thường dùng để sử dụng autograd, qua symbolic differentiation như là
<a href="http://deeplearning.net/software/theano/index.html">Theano</a> hoặc reverse
differentiation như là
<a href="https://pytorch.org/docs/stable/notes/autograd.html">Pytorch</a>. SINGA dùng cách
Pytorch, lưu trữ computation graph rồi áp dụng backward propagation tự động sau
khi forward propagation. Thuật toán autograd được giải thích cụ thể ở
<a href="https://pytorch.org/docs/stable/notes/autograd.html">đây</a>. Chúng tôi giải thích
các modules liên quan trong Singa và đưa ra ví dụ để minh hoạ cách sử dụng.</p>
<h2><a class="anchor" aria-hidden="true" id="các-module-liên-quan"></a><a href="#các-module-liên-quan" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Các Module liên quan</h2>
<p>Autograd gồm ba classes với tên gọi <code>singa.tensor.Tensor</code>,
<code>singa.autograd.Operation</code>, và <code>singa.autograd.Layer</code>. Trong phần tiếp theo của
văn kiện này, chúng tôi dùng tensor, operation và layer để chỉ một chương trình
(instance) trong class tương ứng.</p>
<h3><a class="anchor" aria-hidden="true" id="tensor"></a><a href="#tensor" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Tensor</h3>
<p>Ba tính năng của Tensor được sử dụng bởi autograd,</p>
<ul>
<li><code>.creator</code> là một chương trình <code>Operation</code>. Chương trình này lưu trữ tác vụ
tạo ra Tensor instance.</li>
<li><code>.requires_grad</code> là một biến kiểu bool. Biến được sử dụng để chỉ rằng thuật
toán autograd cần tính ra độ dốc (gradient) của tensor. (như owner). Ví dụ,
khi chạy backpropagation, thì cần phải tính ra độ dốc của tensor cho ma trận
trọng lượng (weight matrix) của lớp tuyến tính (linear layer) và bản đồ tính
năng (feature map) của convolution layer (không phải lớp cuối).</li>
<li><code>.stores_grad</code> là một biến kiểu bool. Biến được sử dụng để chỉ rằng độ dốc của
owner tensor cần được lưu và tạo ra bởi hàm backward. Ví dụ, độ dốc của
feature maps được tính thông qua backpropagation, nhưng không được bao gồm
trong kết quả của hàm backward.</li>
</ul>
<p>Lập trình viên có thể thay đổi <code>requires_grad</code> và <code>stores_grad</code> của chương trình
Tensor. Ví dụ nếu hàm sau để là True, độ dốc tương ứng sẽ được bao gồm trong kết
quả của hàm backward. Cần lưu ý rằng nếu <code>stores_grad</code> để là True, thì
<code>requires_grad</code> cũng phải là True, và ngược lại.</p>
<h3><a class="anchor" aria-hidden="true" id="operation"></a><a href="#operation" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Operation</h3>
<p>Hàm chạy một hoặc một vài chương trình <code>Tensor</code> instances ở đầu vào, sau đó đầu
ra là một hoặc một vài chương trình <code>Tensor</code> instances. Ví dụ, hàm ReLU có thể
được sử dụng như một subclass của một hàm Operation cụ thể. Khi gọi một chương
trình <code>Operation</code> (sau cài đặt), cần thực hiện hai bước sau:</p>
<ol>
<li>Ghi lại hàm operations nguồn, vd. biến <code>creator</code>của tensor đầu vào.</li>
<li>làm tính toán bằng cách gọi hàm thành viên <code>.forward()</code></li>
</ol>
<p>Có hai hàm thành viên cho forwarding và backwarding, vd. <code>.forward()</code> và
<code>.backward()</code>. Đầu vào là <code>Tensor.data</code> (thuộc loại <code>CTensor</code>), và đầu ra là
<code>Ctensor</code>. Nếu muốn thêm một hàm operation thì subclass <code>operation</code> cần chạy
riêng <code>.forward()</code> và <code>.backward()</code>. Hàm <code>backward()</code> được tự động gọi bởi hàm
<code>backward()</code> của autograd trong quá trình chạy backward để thực hiện độ dốc của
đầu vào (theo mục <code>require_grad</code>).</p>
<h3><a class="anchor" aria-hidden="true" id="layer"></a><a href="#layer" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Layer</h3>
<p>Với các hàm yêu cầu tham số (parameter), chúng tôi gói chúng lại thành một class
mới, <code>Layer</code>. Ví dụ hàm convolution operation thì được nhóm vào trong
convolution layer. <code>Layer</code> quản lý (hoặc lưu trữ) các tham số và sẽ gọi các hàm
<code>Operation</code> tương ứng để thực hiện việc chuyển đổi.</p>
<h2><a class="anchor" aria-hidden="true" id="ví-dụ"></a><a href="#ví-dụ" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Ví dụ</h2>
<p>Chúng tôi cung cấp nhiều ví dụ trong
<a href="https://github.com/apache/singa/tree/master/examples/autograd">mục ví dụ</a>.
