Skip to content

P
Paddle

PaddlePaddle / Paddle
大约 1 年 前同步成功

通知 2299
Star 20931
Fork 5422
P
Paddle
切换分支/标签
查找文件 普通视图历史永久链接Permalink
sequence_decoder.html 36.1 KB
Newer Older
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 


<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
  <meta charset="utf-8">
  
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  
  <title>Design: Sequence Decoder Generating LoDTensors &mdash; PaddlePaddle  documentation</title>
  

  
  

  

  
  
    

  

  
  
    <link rel="stylesheet" href="../../_static/css/theme.css" type="text/css" />
  

  
  
        <link rel="index" title="Index"
              href="../../genindex.html"/>
        <link rel="search" title="Search" href="../../search.html"/>
    <link rel="top" title="PaddlePaddle  documentation" href="../../index.html"/> 

  <link rel="stylesheet" href="https://cdn.jsdelivr.net/perfect-scrollbar/0.6.14/css/perfect-scrollbar.min.css" type="text/css" />
  <link rel="stylesheet" href="../../_static/css/override.css" type="text/css" />
  <script>
  var _hmt = _hmt || [];
  (function() {
    var hm = document.createElement("script");
    hm.src = "//hm.baidu.com/hm.js?b9a314ab40d04d805655aab1deee08ba";
    var s = document.getElementsByTagName("script")[0]; 
    s.parentNode.insertBefore(hm, s);
  })();
  </script>

  

  
  <script src="../../_static/js/modernizr.min.js"></script>

</head>

<body class="wy-body-for-nav" role="document">

  
  <header class="site-header">
    <div class="site-logo">
      <a href="/"><img src="../../_static/images/PP_w.png"></a>
    </div>
    <div class="site-nav-links">
      <div class="site-menu">
        <a class="fork-on-github" href="https://github.com/PaddlePaddle/Paddle" target="_blank"><i class="fa fa-github"></i>Fork me on Github</a>
        <div class="language-switcher dropdown">
          <a type="button" data-toggle="dropdown">
            <span>English</span>
            <i class="fa fa-angle-up"></i>
            <i class="fa fa-angle-down"></i>
          </a>
          <ul class="dropdown-menu">
            <li><a href="/doc_cn">中文</a></li>
            <li><a href="/doc">English</a></li>
          </ul>
        </div>
        <ul class="site-page-links">
          <li><a href="/">Home</a></li>
        </ul>
      </div>
      <div class="doc-module">
        
        <ul>
<li class="toctree-l1"><a class="reference internal" href="../../getstarted/index_en.html">GET STARTED</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../howto/index_en.html">HOW TO</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../api/index_en.html">API</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../mobile/index_en.html">MOBILE</a></li>
</ul>

        
<div role="search">
  <form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
    <input type="text" name="q" placeholder="Search docs" />
    <input type="hidden" name="check_keywords" value="yes" />
    <input type="hidden" name="area" value="default" />
  </form>
</div>        
      </div>
    </div>
  </header>
  
