Skip to content

P
Paddle

Crayon鑫 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
“29b500e79aad6c9ddc859aa14156ddd4e156b90c”上不存在“develop/doc/py-modindex.html”
提交 4e88e677 编写于 作者: T Travis CI
浏览文件
操作

Deploy to GitHub Pages: 53cb4df0

上级 3991a9b9
显示空白变更内容
内联 并排
Showing with 1453 addition and 2 deletion
develop/doc/_sources/design/ops/sequence_decoder.md.txt 0 → 100644
浏览文件 @ 4e88e677
# Design: Sequence Decoder Generating LoDTensors
In tasks such as machine translation and image to text,
a [sequence decoder](https://github.com/PaddlePaddle/book/blob/develop/08.machine_translation/README.md) is necessary to generate sequences.
This documentation describes how to implement the sequence decoder as an operator.
## Beam Search based Decoder
The [beam search algorithm](https://en.wikipedia.org/wiki/Beam_search) 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.
In the old version of PaddlePaddle, a C++ class `RecurrentGradientMachine` implements the general sequence decoder based on beam search,
due to the complexity, the implementation relays on a lot of special data structures,
quite trivial and hard to be customized by users.
There are a lot of heuristic tricks in the sequence generation tasks,
so the flexibility of sequence decoder is very important to users.
During PaddlePaddle's refactoring work,
some new concept is proposed such as [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md) and [TensorArray](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/tensor_array.md) that can better support sequence usage,
and they can help to make the implementation of beam search based sequence decoder **more transparent and modular** .
For example, the RNN sates, candidates IDs and probabilities of beam search can be represented as `LoDTensors`;
the selected candidate's IDs in each time step can be stored in a `TensorArray`, and `Packed` to the sentences translated.
## Changing LoD's absolute offset to relative offsets
The current `LoDTensor` is designed to store levels of variable-length sequences,
it stores several arrays of integers each represents a level.
The integers in each level represents the begin and end (not inclusive) offset of a sequence **in the underlying tensor**,
let's call this format the **absolute-offset LoD** for clear.
The relative-offset LoD can fast retrieve any sequence but fails to represent empty sequences, for example, a two-level LoD is as follows
```python
[[0, 3, 9]
[0, 2, 3, 3, 3, 9]]
```
The first level tells that there are two sequences:
- the first's offset is `[0, 3)`
- the second's offset is `[3, 9)`
while on the second level, there are several empty sequences that both begin and end at `3`.
It is impossible to tell how many empty second-level sequences exist in the first-level sequences.
There are many scenarios that relay on empty sequence representation,
such as machine translation or image to text, one instance has no translations or the empty candidate set for a prefix.
So let's introduce another format of LoD,
it stores **the offsets of the lower level sequences** and is called **relative-offset** LoD.
For example, to represent the same sequences of the above data
```python
[[0, 3, 6]
[0, 2, 3, 3, 3, 9]]
```
the first level represents that there are two sequences,
their offsets in the second-level LoD is `[0, 3)` and `[3, 5)`.
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.
The following demos are based on relative-offset LoD.
## Usage in a simple machine translation model
Let's start from a simple machine translation model that is simplified from [machine translation chapter](https://github.com/PaddlePaddle/book/tree/develop/08.machine_translation) to draw a simple blueprint of what a sequence decoder can do and how to use it.
The model has an encoder that learns the semantic vector from a sequence,
and a decoder which uses the sequence decoder to generate new sentences.