Chúng tôi đưa ra giải thích cụ thể trong hai ví dụ tiêu biểu ở đây.</p>
<h3><a class="anchor" aria-hidden="true" id="dùng-hàm-operation"></a><a href="#dùng-hàm-operation" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Dùng hàm Operation</h3>
<p>Code dưới đây áp dụng model MLP, chỉ dùng hàm Operation (không dùng hàm Layer).</p>
<h4><a class="anchor" aria-hidden="true" id="thêm-packages"></a><a href="#thêm-packages" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Thêm packages</h4>
<pre><code class="hljs css language-python"><span class="hljs-keyword">from</span> singa.tensor <span class="hljs-keyword">import</span> Tensor
<span class="hljs-keyword">from</span> singa <span class="hljs-keyword">import</span> autograd
<span class="hljs-keyword">from</span> singa <span class="hljs-keyword">import</span> opt
</code></pre>
<h4><a class="anchor" aria-hidden="true" id="tạo-ma-trận-trọng-lượng-weight-matrix-và-bias-vector"></a><a href="#tạo-ma-trận-trọng-lượng-weight-matrix-và-bias-vector" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Tạo ma trận trọng lượng (weight matrix) và bias vector</h4>
<p>Tham số tensors được tạo bởi cả <code>requires_grad</code> và <code>stores_grad</code> ở giá trị
<code>True</code>.</p>
<pre><code class="hljs css language-python">w0 = Tensor(shape=(<span class="hljs-number">2</span>, <span class="hljs-number">3</span>), requires_grad=<span class="hljs-literal">True</span>, stores_grad=<span class="hljs-literal">True</span>)
w0.gaussian(<span class="hljs-number">0.0</span>, <span class="hljs-number">0.1</span>)
b0 = Tensor(shape=(<span class="hljs-number">1</span>, <span class="hljs-number">3</span>), requires_grad=<span class="hljs-literal">True</span>, stores_grad=<span class="hljs-literal">True</span>)
b0.set_value(<span class="hljs-number">0.0</span>)
w1 = Tensor(shape=(<span class="hljs-number">3</span>, <span class="hljs-number">2</span>), requires_grad=<span class="hljs-literal">True</span>, stores_grad=<span class="hljs-literal">True</span>)
w1.gaussian(<span class="hljs-number">0.0</span>, <span class="hljs-number">0.1</span>)
b1 = Tensor(shape=(<span class="hljs-number">1</span>, <span class="hljs-number">2</span>), requires_grad=<span class="hljs-literal">True</span>, stores_grad=<span class="hljs-literal">True</span>)
b1.set_value(<span class="hljs-number">0.0</span>)
</code></pre>
<h4><a class="anchor" aria-hidden="true" id="training"></a><a href="#training" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Training</h4>
<pre><code class="hljs css language-python">inputs = Tensor(data=data) <span class="hljs-comment"># data matrix</span>
target = Tensor(data=label) <span class="hljs-comment"># label vector</span>
autograd.training = <span class="hljs-literal">True</span> <span class="hljs-comment"># cho training</span>
sgd = opt.SGD(<span class="hljs-number">0.05</span>) <span class="hljs-comment"># optimizer</span>
<span class="hljs-keyword">for</span> i <span class="hljs-keyword">in</span> range(<span class="hljs-number">10</span>):
x = autograd.matmul(inputs, w0) <span class="hljs-comment"># matrix multiplication</span>
x = autograd.add_bias(x, b0) <span class="hljs-comment"># add the bias vector</span>
x = autograd.relu(x) <span class="hljs-comment"># ReLU activation operation</span>
x = autograd.matmul(x, w1)
x = autograd.add_bias(x, b1)
loss = autograd.softmax_cross_entropy(x, target)
<span class="hljs-keyword">for</span> p, g <span class="hljs-keyword">in</span> autograd.backward(loss):
sgd.update(p, g)
</code></pre>
<h3><a class="anchor" aria-hidden="true" id="hàm-operation--layer"></a><a href="#hàm-operation--layer" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Hàm Operation + Layer</h3>
<p><a href="https://github.com/apache/singa/blob/master/examples/autograd/mnist_cnn.py">Ví dụ</a>
sau đây áp dụng CNN model sử dụng các lớp (layers) tạo từ autograd module.</p>