  <div class="main-content-wrap">

    
    <nav class="doc-menu-vertical" role="navigation">
        
          
          <ul>
<li class="toctree-l1"><a class="reference internal" href="../../getstarted/index_en.html">GET STARTED</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../getstarted/build_and_install/index_en.html">Install and Build</a><ul>
T
Deploy to GitHub Pages: dc82a30908d0d75948491b0a669abfd690b4acce  
Travis CI 已提交
111 
<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/pip_install_en.html">Install PaddlePaddle Using pip</a></li>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
112 
<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/docker_install_en.html">PaddlePaddle in Docker Containers</a></li>
T
Deploy to GitHub Pages: dc82a30908d0d75948491b0a669abfd690b4acce  
Travis CI 已提交
113 
<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/build_from_source_en.html">Build PaddlePaddle from Sources</a></li>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 
</ul>
</li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../howto/index_en.html">HOW TO</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/cmd_parameter/index_en.html">Set Command-line Parameters</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/use_case_en.html">Use Case</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/arguments_en.html">Argument Outline</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/detail_introduction_en.html">Detail Description</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/cluster/cluster_train_en.html">PaddlePaddle Distributed Training</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/k8s/k8s_en.html">Paddle On Kubernetes</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/k8s/k8s_aws_en.html">Distributed PaddlePaddle Training on AWS with Kubernetes</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/build_en.html">Build PaddlePaddle from Source Code and Run Unit Test</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/new_layer_en.html">Write New Layers</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/contribute_to_paddle_en.html">Contribute Code</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/deep_model/rnn/index_en.html">RNN Models</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/rnn_config_en.html">RNN Configuration</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/optimization/gpu_profiling_en.html">Tune GPU Performance</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../api/index_en.html">API</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../api/v2/model_configs.html">Model Configuration</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/activation.html">Activation</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/layer.html">Layers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/evaluators.html">Evaluators</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/optimizer.html">Optimizer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/pooling.html">Pooling</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/networks.html">Networks</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/config/attr.html">Parameter Attribute</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../api/v2/data.html">Data Reader Interface and DataSets</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/data/data_reader.html">Data Reader Interface</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/data/image.html">Image Interface</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../api/v2/data/dataset.html">Dataset</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../api/v2/run_logic.html">Training and Inference</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../mobile/index_en.html">MOBILE</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../mobile/cross_compiling_for_android_en.html">Build PaddlePaddle for Android</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../mobile/cross_compiling_for_raspberry_en.html">Build PaddlePaddle for Raspberry Pi</a></li>
</ul>
</li>
</ul>

        
    </nav>
    
    <section class="doc-content-wrap">

      

 







<div role="navigation" aria-label="breadcrumbs navigation">
  <ul class="wy-breadcrumbs">
      
    <li>Design: Sequence Decoder Generating LoDTensors</li>
  </ul>
</div>
      
      <div class="wy-nav-content" id="doc-content">
        <div class="rst-content">
          <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
           <div itemprop="articleBody">
            