**Encoder**
```python
import paddle as pd
dict_size = 8000
source_dict_size = dict_size
target_dict_size = dict_size
word_vector_dim = 128
encoder_dim = 128
decoder_dim = 128
beam_size = 5
max_length = 120
# encoder
src_word_id = pd.data(
name='source_language_word',
type=pd.data.integer_value_sequence(source_dict_dim))
src_embedding = pd.embedding(size=source_dict_size, size=word_vector_dim)
src_word_vec = pd.lookup(src_embedding, src_word_id)
encoder_out_seq = pd.gru(input=src_word_vec, size=encoder_dim)
encoder_ctx = pd.last_seq(encoder_out_seq)
# encoder_ctx_proj is the learned semantic vector
encoder_ctx_proj = pd.fc(
encoder_ctx, size=decoder_dim, act=pd.activation.Tanh(), bias=None)
```
**Decoder**
```python
def generate():
decoder = pd.while_loop()
with decoder.step():
decoder_mem = decoder.memory(init=encoder_ctx) # mark the memory
generated_ids = decoder.memory() # TODO init to batch_size <s>s
generated_scores = decoder.memory() # TODO init to batch_size 1s or 0s
target_word = pd.lookup(trg_embedding, gendrated_ids)
# expand encoder_ctx's batch to fit target_word's lod
# for example
# decoder_mem.lod is
# [[0 1 3],
# [0 1 3 6]]
# its tensor content is [a1 a2 a3 a4 a5]
# which means there are 2 sentences to translate
# - the first sentence has 1 translation prefixes, the offsets are [0, 1)
# - the second sentence has 2 translation prefixes, the offsets are [1, 3) and [3, 6)
# the target_word.lod is
# [[0, 1, 6]
# [0, 2, 4, 7, 9 12]]
# which means 2 sentences to translate, each has 1 and 5 prefixes
# the first prefix has 2 candidates
# the following has 2, 3, 2, 3 candidates
# the encoder_ctx_expanded's content will be
# [a1 a1 a2 a2 a3 a3 a3 a4 a4 a5 a5 a5]
encoder_ctx_expanded = pd.lod_expand(encoder_ctx, target_word)
decoder_input = pd.fc(
act=pd.activation.Linear(),
input=[target_word, encoder_ctx],
size=3 * decoder_dim)
gru_out, cur_mem = pd.gru_step(
decoder_input, mem=decoder_mem, size=decoder_dim)
scores = pd.fc(
gru_out,
size=trg_dic_size,
bias=None,
act=pd.activation.Softmax())
# K is an config
topk_scores, topk_ids = pd.top_k(scores, K)
topk_generated_scores = pd.add_scalar(topk_scores, generated_scores)
selected_ids, selected_generation_scores = decoder.beam_search(
topk_ids, topk_generated_scores)
# update the states
decoder_mem.update(cur_mem) # tells how to update state
generated_ids.update(selected_ids)
generated_scores.update(selected_generation_scores)
decoder.output(selected_ids)
decoder.output(selected_generation_scores)
translation_ids, translation_scores = decoder()
```
The `decoder.beam_search` is a operator that given the candidates and the scores of translations including the candidates,
return the result of the beam search algorithm.
In this way, users can customize anything on the inputs or outputs of beam search, for example, two ways to prune some translation prefixes
1. meke the correspondind elements in `topk_generated_scores` zero or some small values, beam_search will discard this candidate.
2. remove some specific candidate in `selected_ids`
3. get the final `translation_ids`, remove the translation sequence in it.
The implementation of sequence decoder can reuse the C++ class [RNNAlgorithm](https://github.com/Superjom/Paddle/blob/68cac3c0f8451fe62a4cdf156747d6dc0ee000b3/paddle/operators/dynamic_recurrent_op.h#L30),
so the python syntax is quite similar to a [RNN](https://github.com/Superjom/Paddle/blob/68cac3c0f8451fe62a4cdf156747d6dc0ee000b3/doc/design/block.md#blocks-with-for-and-rnnop).
Both of them are two-level `LoDTensors`
- the first level represents `batch_size` of (source) sentences;
- the second level represents the candidate ID sets for translation prefix.
for example, 3 source sentences to translate, and has 2, 3, 1 candidates.
Unlike an RNN, in sequence decoder, the previous state and the current state have different LoD and shape,
a `lod_expand` operator is used to expand the LoD of the previous state to fit the current state.