<h4><a class="anchor" aria-hidden="true" id="tạo-layers"></a><a href="#tạo-layers" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Tạo layers</h4>
<pre><code class="hljs css language-python">conv1 = autograd.Conv2d(<span class="hljs-number">1</span>, <span class="hljs-number">32</span>, <span class="hljs-number">3</span>, padding=<span class="hljs-number">1</span>, bias=<span class="hljs-literal">False</span>)
bn1 = autograd.BatchNorm2d(<span class="hljs-number">32</span>)
pooling1 = autograd.MaxPool2d(<span class="hljs-number">3</span>, <span class="hljs-number">1</span>, padding=<span class="hljs-number">1</span>)
conv21 = autograd.Conv2d(<span class="hljs-number">32</span>, <span class="hljs-number">16</span>, <span class="hljs-number">3</span>, padding=<span class="hljs-number">1</span>)
conv22 = autograd.Conv2d(<span class="hljs-number">32</span>, <span class="hljs-number">16</span>, <span class="hljs-number">3</span>, padding=<span class="hljs-number">1</span>)
bn2 = autograd.BatchNorm2d(<span class="hljs-number">32</span>)
linear = autograd.Linear(<span class="hljs-number">32</span> * <span class="hljs-number">28</span> * <span class="hljs-number">28</span>, <span class="hljs-number">10</span>)
pooling2 = autograd.AvgPool2d(<span class="hljs-number">3</span>, <span class="hljs-number">1</span>, padding=<span class="hljs-number">1</span>)
</code></pre>
<h4><a class="anchor" aria-hidden="true" id="định-nghĩa-hàm-forward"></a><a href="#định-nghĩa-hàm-forward" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Định nghĩa hàm forward</h4>
<p>Hàm trong forward pass sẽ được tự đông lưu cho backward propagation.</p>
<pre><code class="hljs css language-python"><span class="hljs-function"><span class="hljs-keyword">def</span> <span class="hljs-title">forward</span><span class="hljs-params">(x, t)</span>:</span>
<span class="hljs-comment"># x là input data (batch hình ảnh)</span>
<span class="hljs-comment"># t là label vector (batch số nguyên)</span>
y = conv1(x) <span class="hljs-comment"># Conv layer</span>
y = autograd.relu(y) <span class="hljs-comment"># ReLU operation</span>
y = bn1(y) <span class="hljs-comment"># BN layer</span>
y = pooling1(y) <span class="hljs-comment"># Pooling Layer</span>
<span class="hljs-comment"># hai convolution layers song song</span>
y1 = conv21(y)
y2 = conv22(y)
y = autograd.cat((y1, y2), <span class="hljs-number">1</span>) <span class="hljs-comment"># cat operation</span>
y = autograd.relu(y) <span class="hljs-comment"># ReLU operation</span>
y = bn2(y)
y = pooling2(y)
y = autograd.flatten(y) <span class="hljs-comment"># flatten operation</span>
y = linear(y) <span class="hljs-comment"># Linear layer</span>
loss = autograd.softmax_cross_entropy(y, t) <span class="hljs-comment"># operation</span>
<span class="hljs-keyword">return</span> loss, y
</code></pre>
<h4><a class="anchor" aria-hidden="true" id="training-1"></a><a href="#training-1" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Training</h4>
<pre><code class="hljs css language-python">autograd.training = <span class="hljs-literal">True</span>
<span class="hljs-keyword">for</span> epoch <span class="hljs-keyword">in</span> range(epochs):
<span class="hljs-keyword">for</span> i <span class="hljs-keyword">in</span> range(batch_number):
inputs = tensor.Tensor(device=dev, data=x_train[
i * batch_sz:(<span class="hljs-number">1</span> + i) * batch_sz], stores_grad=<span class="hljs-literal">False</span>)
targets = tensor.Tensor(device=dev, data=y_train[
i * batch_sz:(<span class="hljs-number">1</span> + i) * batch_sz], requires_grad=<span class="hljs-literal">False</span>, stores_grad=<span class="hljs-literal">False</span>)
loss, y = forward(inputs, targets) <span class="hljs-comment"># forward the net</span>
<span class="hljs-keyword">for</span> p, gp <span class="hljs-keyword">in</span> autograd.backward(loss): <span class="hljs-comment"># auto backward</span>