  <div class="section" id="design-sequence-decoder-generating-lodtensors">
<span id="design-sequence-decoder-generating-lodtensors"></span><h1>Design: Sequence Decoder Generating LoDTensors<a class="headerlink" href="#design-sequence-decoder-generating-lodtensors" title="Permalink to this headline"></a></h1>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
194 195 
<p>In tasks such as machine translation and visual captioning,
a <a class="reference external" href="https://github.com/PaddlePaddle/book/blob/develop/08.machine_translation/README.md">sequence decoder</a> is necessary to generate sequences, one word at a time.</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
196 197 198 
<p>This documentation describes how to implement the sequence decoder as an operator.</p>
<div class="section" id="beam-search-based-decoder">
<span id="beam-search-based-decoder"></span><h2>Beam Search based Decoder<a class="headerlink" href="#beam-search-based-decoder" title="Permalink to this headline"></a></h2>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
199 200 201 202 203 
<p>The <a class="reference external" href="https://en.wikipedia.org/wiki/Beam_search">beam search algorithm</a> is necessary when generating sequences. It is a heuristic search algorithm that explores the paths by expanding the most promising node in a limited set.</p>
<p>In the old version of PaddlePaddle, the C++ class <code class="docutils literal"><span class="pre">RecurrentGradientMachine</span></code> implements the general sequence decoder based on beam search, due to the complexity involved, the implementation relies on a lot of special data structures that are quite trivial and hard to be customized by users.</p>
<p>There are a lot of heuristic tricks in the sequence generation tasks, so the flexibility of sequence decoder is very important to users.</p>
<p>During the refactoring of PaddlePaddle, some new concepts are proposed such as:  <a class="reference external" href="https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md">LoDTensor</a> and <a class="reference external" href="https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/tensor_array.md">TensorArray</a> that can better support the sequence usage, and they can also help make the implementation of beam search based sequence decoder <strong>more transparent and modular</strong> .</p>
<p>For example, the RNN states, candidates IDs and probabilities of beam search can be represented all as <code class="docutils literal"><span class="pre">LoDTensors</span></code>;
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
204 205 206 207 
the selected candidate&#8217;s IDs in each time step can be stored in a <code class="docutils literal"><span class="pre">TensorArray</span></code>, and <code class="docutils literal"><span class="pre">Packed</span></code> to the sentences translated.</p>
</div>
<div class="section" id="changing-lod-s-absolute-offset-to-relative-offsets">
<span id="changing-lod-s-absolute-offset-to-relative-offsets"></span><h2>Changing LoD&#8217;s absolute offset to relative offsets<a class="headerlink" href="#changing-lod-s-absolute-offset-to-relative-offsets" title="Permalink to this headline"></a></h2>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
208 209 210 211 
<p>The current <code class="docutils literal"><span class="pre">LoDTensor</span></code> is designed to store levels of variable-length sequences. It stores several arrays of integers where each represents a level.</p>
<p>The integers in each level represent the begin and end (not inclusive) offset of a sequence <strong>in the underlying tensor</strong>,
let&#8217;s call this format the <strong>absolute-offset LoD</strong> for clarity.</p>
<p>The relative-offset LoD can retrieve any sequence very quickly but fails to represent empty sequences, for example, a two-level LoD is as follows</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
212 213 214 215 216 217 218 219 220 221 222 
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="p">[[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">9</span><span class="p">]</span>
 <span class="p">[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">9</span><span class="p">]]</span>
</pre></div>
</div>
<p>The first level tells that there are two sequences:</p>
<ul class="simple">
<li>the first&#8217;s offset is <code class="docutils literal"><span class="pre">[0,</span> <span class="pre">3)</span></code></li>
<li>the second&#8217;s offset is <code class="docutils literal"><span class="pre">[3,</span> <span class="pre">9)</span></code></li>
</ul>
<p>while on the second level, there are several empty sequences that both begin and end at <code class="docutils literal"><span class="pre">3</span></code>.
It is impossible to tell how many empty second-level sequences exist in the first-level sequences.</p>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
223 
<p>There are many scenarios that rely on empty sequence representation, for example in machine translation or visual captioning, one instance has no translation or the empty candidate set for a prefix.</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
224 225 226 227 228 229 230 231 232 233 234 
<p>So let&#8217;s introduce another format of LoD,
it stores <strong>the offsets of the lower level sequences</strong> and is called <strong>relative-offset</strong> LoD.</p>
<p>For example, to represent the same sequences of the above data</p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="p">[[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">6</span><span class="p">]</span>
 <span class="p">[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">3</span><span class="p">,</span> <span class="mi">9</span><span class="p">]]</span>
</pre></div>
</div>
<p>the first level represents that there are two sequences,
their offsets in the second-level LoD is <code class="docutils literal"><span class="pre">[0,</span> <span class="pre">3)</span></code> and <code class="docutils literal"><span class="pre">[3,</span> <span class="pre">5)</span></code>.</p>
<p>The second level is the same with the relative offset example because the lower level is a tensor.
It is easy to find out the second sequence in the first-level LoD has two empty sequences.</p>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
235 
<p>The following examples are based on relative-offset LoD.</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
236 237 238 
</div>
<div class="section" id="usage-in-a-simple-machine-translation-model">
<span id="usage-in-a-simple-machine-translation-model"></span><h2>Usage in a simple machine translation model<a class="headerlink" href="#usage-in-a-simple-machine-translation-model" title="Permalink to this headline"></a></h2>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
239 240 
<p>Let&#8217;s start from a simple machine translation model that is simplified from the <a class="reference external" href="https://github.com/PaddlePaddle/book/tree/develop/08.machine_translation">machine translation chapter</a> to draw a blueprint of what a sequence decoder can do and how to use it.</p>
<p>The model has an encoder that learns the semantic vector from a sequence, and a decoder which uses the sequence encoder to generate new sentences.</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 
<p><strong>Encoder</strong></p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">paddle</span> <span class="kn">as</span> <span class="nn">pd</span>