For example, the previous state
* LoD is `[0, 1, 3][0, 2, 5, 6]`
* content of tensor is `a1 a2 b1 b2 b3 c1`
the current state stored in `encoder_ctx_expanded`
* LoD is `[0, 2, 7][0 3 5 8 9 11 11]`
* the content is
- a1 a1 a1 (a1 has 3 candidates, so the state should be copied 3 times for each candidates)
- a2 a2
- b1 b1 b1
- b2
- b3 b3
- None (c1 has 0 candidates, so c1 is dropped)
Benefit from the relative offset LoD, empty candidate set can be represented naturally.
the status in each time step can be stored in `TensorArray`, and `Pack`ed to a final LoDTensor, the corresponding syntax is
```python
decoder.output(selected_ids)
decoder.output(selected_generation_scores)
```
the `selected_ids` is the candidate ids for the prefixes,
it will be `Packed` by `TensorArray` to a two-level `LoDTensor`,
the first level represents the source sequences,
the second level represents generated sequences.
Pack the `selected_scores` will get a `LoDTensor` that stores scores of each candidate of translations.
Pack the `selected_generation_scores` will get a `LoDTensor`, and each tail is the probability of the translation.
## LoD and shape changes during decoding
<p align="center">
<img src="./images/LOD-and-shape-changes-during-decoding.jpg"/>
</p>
According the image above, the only phrase to change LoD is beam search.
## Beam search design
The beam search algorthm will be implemented as one method of the sequence decoder, it has 3 inputs
1. `topk_ids`, top K candidate ids for each prefix.
2. `topk_scores`, the corresponding scores for `topk_ids`
3. `generated_scores`, the score of the prefixes.
All of the are LoDTensors, so that the sequence affilication is clear.
Beam search will keep a beam for each prefix and select a smaller candidate set for each prefix.
It will return three variables
1. `selected_ids`, the final candidate beam search function selected for the next step.
2. `selected_scores`, the scores for the candidates.
3. `generated_scores`, the updated scores for each prefixes (with the new candidates appended).
## Introducing the LoD-based `Pack` and `Unpack` methods in `TensorArray`
The `selected_ids`, `selected_scores` and `generated_scores` are LoDTensors,
and they exist in each time step,
so it is natural to store them in arrays.
Currently, PaddlePaddle has a module called `TensorArray` which can store an array of tensors,
the results of beam search are better to store in a `TensorArray`.
The `Pack` and `UnPack` in `TensorArray` are used to package tensors in the array to a `LoDTensor` or split the `LoDTensor` to an array of tensors.
It needs some extensions to support pack or unpack an array of `LoDTensors`.
develop/doc/design/ops/sequence_decoder.html 0 → 100644
浏览文件 @ 4e88e677
<!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>
<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/docker_install_en.html">PaddlePaddle in Docker Containers</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/build_from_source_en.html">Installing from Sources</a></li>
</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>
<p>In tasks such as machine translation and image to text,
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.</p>
<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>
<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, a 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, the implementation relays on a lot of special data structures,
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 PaddlePaddle&#8217;s refactoring work,
some new concept is 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 sequence usage,
and they can help to make the implementation of beam search based sequence decoder <strong>more transparent and modular</strong> .</p>
<p>For example, the RNN sates, candidates IDs and probabilities of beam search can be represented as <code class="docutils literal"><span class="pre">LoDTensors</span></code>;
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>
<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 each represents a level.</p>
<p>The integers in each level represents 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 clear.</p>
<p>The relative-offset LoD can fast retrieve any sequence but fails to represent empty sequences, for example, a two-level LoD is as follows</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">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>
<p>There are many scenarios that relay on empty sequence representation,
such as machine translation or image to text, one instance has no translations or the empty candidate set for a prefix.</p>
<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>
<p>The following demos are based on relative-offset LoD.</p>
</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>
<p>Let&#8217;s start from a simple machine translation model that is simplified from <a class="reference external" href="https://github.com/PaddlePaddle/book/tree/develop/08.machine_translation">machine translation chapter</a> to draw a simple 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 decoder to generate new sentences.</p>
<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>