sgd.update(p, gp)
</code></pre>
<h3><a class="anchor" aria-hidden="true" id="sử-dụng-model-api"></a><a href="#sử-dụng-model-api" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Sử dụng Model API</h3>
<p><a href="https://github.com/apache/singa/blob/master/examples/cnn/model/cnn.py">Ví dụ</a>
sau áp dụng CNN model sử dụng <a href="./graph">Model API</a>.</p>
<h4><a class="anchor" aria-hidden="true" id="định-nghiã-subclass-của-model"></a><a href="#định-nghiã-subclass-của-model" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Định nghiã subclass của Model</h4>
<p>Model class được định nghĩa là subclass của Model. Theo đó, tất cả các hàm
operations được sử dụng trong bước training sẽ tạo thành một computational graph
và được phân tích. Hàm operation trong graph sẽ được lên lịch trình và thực hiện
hiệu quả. Layers cũng có thể được bao gồm trong model class.</p>
<pre><code class="hljs css language-python"><span class="hljs-class"><span class="hljs-keyword">class</span> <span class="hljs-title">MLP</span><span class="hljs-params">(model.Model)</span>:</span> <span class="hljs-comment"># model là subclass của Model</span>
<span class="hljs-function"><span class="hljs-keyword">def</span> <span class="hljs-title">__init__</span><span class="hljs-params">(self, data_size=<span class="hljs-number">10</span>, perceptron_size=<span class="hljs-number">100</span>, num_classes=<span class="hljs-number">10</span>)</span>:</span>
super(MLP, self).__init__()
<span class="hljs-comment"># taọ operators, layers và các object khác</span>
self.relu = layer.ReLU()
self.linear1 = layer.Linear(perceptron_size)
self.linear2 = layer.Linear(num_classes)
self.softmax_cross_entropy = layer.SoftMaxCrossEntropy()
<span class="hljs-function"><span class="hljs-keyword">def</span> <span class="hljs-title">forward</span><span class="hljs-params">(self, inputs)</span>:</span> <span class="hljs-comment"># định nghĩa forward function</span>
y = self.linear1(inputs)
y = self.relu(y)
y = self.linear2(y)
<span class="hljs-keyword">return</span> y
<span class="hljs-function"><span class="hljs-keyword">def</span> <span class="hljs-title">train_one_batch</span><span class="hljs-params">(self, x, y)</span>:</span>
out = self.forward(x)
loss = self.softmax_cross_entropy(out, y)
self.optimizer(loss)
<span class="hljs-keyword">return</span> out, loss
<span class="hljs-function"><span class="hljs-keyword">def</span> <span class="hljs-title">set_optimizer</span><span class="hljs-params">(self, optimizer)</span>:</span> <span class="hljs-comment"># đính kèm optimizer</span>
self.optimizer = optimizer
</code></pre>
<h4><a class="anchor" aria-hidden="true" id="training-2"></a><a href="#training-2" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Training</h4>
<pre><code class="hljs css language-python"><span class="hljs-comment"># tạo hàm model instance</span>
model = MLP()
<span class="hljs-comment"># tạo optimizer và đính vào model</span>
sgd = opt.SGD(lr=<span class="hljs-number">0.005</span>, momentum=<span class="hljs-number">0.9</span>, weight_decay=<span class="hljs-number">1e-5</span>)
model.set_optimizer(sgd)
<span class="hljs-comment"># input và target placeholders cho model</span>
tx = tensor.Tensor((batch_size, <span class="hljs-number">1</span>, IMG_SIZE, IMG_SIZE), dev, tensor.float32)
ty = tensor.Tensor((batch_size, num_classes), dev, tensor.int32)
<span class="hljs-comment"># tổng hợp model trước khi training</span>
model.compile([tx], is_train=<span class="hljs-literal">True</span>, use_graph=<span class="hljs-literal">True</span>, sequential=<span class="hljs-literal">False</span>)
<span class="hljs-comment"># train model theo bước lặp (iterative)</span>
<span class="hljs-keyword">for</span> b <span class="hljs-keyword">in</span> range(num_train_batch):
<span class="hljs-comment"># generate the next mini-batch</span>
x, y = ...