<span class="n">dict_size</span> <span class="o">=</span> <span class="mi">8000</span>
<span class="n">source_dict_size</span> <span class="o">=</span> <span class="n">dict_size</span>
<span class="n">target_dict_size</span> <span class="o">=</span> <span class="n">dict_size</span>
<span class="n">word_vector_dim</span> <span class="o">=</span> <span class="mi">128</span>
<span class="n">encoder_dim</span> <span class="o">=</span> <span class="mi">128</span>
<span class="n">decoder_dim</span> <span class="o">=</span> <span class="mi">128</span>
<span class="n">beam_size</span> <span class="o">=</span> <span class="mi">5</span>
<span class="n">max_length</span> <span class="o">=</span> <span class="mi">120</span>

<span class="c1"># encoder</span>
<span class="n">src_word_id</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">data</span><span class="p">(</span>
    <span class="n">name</span><span class="o">=</span><span class="s1">&#39;source_language_word&#39;</span><span class="p">,</span>
    <span class="nb">type</span><span class="o">=</span><span class="n">pd</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">integer_value_sequence</span><span class="p">(</span><span class="n">source_dict_dim</span><span class="p">))</span>
<span class="n">src_embedding</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">embedding</span><span class="p">(</span><span class="n">size</span><span class="o">=</span><span class="n">source_dict_size</span><span class="p">,</span> <span class="n">size</span><span class="o">=</span><span class="n">word_vector_dim</span><span class="p">)</span>

<span class="n">src_word_vec</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">lookup</span><span class="p">(</span><span class="n">src_embedding</span><span class="p">,</span> <span class="n">src_word_id</span><span class="p">)</span>

<span class="n">encoder_out_seq</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">gru</span><span class="p">(</span><span class="nb">input</span><span class="o">=</span><span class="n">src_word_vec</span><span class="p">,</span> <span class="n">size</span><span class="o">=</span><span class="n">encoder_dim</span><span class="p">)</span>

<span class="n">encoder_ctx</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">last_seq</span><span class="p">(</span><span class="n">encoder_out_seq</span><span class="p">)</span>
<span class="c1"># encoder_ctx_proj is the learned semantic vector</span>
<span class="n">encoder_ctx_proj</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">fc</span><span class="p">(</span>
    <span class="n">encoder_ctx</span><span class="p">,</span> <span class="n">size</span><span class="o">=</span><span class="n">decoder_dim</span><span class="p">,</span> <span class="n">act</span><span class="o">=</span><span class="n">pd</span><span class="o">.</span><span class="n">activation</span><span class="o">.</span><span class="n">Tanh</span><span class="p">(),</span> <span class="n">bias</span><span class="o">=</span><span class="bp">None</span><span class="p">)</span>
</pre></div>
</div>
<p><strong>Decoder</strong></p>
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">generate</span><span class="p">():</span>
    <span class="n">decoder</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">while_loop</span><span class="p">()</span>
    <span class="k">with</span> <span class="n">decoder</span><span class="o">.</span><span class="n">step</span><span class="p">():</span>
        <span class="n">decoder_mem</span> <span class="o">=</span> <span class="n">decoder</span><span class="o">.</span><span class="n">memory</span><span class="p">(</span><span class="n">init</span><span class="o">=</span><span class="n">encoder_ctx</span><span class="p">)</span>  <span class="c1"># mark the memory</span>
        <span class="n">generated_ids</span> <span class="o">=</span> <span class="n">decoder</span><span class="o">.</span><span class="n">memory</span><span class="p">()</span> <span class="c1"># TODO init to batch_size &lt;s&gt;s</span>
        <span class="n">generated_scores</span> <span class="o">=</span> <span class="n">decoder</span><span class="o">.</span><span class="n">memory</span><span class="p">()</span> <span class="c1"># TODO init to batch_size 1s or 0s</span>