<span class="c1"># the target_word.lod is </span>
<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>
<p>The <code class="docutils literal"><span class="pre">decoder.beam_search</span></code> is a operator that given the candidates and the scores of translations including the candidates,
return the result of the beam search algorithm.</p>
<p>In this way, users can customize anything on the inputs or outputs of beam search, for example, two ways to prune some translation prefixes</p>
<ol class="simple">
<li>meke the correspondind 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>
</ol>
<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 a <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>
<ul class="simple">
<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>
</ul>
<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,
a <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>
<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>
<p>the current state stored in <code class="docutils literal"><span class="pre">encoder_ctx_expanded</span></code></p>
<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>
<p>Benefit from the relative offset LoD, 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>
<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>
<p>the <code class="docutils literal"><span class="pre">selected_ids</span></code> is the candidate ids for the prefixes,
it 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>,
the first level represents the source sequences,
the second level represents generated sequences.</p>
<p>Pack 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 candidate of translations.</p>
<p>Pack 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>
</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"/>
</p><p>According the image above, the only phrase to change LoD is beam search.</p>
</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>
<p>The beam search algorthm will be implemented as one method of the sequence decoder, it has 3 inputs</p>
<ol class="simple">
<li><code class="docutils literal"><span class="pre">topk_ids</span></code>, top K candidate ids for each prefix.</li>
<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>
<p>All of the are LoDTensors, so that the sequence affilication 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>
<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>
<li><code class="docutils literal"><span class="pre">generated_scores</span></code>, the updated scores for each prefixes (with the new candidates appended).</li>
</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>
<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,
and they exist in each time step,
so it is natural to store them in arrays.</p>
<p>Currently, PaddlePaddle has a module called <code class="docutils literal"><span class="pre">TensorArray</span></code> which can store an array of tensors,
the results of beam search are better to store 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 package tensors in the array to a <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 pack or unpack an array of <code class="docutils literal"><span class="pre">LoDTensors</span></code>.</p>
</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>
\ No newline at end of file
develop/doc/searchindex.js
浏览文件 @ 4e88e677
因为 它太大了无法显示 source diff 。你可以改为 查看blob。
develop/doc_cn/_sources/design/ops/sequence_decoder.md.txt 0 → 100644
浏览文件 @ 4e88e677
# Design: Sequence Decoder Generating LoDTensors
In tasks such as machine translation and image to text,
a [sequence decoder](https://github.com/PaddlePaddle/book/blob/develop/08.machine_translation/README.md) is necessary to generate sequences.
This documentation describes how to implement the sequence decoder as an operator.
## Beam Search based Decoder
The [beam search algorithm](https://en.wikipedia.org/wiki/Beam_search) 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.
In the old version of PaddlePaddle, a C++ class `RecurrentGradientMachine` implements the general sequence decoder based on beam search,
due to the complexity, the implementation relays on a lot of special data structures,
quite trivial and hard to be customized by users.
There are a lot of heuristic tricks in the sequence generation tasks,
so the flexibility of sequence decoder is very important to users.
During PaddlePaddle's refactoring work,
some new concept is proposed such as [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md) and [TensorArray](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/tensor_array.md) that can better support sequence usage,
and they can help to make the implementation of beam search based sequence decoder **more transparent and modular** .
For example, the RNN sates, candidates IDs and probabilities of beam search can be represented as `LoDTensors`;
the selected candidate's IDs in each time step can be stored in a `TensorArray`, and `Packed` to the sentences translated.
## Changing LoD's absolute offset to relative offsets
The current `LoDTensor` is designed to store levels of variable-length sequences,
it stores several arrays of integers each represents a level.