<span class="hljs-comment"># Copy the data into input tensors</span>
tx.copy_from_numpy(x)
ty.copy_from_numpy(y)
<span class="hljs-comment"># Training with one batch</span>
out, loss = model(tx, ty)
</code></pre>
<h4><a class="anchor" aria-hidden="true" id="lưu-model-checkpoint"></a><a href="#lưu-model-checkpoint" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Lưu model checkpoint</h4>
<pre><code class="hljs css language-python"><span class="hljs-comment"># xác định đường dẫn để lưu checkpoint</span>
checkpointpath=<span class="hljs-string">"checkpoint.zip"</span>
<span class="hljs-comment"># lưu checkpoint</span>
model.save_states(fpath=checkpointpath)
</code></pre>
<h4><a class="anchor" aria-hidden="true" id="tải-model-checkpoint"></a><a href="#tải-model-checkpoint" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Tải model checkpoint</h4>
<pre><code class="hljs css language-python"><span class="hljs-comment"># xác định đường dẫn để lưu checkpoint</span>
checkpointpath=<span class="hljs-string">"checkpoint.zip"</span>
<span class="hljs-comment"># lưu checkpoint</span>
<span class="hljs-keyword">import</span> os
<span class="hljs-keyword">if</span> os.path.exists(checkpointpath):
model.load_states(fpath=checkpointpath)
</code></pre>
<h3><a class="anchor" aria-hidden="true" id="python-api"></a><a href="#python-api" aria-hidden="true" class="hash-link"><svg class="hash-link-icon" aria-hidden="true" height="16" version="1.1" viewBox="0 0 16 16" width="16"><path fill-rule="evenodd" d="M4 9h1v1H4c-1.5 0-3-1.69-3-3.5S2.55 3 4 3h4c1.45 0 3 1.69 3 3.5 0 1.41-.91 2.72-2 3.25V8.59c.58-.45 1-1.27 1-2.09C10 5.22 8.98 4 8 4H4c-.98 0-2 1.22-2 2.5S3 9 4 9zm9-3h-1v1h1c1 0 2 1.22 2 2.5S13.98 12 13 12H9c-.98 0-2-1.22-2-2.5 0-.83.42-1.64 1-2.09V6.25c-1.09.53-2 1.84-2 3.25C6 11.31 7.55 13 9 13h4c1.45 0 3-1.69 3-3.5S14.5 6 13 6z"></path></svg></a>Python API</h3>
<p>Xem
<a href="https://singa.readthedocs.io/en/latest/autograd.html#module-singa.autograd">tại đây</a>
để thêm thông tin chi tiết về Python API.</p>
</span></div></article></div><div class="docs-prevnext"><a class="docs-prev button" href="/docs/4.0.0_Viet/tensor"><span class="arrow-prev">← </span><span>Tensor</span></a><a class="docs-next button" href="/docs/4.0.0_Viet/optimizer"><span>Optimizer</span><span class="arrow-next"> →</span></a></div></div></div><nav class="onPageNav"><ul class="toc-headings"><li><a href="#các-module-liên-quan">Các Module liên quan</a><ul class="toc-headings"><li><a href="#tensor">Tensor</a></li><li><a href="#operation">Operation</a></li><li><a href="#layer">Layer</a></li></ul></li><li><a href="#ví-dụ">Ví dụ</a><ul class="toc-headings"><li><a href="#dùng-hàm-operation">Dùng hàm Operation</a></li><li><a href="#hàm-operation--layer">Hàm Operation + Layer</a></li><li><a href="#sử-dụng-model-api">Sử dụng Model API</a></li><li><a href="#python-api">Python API</a></li></ul></li></ul></nav></div><footer class="nav-footer" id="footer"><section class="sitemap"><a href="/" class="nav-home"><img src="/img/singa-logo-square.png" alt="Apache SINGA" width="66" height="58"/></a><div><h5>Docs</h5><a