        <span class="n">target_word</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">lookup</span><span class="p">(</span><span class="n">trg_embedding</span><span class="p">,</span> <span class="n">gendrated_ids</span><span class="p">)</span>
        <span class="c1"># expand encoder_ctx&#39;s batch to fit target_word&#39;s lod</span>
        <span class="c1"># for example</span>
        <span class="c1"># decoder_mem.lod is</span>
        <span class="c1"># [[0 1 3],</span>
        <span class="c1">#  [0 1 3 6]]</span>
        <span class="c1"># its tensor content is [a1 a2 a3 a4 a5]</span>
        <span class="c1"># which means there are 2 sentences to translate</span>
        <span class="c1">#   - the first sentence has 1 translation prefixes, the offsets are [0, 1)</span>
        <span class="c1">#   - the second sentence has 2 translation prefixes, the offsets are [1, 3) and [3, 6)</span>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
287 
        <span class="c1"># the target_word.lod is</span>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 
        <span class="c1"># [[0, 1, 6]</span>
        <span class="c1">#  [0, 2, 4, 7, 9 12]]</span>
        <span class="c1"># which means 2 sentences to translate, each has 1 and 5 prefixes</span>
        <span class="c1"># the first prefix has 2 candidates</span>
        <span class="c1"># the following has 2, 3, 2, 3 candidates</span>
        <span class="c1"># the encoder_ctx_expanded&#39;s content will be</span>
        <span class="c1"># [a1 a1 a2 a2 a3 a3 a3 a4 a4 a5 a5 a5]</span>
        <span class="n">encoder_ctx_expanded</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">lod_expand</span><span class="p">(</span><span class="n">encoder_ctx</span><span class="p">,</span> <span class="n">target_word</span><span class="p">)</span>
        <span class="n">decoder_input</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">fc</span><span class="p">(</span>
            <span class="n">act</span><span class="o">=</span><span class="n">pd</span><span class="o">.</span><span class="n">activation</span><span class="o">.</span><span class="n">Linear</span><span class="p">(),</span>
            <span class="nb">input</span><span class="o">=</span><span class="p">[</span><span class="n">target_word</span><span class="p">,</span> <span class="n">encoder_ctx</span><span class="p">],</span>
            <span class="n">size</span><span class="o">=</span><span class="mi">3</span> <span class="o">*</span> <span class="n">decoder_dim</span><span class="p">)</span>
        <span class="n">gru_out</span><span class="p">,</span> <span class="n">cur_mem</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">gru_step</span><span class="p">(</span>
            <span class="n">decoder_input</span><span class="p">,</span> <span class="n">mem</span><span class="o">=</span><span class="n">decoder_mem</span><span class="p">,</span> <span class="n">size</span><span class="o">=</span><span class="n">decoder_dim</span><span class="p">)</span>
        <span class="n">scores</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">fc</span><span class="p">(</span>
            <span class="n">gru_out</span><span class="p">,</span>
            <span class="n">size</span><span class="o">=</span><span class="n">trg_dic_size</span><span class="p">,</span>
            <span class="n">bias</span><span class="o">=</span><span class="bp">None</span><span class="p">,</span>
            <span class="n">act</span><span class="o">=</span><span class="n">pd</span><span class="o">.</span><span class="n">activation</span><span class="o">.</span><span class="n">Softmax</span><span class="p">())</span>
        <span class="c1"># K is an config</span>
        <span class="n">topk_scores</span><span class="p">,</span> <span class="n">topk_ids</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">top_k</span><span class="p">(</span><span class="n">scores</span><span class="p">,</span> <span class="n">K</span><span class="p">)</span>
        <span class="n">topk_generated_scores</span> <span class="o">=</span> <span class="n">pd</span><span class="o">.</span><span class="n">add_scalar</span><span class="p">(</span><span class="n">topk_scores</span><span class="p">,</span> <span class="n">generated_scores</span><span class="p">)</span>