The integers in each level represents the begin and end (not inclusive) offset of a sequence **in the underlying tensor**,
let's call this format the **absolute-offset LoD** for clear.
The relative-offset LoD can fast retrieve any sequence but fails to represent empty sequences, for example, a two-level LoD is as follows
```python
[[0, 3, 9]
[0, 2, 3, 3, 3, 9]]
```
The first level tells that there are two sequences:
- the first's offset is `[0, 3)`
- the second's offset is `[3, 9)`
while on the second level, there are several empty sequences that both begin and end at `3`.
It is impossible to tell how many empty second-level sequences exist in the first-level sequences.
There are many scenarios that relay on empty sequence representation,
such as machine translation or image to text, one instance has no translations or the empty candidate set for a prefix.
So let's introduce another format of LoD,
it stores **the offsets of the lower level sequences** and is called **relative-offset** LoD.
For example, to represent the same sequences of the above data
```python
[[0, 3, 6]
[0, 2, 3, 3, 3, 9]]
```
the first level represents that there are two sequences,
their offsets in the second-level LoD is `[0, 3)` and `[3, 5)`.
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.
The following demos are based on relative-offset LoD.
## Usage in a simple machine translation model
Let's start from a simple machine translation model that is simplified from [machine translation chapter](https://github.com/PaddlePaddle/book/tree/develop/08.machine_translation) to draw a simple blueprint of what a sequence decoder can do and how to use it.
The model has an encoder that learns the semantic vector from a sequence,
and a decoder which uses the sequence decoder to generate new sentences.
**Encoder**
```python
import paddle as pd
dict_size = 8000
source_dict_size = dict_size
target_dict_size = dict_size
word_vector_dim = 128
encoder_dim = 128
decoder_dim = 128
beam_size = 5
max_length = 120
# encoder
src_word_id = pd.data(
name='source_language_word',
type=pd.data.integer_value_sequence(source_dict_dim))
src_embedding = pd.embedding(size=source_dict_size, size=word_vector_dim)
src_word_vec = pd.lookup(src_embedding, src_word_id)
encoder_out_seq = pd.gru(input=src_word_vec, size=encoder_dim)
encoder_ctx = pd.last_seq(encoder_out_seq)
# encoder_ctx_proj is the learned semantic vector
encoder_ctx_proj = pd.fc(
encoder_ctx, size=decoder_dim, act=pd.activation.Tanh(), bias=None)
```
**Decoder**
```python
def generate():
decoder = pd.while_loop()
with decoder.step():
decoder_mem = decoder.memory(init=encoder_ctx) # mark the memory
generated_ids = decoder.memory() # TODO init to batch_size <s>s
generated_scores = decoder.memory() # TODO init to batch_size 1s or 0s
target_word = pd.lookup(trg_embedding, gendrated_ids)
# expand encoder_ctx's batch to fit target_word's lod
# for example
# decoder_mem.lod is
# [[0 1 3],
# [0 1 3 6]]
# its tensor content is [a1 a2 a3 a4 a5]
# which means there are 2 sentences to translate
# - the first sentence has 1 translation prefixes, the offsets are [0, 1)
# - the second sentence has 2 translation prefixes, the offsets are [1, 3) and [3, 6)
# the target_word.lod is
# [[0, 1, 6]
# [0, 2, 4, 7, 9 12]]
# which means 2 sentences to translate, each has 1 and 5 prefixes
# the first prefix has 2 candidates
# the following has 2, 3, 2, 3 candidates
# the encoder_ctx_expanded's content will be
# [a1 a1 a2 a2 a3 a3 a3 a4 a4 a5 a5 a5]
encoder_ctx_expanded = pd.lod_expand(encoder_ctx, target_word)
decoder_input = pd.fc(
act=pd.activation.Linear(),
input=[target_word, encoder_ctx],
size=3 * decoder_dim)
gru_out, cur_mem = pd.gru_step(
decoder_input, mem=decoder_mem, size=decoder_dim)
scores = pd.fc(
gru_out,
size=trg_dic_size,
bias=None,
act=pd.activation.Softmax())
# K is an config
topk_scores, topk_ids = pd.top_k(scores, K)
topk_generated_scores = pd.add_scalar(topk_scores, generated_scores)
selected_ids, selected_generation_scores = decoder.beam_search(
topk_ids, topk_generated_scores)
# update the states
decoder_mem.update(cur_mem) # tells how to update state
generated_ids.update(selected_ids)
generated_scores.update(selected_generation_scores)
decoder.output(selected_ids)
decoder.output(selected_generation_scores)
translation_ids, translation_scores = decoder()
```
The `decoder.beam_search` is a operator that given the candidates and the scores of translations including the candidates,
return the result of the beam search algorithm.