href="/docs/installation">Getting Started</a><a href="/docs/device">Guides</a><a href="/en/https://apache-singa.readthedocs.io/en/latest/">API Reference</a><a href="/docs/examples">Examples</a><a href="/docs/download-singa">Development</a></div><div><h5>Community</h5><a href="/en/users.html">User Showcase</a><a href="/docs/history-singa">SINGA History</a><a href="/docs/team-list">SINGA Team</a><a href="/blog">SINGA News</a><a href="https://github.com/apache/singa">GitHub</a><div class="social"><a class="github-button" href="https://github.com/apache/singa" data-count-href="/apache/singa/stargazers" data-show-count="true" data-count-aria-label="# stargazers on GitHub" aria-label="Star this project on GitHub">apache/singa-doc</a></div><div class="social"><a href="https://twitter.com/ApacheSINGA" class="twitter-follow-button">Follow @ApacheSINGA</a></div></div><div><h5>Apache Software Foundation</h5><a href="https://apache.org/" target="_blank" rel="noreferrer noopener">Foundation</a><a href="http://www.apache.org/licenses/" target="_blank" rel="noreferrer noopener">License</a><a href="http://www.apache.org/foundation/sponsorship.html" target="_blank" rel="noreferrer noopener">Sponsorship</a><a href="http://www.apache.org/foundation/thanks.html" target="_blank" rel="noreferrer noopener">Thanks</a><a href="http://www.apache.org/events/current-event" target="_blank" rel="noreferrer noopener">Events</a><a href="http://www.apache.org/security/" target="_blank" rel="noreferrer noopener">Security</a></div></section><div style="width:100%;text-align:center"><a href="https://apache.org/" target="_blank" rel="noreferrer noopener" class="ApacheOpenSource"><img src="/img/asf_logo_wide.svg" alt="Apache Open Source"/></a><section class="copyright" style="max-width:60%;margin:0 auto">Copyright © 2023
The Apache Software Foundation. All rights reserved.
Apache SINGA, Apache, the Apache feather logo, and
the Apache SINGA project logos are trademarks of The
Apache Software Foundation. All other marks mentioned
may be trademarks or registered trademarks of their
respective owners.</section></div></footer></div><script type="text/javascript" src="https://cdn.jsdelivr.net/docsearch.js/1/docsearch.min.js"></script><script>window.twttr=(function(d,s, id){var js,fjs=d.getElementsByTagName(s)[0],t=window.twttr||{};if(d.getElementById(id))return t;js=d.createElement(s);js.id=id;js.src='https://platform.twitter.com/widgets.js';fjs.parentNode.insertBefore(js, fjs);t._e = [];t.ready = function(f) {t._e.push(f);};return t;}(document, 'script', 'twitter-wjs'));</script><script>
document.addEventListener('keyup', function(e) {
if (e.target !== document.body) {
return;
}
// keyCode for '/' (slash)
if (e.keyCode === 191) {
const search = document.getElementById('search_input_react');
search && search.focus();
}
});
</script><script>
var search = docsearch({
apiKey: '45202133606c0b5fa6d21cddc4725dd8',
indexName: 'apache_singa',
inputSelector: '#search_input_react',
algoliaOptions: {"facetFilters":["language:en","version:3.0.0"]}
});
</script></body></html>