        <span class="n">selected_ids</span><span class="p">,</span> <span class="n">selected_generation_scores</span> <span class="o">=</span> <span class="n">decoder</span><span class="o">.</span><span class="n">beam_search</span><span class="p">(</span>
            <span class="n">topk_ids</span><span class="p">,</span> <span class="n">topk_generated_scores</span><span class="p">)</span>

        <span class="c1"># update the states</span>
        <span class="n">decoder_mem</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">cur_mem</span><span class="p">)</span>  <span class="c1"># tells how to update state</span>
        <span class="n">generated_ids</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">selected_ids</span><span class="p">)</span>
        <span class="n">generated_scores</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">selected_generation_scores</span><span class="p">)</span>

        <span class="n">decoder</span><span class="o">.</span><span class="n">output</span><span class="p">(</span><span class="n">selected_ids</span><span class="p">)</span>
        <span class="n">decoder</span><span class="o">.</span><span class="n">output</span><span class="p">(</span><span class="n">selected_generation_scores</span><span class="p">)</span>

<span class="n">translation_ids</span><span class="p">,</span> <span class="n">translation_scores</span> <span class="o">=</span> <span class="n">decoder</span><span class="p">()</span>
</pre></div>
</div>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
325 326 327 
<p>The <code class="docutils literal"><span class="pre">decoder.beam_search</span></code> is an operator that, given the candidates and the scores of translations including the candidates,
returns the result of the beam search algorithm.</p>
<p>In this way, users can customize anything on the input or output of beam search, for example:</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
328 
<ol class="simple">
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
329 330 331 
<li>Make the corresponding elements in <code class="docutils literal"><span class="pre">topk_generated_scores</span></code> zero or some small values, beam_search will discard this candidate.</li>
<li>Remove some specific candidate in <code class="docutils literal"><span class="pre">selected_ids</span></code>.</li>
<li>Get the final <code class="docutils literal"><span class="pre">translation_ids</span></code>, remove the translation sequence in it.</li>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
332 
</ol>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
333 334 335 
<p>The implementation of sequence decoder can reuse the C++ class:  <a class="reference external" href="https://github.com/Superjom/Paddle/blob/68cac3c0f8451fe62a4cdf156747d6dc0ee000b3/paddle/operators/dynamic_recurrent_op.h#L30">RNNAlgorithm</a>,
so the python syntax is quite similar to that of an  <a class="reference external" href="https://github.com/Superjom/Paddle/blob/68cac3c0f8451fe62a4cdf156747d6dc0ee000b3/doc/design/block.md#blocks-with-for-and-rnnop">RNN</a>.</p>
<p>Both of them are two-level <code class="docutils literal"><span class="pre">LoDTensors</span></code>:</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
336 
<ul class="simple">
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
337 338 
<li>The first level represents <code class="docutils literal"><span class="pre">batch_size</span></code> of (source) sentences.</li>
<li>The second level represents the candidate ID sets for translation prefix.</li>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
339 
</ul>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
340 341 342 
<p>For example, 3 source sentences to translate, and has 2, 3, 1 candidates.</p>
<p>Unlike an RNN, in sequence decoder, the previous state and the current state have different LoD and shape, and an <code class="docutils literal"><span class="pre">lod_expand</span></code> operator is used to expand the LoD of the previous state to fit the current state.</p>
<p>For example, the previous state:</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
343 344 345 346 
<ul class="simple">
<li>LoD is <code class="docutils literal"><span class="pre">[0,</span> <span class="pre">1,</span> <span class="pre">3][0,</span> <span class="pre">2,</span> <span class="pre">5,</span> <span class="pre">6]</span></code></li>
<li>content of tensor is <code class="docutils literal"><span class="pre">a1</span> <span class="pre">a2</span> <span class="pre">b1</span> <span class="pre">b2</span> <span class="pre">b3</span> <span class="pre">c1</span></code></li>
</ul>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
347 
<p>the current state is stored in <code class="docutils literal"><span class="pre">encoder_ctx_expanded</span></code>:</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
348 349 350 351 352 353 354 355 356 357 358 359 
<ul class="simple">
<li>LoD is <code class="docutils literal"><span class="pre">[0,</span> <span class="pre">2,</span> <span class="pre">7][0</span> <span class="pre">3</span> <span class="pre">5</span> <span class="pre">8</span> <span class="pre">9</span> <span class="pre">11</span> <span class="pre">11]</span></code></li>
<li>the content is<ul>
<li>a1 a1 a1 (a1 has 3 candidates, so the state should be copied 3 times for each candidates)</li>
<li>a2 a2</li>
<li>b1 b1 b1</li>
<li>b2</li>
<li>b3 b3</li>
<li>None (c1 has 0 candidates, so c1 is dropped)</li>
</ul>
</li>
</ul>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
360 361 
<p>The benefit from the relative offset LoD is that the empty candidate set can be represented naturally.</p>
<p>The status in each time step can be stored in <code class="docutils literal"><span class="pre">TensorArray</span></code>, and <code class="docutils literal"><span class="pre">Pack</span></code>ed to a final LoDTensor. The corresponding syntax is:</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
362 363 364 365 