In this way, users can customize anything on the inputs or outputs of beam search, for example, two ways to prune some translation prefixes
1. meke the correspondind elements in `topk_generated_scores` zero or some small values, beam_search will discard this candidate.
2. remove some specific candidate in `selected_ids`
3. get the final `translation_ids`, remove the translation sequence in it.
The implementation of sequence decoder can reuse the C++ class [RNNAlgorithm](https://github.com/Superjom/Paddle/blob/68cac3c0f8451fe62a4cdf156747d6dc0ee000b3/paddle/operators/dynamic_recurrent_op.h#L30),
so the python syntax is quite similar to a [RNN](https://github.com/Superjom/Paddle/blob/68cac3c0f8451fe62a4cdf156747d6dc0ee000b3/doc/design/block.md#blocks-with-for-and-rnnop).
Both of them are two-level `LoDTensors`
- the first level represents `batch_size` of (source) sentences;
- the second level represents the candidate ID sets for translation prefix.
for example, 3 source sentences to translate, and has 2, 3, 1 candidates.
Unlike an RNN, in sequence decoder, the previous state and the current state have different LoD and shape,
a `lod_expand` operator is used to expand the LoD of the previous state to fit the current state.
For example, the previous state
* LoD is `[0, 1, 3][0, 2, 5, 6]`
* content of tensor is `a1 a2 b1 b2 b3 c1`
the current state stored in `encoder_ctx_expanded`
* LoD is `[0, 2, 7][0 3 5 8 9 11 11]`
* the content is
- a1 a1 a1 (a1 has 3 candidates, so the state should be copied 3 times for each candidates)
- a2 a2
- b1 b1 b1
- b2
- b3 b3
- None (c1 has 0 candidates, so c1 is dropped)
Benefit from the relative offset LoD, empty candidate set can be represented naturally.
the status in each time step can be stored in `TensorArray`, and `Pack`ed to a final LoDTensor, the corresponding syntax is
```python
decoder.output(selected_ids)
decoder.output(selected_generation_scores)
```
the `selected_ids` is the candidate ids for the prefixes,
it will be `Packed` by `TensorArray` to a two-level `LoDTensor`,
the first level represents the source sequences,
the second level represents generated sequences.
Pack the `selected_scores` will get a `LoDTensor` that stores scores of each candidate of translations.
Pack the `selected_generation_scores` will get a `LoDTensor`, and each tail is the probability of the translation.
## LoD and shape changes during decoding
<p align="center">
<img src="./images/LOD-and-shape-changes-during-decoding.jpg"/>
</p>
According the image above, the only phrase to change LoD is beam search.
## Beam search design
The beam search algorthm will be implemented as one method of the sequence decoder, it has 3 inputs
1. `topk_ids`, top K candidate ids for each prefix.
2. `topk_scores`, the corresponding scores for `topk_ids`
3. `generated_scores`, the score of the prefixes.
All of the are LoDTensors, so that the sequence affilication is clear.
Beam search will keep a beam for each prefix and select a smaller candidate set for each prefix.
It will return three variables
1. `selected_ids`, the final candidate beam search function selected for the next step.
2. `selected_scores`, the scores for the candidates.
3. `generated_scores`, the updated scores for each prefixes (with the new candidates appended).
## Introducing the LoD-based `Pack` and `Unpack` methods in `TensorArray`
The `selected_ids`, `selected_scores` and `generated_scores` are LoDTensors,
and they exist in each time step,
so it is natural to store them in arrays.
Currently, PaddlePaddle has a module called `TensorArray` which can store an array of tensors,
the results of beam search are better to store in a `TensorArray`.
The `Pack` and `UnPack` in `TensorArray` are used to package tensors in the array to a `LoDTensor` or split the `LoDTensor` to an array of tensors.
It needs some extensions to support pack or unpack an array of `LoDTensors`.
develop/doc_cn/design/ops/sequence_decoder.html 0 → 100644
浏览文件 @ 4e88e677
<!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 文档</title>
<link rel="stylesheet" href="../../_static/css/theme.css" type="text/css" />
<link rel="index" title="索引"
href="../../genindex.html"/>
<link rel="search" title="搜索" href="../../search.html"/>
<link rel="top" title="PaddlePaddle 文档" 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_cn.html">新手入门</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../howto/index_cn.html">进阶指南</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../api/index_cn.html">API</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../faq/index_cn.html">FAQ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../mobile/index_cn.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_cn.html">新手入门</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../getstarted/build_and_install/index_cn.html">安装与编译</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/docker_install_cn.html">PaddlePaddle的Docker容器使用方式</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../getstarted/build_and_install/cmake/build_from_source_cn.html">PaddlePaddle的编译选项</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../getstarted/concepts/use_concepts_cn.html">基本使用概念</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../howto/index_cn.html">进阶指南</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/cmd_parameter/index_cn.html">设置命令行参数</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/use_case_cn.html">使用案例</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/arguments_cn.html">参数概述</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/usage/cmd_parameter/detail_introduction_cn.html">细节描述</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/cluster/cluster_train_cn.html">PaddlePaddle分布式训练</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/k8s/k8s_basis_cn.html">Kubernetes 简介</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/k8s/k8s_cn.html">Kubernetes单机训练</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/usage/k8s/k8s_distributed_cn.html">Kubernetes分布式训练</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/build_cn.html">编译PaddlePaddle和运行单元测试</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/dev/write_docs_cn.html">如何贡献/修改文档</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/deep_model/rnn/index_cn.html">RNN相关模型</a><ul>
<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/rnn_config_cn.html">RNN配置</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/recurrent_group_cn.html">Recurrent Group教程</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/hierarchical_layer_cn.html">支持双层序列作为输入的Layer</a></li>
<li class="toctree-l3"><a class="reference internal" href="../../howto/deep_model/rnn/hrnn_rnn_api_compare_cn.html">单双层RNN API对比介绍</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="../../howto/optimization/gpu_profiling_cn.html">GPU性能分析与调优</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../api/index_cn.html">API</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../api/v2/model_configs.html">模型配置</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">数据访问</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">训练与应用</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../faq/index_cn.html">FAQ</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../faq/build_and_install/index_cn.html">编译安装与单元测试</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../faq/model/index_cn.html">模型配置</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../faq/parameter/index_cn.html">参数设置</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../faq/local/index_cn.html">本地训练与预测</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../faq/cluster/index_cn.html">集群训练与预测</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="../../mobile/index_cn.html">MOBILE</a><ul>
<li class="toctree-l2"><a class="reference internal" href="../../mobile/cross_compiling_for_android_cn.html">构建Android平台上的PaddlePaddle库</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../mobile/cross_compiling_for_ios_cn.html">构建iOS平台上的PaddlePaddle库</a></li>
<li class="toctree-l2"><a class="reference internal" href="../../mobile/cross_compiling_for_raspberry_cn.html">构建Raspberry Pi平台上的PaddlePaddle库</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="永久链接至标题"></a></h1>
<p>In tasks such as machine translation and image to text,
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.</p>
<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="永久链接至标题"></a></h2>
<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, a 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, the implementation relays on a lot of special data structures,