<div class="highlight-python"><div class="highlight"><pre><span></span><span class="n">decoder</span><span class="o">.</span><span class="n">output</span><span class="p">(</span><span class="n">selected_ids</span><span class="p">)</span>
<span class="n">decoder</span><span class="o">.</span><span class="n">output</span><span class="p">(</span><span class="n">selected_generation_scores</span><span class="p">)</span>
</pre></div>
</div>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
366 367 368 
<p>The <code class="docutils literal"><span class="pre">selected_ids</span></code> are the candidate ids for the prefixes, and will be <code class="docutils literal"><span class="pre">Packed</span></code> by <code class="docutils literal"><span class="pre">TensorArray</span></code> to a two-level <code class="docutils literal"><span class="pre">LoDTensor</span></code>, where the first level represents the source sequences and the second level represents generated sequences.</p>
<p>Packing the <code class="docutils literal"><span class="pre">selected_scores</span></code> will get a <code class="docutils literal"><span class="pre">LoDTensor</span></code> that stores scores of each translation candidate.</p>
<p>Packing the <code class="docutils literal"><span class="pre">selected_generation_scores</span></code> will get a <code class="docutils literal"><span class="pre">LoDTensor</span></code>, and each tail is the probability of the translation.</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
369 370 371 372 373 
</div>
<div class="section" id="lod-and-shape-changes-during-decoding">
<span id="lod-and-shape-changes-during-decoding"></span><h2>LoD and shape changes during decoding<a class="headerlink" href="#lod-and-shape-changes-during-decoding" title="Permalink to this headline"></a></h2>
<p align="center">
  <img src="./images/LOD-and-shape-changes-during-decoding.jpg"/>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
374 
</p><p>According to the image above, the only phase that changes the LoD is beam search.</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
375 376 377 
</div>
<div class="section" id="beam-search-design">
<span id="beam-search-design"></span><h2>Beam search design<a class="headerlink" href="#beam-search-design" title="Permalink to this headline"></a></h2>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
378 
<p>The beam search algorithm will be implemented as one method of the sequence decoder and has 3 inputs:</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
379 
<ol class="simple">
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
380 
<li><code class="docutils literal"><span class="pre">topk_ids</span></code>, the top K candidate ids for each prefix.</li>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
381 382 383 
<li><code class="docutils literal"><span class="pre">topk_scores</span></code>, the corresponding scores for <code class="docutils literal"><span class="pre">topk_ids</span></code></li>
<li><code class="docutils literal"><span class="pre">generated_scores</span></code>, the score of the prefixes.</li>
</ol>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
384 385 
<p>All of these are LoDTensors, so that the sequence affiliation is clear. Beam search will keep a beam for each prefix and select a smaller candidate set for each prefix.</p>
<p>It will return three variables:</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
386 387 388 
<ol class="simple">
<li><code class="docutils literal"><span class="pre">selected_ids</span></code>, the final candidate beam search function selected for the next step.</li>
<li><code class="docutils literal"><span class="pre">selected_scores</span></code>, the scores for the candidates.</li>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
389 
<li><code class="docutils literal"><span class="pre">generated_scores</span></code>, the updated scores for each prefix (with the new candidates appended).</li>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
390 391 392 393 
</ol>
</div>
<div class="section" id="introducing-the-lod-based-pack-and-unpack-methods-in-tensorarray">
<span id="introducing-the-lod-based-pack-and-unpack-methods-in-tensorarray"></span><h2>Introducing the LoD-based <code class="docutils literal"><span class="pre">Pack</span></code> and <code class="docutils literal"><span class="pre">Unpack</span></code> methods in <code class="docutils literal"><span class="pre">TensorArray</span></code><a class="headerlink" href="#introducing-the-lod-based-pack-and-unpack-methods-in-tensorarray" title="Permalink to this headline"></a></h2>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
394 
<p>The <code class="docutils literal"><span class="pre">selected_ids</span></code>, <code class="docutils literal"><span class="pre">selected_scores</span></code> and <code class="docutils literal"><span class="pre">generated_scores</span></code> are LoDTensors that exist at each time step,
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
395 
so it is natural to store them in arrays.</p>
T
Deploy to GitHub Pages: e0e3a8a5bb2a33bf1953c7cebdedcf2ea5869b51  
Travis CI 已提交
396 397 398 
<p>Currently, PaddlePaddle has a module called <code class="docutils literal"><span class="pre">TensorArray</span></code> which can store an array of tensors. It is better to store the results of beam search in a <code class="docutils literal"><span class="pre">TensorArray</span></code>.</p>
<p>The <code class="docutils literal"><span class="pre">Pack</span></code> and <code class="docutils literal"><span class="pre">UnPack</span></code> in <code class="docutils literal"><span class="pre">TensorArray</span></code> are used to pack tensors in the array to an <code class="docutils literal"><span class="pre">LoDTensor</span></code> or split the <code class="docutils literal"><span class="pre">LoDTensor</span></code> to an array of tensors.
It needs some extensions to support the packing or unpacking an array of <code class="docutils literal"><span class="pre">LoDTensors</span></code>.</p>
T
Deploy to GitHub Pages: 53cb4df0a2b9deb6b1fd5e9c8e2027c4ad27b352  
Travis CI 已提交
399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 
</div>
</div>