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 PaddlePaddle&#8217;s refactoring work,
some new concept is 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 sequence usage,
and they can help to make the implementation of beam search based sequence decoder <strong>more transparent and modular</strong> .</p>
<p>For example, the RNN sates, candidates IDs and probabilities of beam search can be represented as <code class="docutils literal"><span class="pre">LoDTensors</span></code>;
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="永久链接至标题"></a></h2>
<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 each represents a level.</p>
<p>The integers in each level represents 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 clear.</p>
<p>The relative-offset LoD can fast retrieve any sequence but fails to represent empty sequences, for example, a two-level LoD is as follows</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">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>
<p>There are many scenarios that relay on empty sequence representation,
such as machine translation or image to text, one instance has no translations or the empty candidate set for a prefix.</p>
<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>
<p>The following demos are based on relative-offset LoD.</p>
</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="永久链接至标题"></a></h2>
<p>Let&#8217;s start from a simple machine translation model that is simplified from <a class="reference external" href="https://github.com/PaddlePaddle/book/tree/develop/08.machine_translation">machine translation chapter</a> to draw a simple 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 decoder to generate new sentences.</p>
<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>
<span class="c1"># the target_word.lod is </span>
<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>
<p>The <code class="docutils literal"><span class="pre">decoder.beam_search</span></code> is a operator that given the candidates and the scores of translations including the candidates,
return the result of the beam search algorithm.</p>
<p>In this way, users can customize anything on the inputs or outputs of beam search, for example, two ways to prune some translation prefixes</p>
<ol class="simple">
<li>meke the correspondind 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>
</ol>
<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 a <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>
<ul class="simple">
<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>
</ul>
<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,
a <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>
<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>
<p>the current state stored in <code class="docutils literal"><span class="pre">encoder_ctx_expanded</span></code></p>
<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>
<p>Benefit from the relative offset LoD, 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>
<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>
<p>the <code class="docutils literal"><span class="pre">selected_ids</span></code> is the candidate ids for the prefixes,
it 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>,
the first level represents the source sequences,
the second level represents generated sequences.</p>
<p>Pack 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 candidate of translations.</p>
<p>Pack 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>
</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="永久链接至标题"></a></h2>
<p align="center">
<img src="./images/LOD-and-shape-changes-during-decoding.jpg"/>
</p><p>According the image above, the only phrase to change LoD is beam search.</p>
</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="永久链接至标题"></a></h2>
<p>The beam search algorthm will be implemented as one method of the sequence decoder, it has 3 inputs</p>
<ol class="simple">
<li><code class="docutils literal"><span class="pre">topk_ids</span></code>, top K candidate ids for each prefix.</li>
<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>
<p>All of the are LoDTensors, so that the sequence affilication 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>
<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>
<li><code class="docutils literal"><span class="pre">generated_scores</span></code>, the updated scores for each prefixes (with the new candidates appended).</li>
</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="永久链接至标题"></a></h2>
<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,
and they exist in each time step,
so it is natural to store them in arrays.</p>
<p>Currently, PaddlePaddle has a module called <code class="docutils literal"><span class="pre">TensorArray</span></code> which can store an array of tensors,
the results of beam search are better to store 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 package tensors in the array to a <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 pack or unpack an array of <code class="docutils literal"><span class="pre">LoDTensors</span></code>.</p>
</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="../../_static/translations.js"></script>
<script type="text/javascript" src="https://cdn.bootcss.com/mathjax/2.7.0/MathJax.js"></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>
\ No newline at end of file
develop/doc_cn/searchindex.js
浏览文件 @ 4e88e677
因为 它太大了无法显示 source diff 。你可以改为 查看blob。
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或