           </div>
          </div>
          <footer>
  

  <hr/>

  <div role="contentinfo">
    <p>
        &copy; Copyright 2016, PaddlePaddle developers.

    </p>
  </div>
  Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/snide/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>. 

</footer>

        </div>
      </div>

    </section>

  </div>
  


  

    <script type="text/javascript">
        var DOCUMENTATION_OPTIONS = {
            URL_ROOT:'../../',
            VERSION:'',
            COLLAPSE_INDEX:false,
            FILE_SUFFIX:'.html',
            HAS_SOURCE:  true,
            SOURCELINK_SUFFIX: ".txt",
        };
    </script>
      <script type="text/javascript" src="../../_static/jquery.js"></script>
      <script type="text/javascript" src="../../_static/underscore.js"></script>
      <script type="text/javascript" src="../../_static/doctools.js"></script>
      <script type="text/javascript" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
       
  

  
  
    <script type="text/javascript" src="../../_static/js/theme.js"></script>
  
  
  <script src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.7/js/bootstrap.min.js" integrity="sha384-Tc5IQib027qvyjSMfHjOMaLkfuWVxZxUPnCJA7l2mCWNIpG9mGCD8wGNIcPD7Txa" crossorigin="anonymous"></script>
  <script src="https://cdn.jsdelivr.net/perfect-scrollbar/0.6.14/js/perfect-scrollbar.jquery.min.js"></script>
  <script src="../../_static/js/paddle_doc_init.js"></script> 

</body>
</html>