[seq2seq]论文实现:Effective Approaches to Attention-based Neural Machine Translation(下)

本文涉及的产品
NLP自然语言处理_基础版,每接口每天50万次
NLP 自学习平台,3个模型定制额度 1个月
NLP自然语言处理_高级版,每接口累计50万次
简介: [seq2seq]论文实现:Effective Approaches to Attention-based Neural Machine Translation(下)

[seq2seq]论文实现:Effective Approaches to Attention-based Neural Machine Translation(上)https://developer.aliyun.com/article/1504074?spm=a2c6h.13148508.setting.41.36834f0eMJOehx

二、论文解读

2.1 RNN模型

介绍seq2seq之前我们需要介绍一下RNN模型,RNN模型表示循环神经网络,具有代表性的有SimpleRNN,GRU,LSTM;其基本实现原理如图:

用公式表达如下:

image.png

其中 W , U , V 三者权重是共享的,所有RNN的参数数量是与 Xt的最后一个维度有关的,维度变化公式如下:

O[outputdim∗1]=V[outputdim∗units]·S[units∗1]+B[outputdim∗1]S[units∗1]=U[units∗xdim]·X[xdim∗1]+W[units∗units]·S[units∗1]+B[units∗1])

\begin{aligned} & O_{[output_{dim}*1]} =V_{[output_{dim}*units]}·S_{[units*1]} + B_{[output_{dim}*1]} \\ & S_{[units*1]} = U_{[units*x_{dim}]}·X_{[x_{dim}*1]}+W_{[units*units]}·S_{[units*1]}+B_{[units*1]}) \end{aligned}

所以,RNN需要的参数数量为(units+x_dim+1)*units + (units+1)*output_dim

2.2 Attention-based Models

论文中提出了两种Attention-based Model,分别是全局注意力模型和局部注意力模型;其结构图如下:

从图中可以看到,其全局和局部的区别在于 a t a_t atc t c_t ct 的不同,在分析之前,我们先定义一些变量: t t t 是时间步, a t a_t at 是模型对其权重向量,其主要是由于 h t h_t hth ‾ s \overline h_s hs计算得到, h s ‾ \overline{h_s} hs 是decoder中第s位置的state, c t c_t ct 被称作为内容向量,由 a t a_t ath s ‾ \overline{h_s} hs计算得到;

接下来我们依次对全局注意力和局部注意力进行分析;

Global attentional model

如图,ct 是由 at hs 计算得到,这里首先定义  at(s)的计算公式为:

at(s)=align(ht,hs¯)=exp(score(ht,hs¯))∑s′exp(score(ht,hs′¯))

论文中这里定义 image.png 有三种方式:

KaTeX parse error: Undefined control sequence: \cases at position 32: …rline{h_s}) = \̲c̲a̲s̲e̲s̲{ h_t^T\overlin…

这里用querykey,value 来解释就相当于 ht query , image.png 做  keyvalue;其流程为 image.png

Local attentional model

全局注意力机制有一个缺点,即它必须关注每个目标词的源端的所有单词,这是昂贵的,并可能使翻译更长的序列不切实际,例如段落或文档。这里使用局部注意力机制进行优化;

所谓局部注意力机制就是说我们不去计算所有位置,而是计算部分位置,那么这部分位置该怎么选择呢,在语言翻译模型中,某部分的target是由某部分的source构成的,在已知target的位置  t 时找到source的位置pt 论文中有两种方式取实现:

  • Monotonic alignmentpt=t
  • Predictive alignment image.png

这里的  vpWp都是参数;

在找到  pt 之后,我们对  [ptD,pt+D]这些位置上的  hs 进行注意力机制计算 at,ct

同时由于词距离pt 越远,则其影响越弱,这里论文中使用高斯分布的方式对 at取值:取值方式如下:

  image.png

根据经验我们一般把 a σ 设置为 image.png ,这就是局部注意力机制;

2.3 Input-feeding Approach

在全局和局部注意力模型中,其注意力部分都是独立进行的,并没有对下一个时间步的过程产生影响,这并不合理,在标准的MT中,通常在翻译过程中会维护一个覆盖集,以跟踪哪些源词已经被翻译过。同样地,在注意nmt中,对齐决策应该共同考虑到过去的对齐信息。我们可以优化一下,把每次的输出作为下一个时间步的输入;如图所示:

2.4 模型效果

论文中模型效果如图所示:

三、过程实现

3.1 导包

这里要用到的包有:tensorflow, keras_nlp, matplotlib, numpy

import tensorflow as tf
import keras_nlp
import matplotlib.pyplot as plt
import numpy as np
plt.rcParams['font.sans-serif']=['SimHei'] 
plt.rcParams['axes.unicode_minus']=False

3.2 数据准备

这里使用的是中英文翻译数据集,进行清洗和dataset构造

# 数据处理
def process_data(x):
    res = tf.strings.split(x, '\t')
    return res[1], res[3]
# 导入数据
dataset = tf.data.TextLineDataset('./data/transformer_data.tsv')
dataset = dataset.map(process_data)
# 建立中英文wordpiece词表
vocab_chinese = keras_nlp.tokenizers.compute_word_piece_vocabulary(
    dataset.map(lambda x, y: x),
    vocabulary_size=20_000,
    lowercase=True,
    strip_accents=True,
    split_on_cjk=True,
    reserved_tokens=["[PAD]", "[START]", "[END]", "[MASK]", "[UNK]"],
)
vocab_english = keras_nlp.tokenizers.compute_word_piece_vocabulary(
    dataset.map(lambda x, y: y),
    vocabulary_size=20_000,
    lowercase=True,
    strip_accents=True,
    split_on_cjk=True,
    reserved_tokens=["[PAD]", "[START]", "[END]", "[MASK]", "[UNK]"],
)
# 构建分词器
chinese_tokenizer = keras_nlp.tokenizers.WordPieceTokenizer(vocabulary=vocab_chinese, oov_token="[UNK]")
english_tokenizer = keras_nlp.tokenizers.WordPieceTokenizer(vocabulary=vocab_english, oov_token="[UNK]")
# 再进行一次数据处理
def process_data_(ch, en, maxtoken=128):
    
    ch = chinese_tokenizer(ch)[:,:maxtoken]
    en = english_tokenizer(tf.strings.lower(en))[:,:maxtoken]
    
    ch = tf.concat([tf.ones(shape=(64,1), dtype='int32'), ch, tf.ones(shape=(64,1), dtype='int32')*2], axis=-1).to_tensor()
    en = tf.concat([tf.ones(shape=(64,1), dtype='int32'), en, tf.ones(shape=(64,1), dtype='int32')*2], axis=-1)
    en_inputs = en[:, :-1].to_tensor()  # Drop the [END] tokens
    en_labels = en[:, 1:].to_tensor() # Drop the [START] tokens
    return (ch, en_inputs), en_labels
    
dataset = dataset.batch(64).map(process_data_)
train_dataset = dataset.take(1000)
val_dataset = dataset.skip(500).take(300)
# 数据准备完毕 查看数据
for (pt, en), en_labels in dataset.take(1):
    break
print(pt.shape)
print(en.shape)
print(en_labels.shape)

3.3 构建相关类

encoder:

class Encoder(tf.keras.layers.Layer):
    def __init__(self, vocabulary_size, d_model, units):
        super().__init__()
        self.embedding = tf.keras.layers.Embedding(vocabulary_size, d_model)
        self.rnn = tf.keras.layers.Bidirectional(
            layer=tf.keras.layers.LSTM(units=units, return_sequences=True, return_state=False),
            merge_mode='sum'
        )
    def call(self, inputs):
        x = inputs
        x = self.embedding(x)
        x = self.rnn(x)
        return x

crossattention:

class CrossAttention(tf.keras.layers.Layer):
    def __init__(self, units, **kwargs):
        super().__init__()
        self.mha = tf.keras.layers.MultiHeadAttention(key_dim=units, num_heads=1, **kwargs)
        self.add = tf.keras.layers.Add()
        self.norm = tf.keras.layers.LayerNormalization()
    def call(self, inputs):
        x, context = inputs
        attention_out, attention_score = self.mha(query=x, value=context, key=context, return_attention_scores=True)
        self.last_attention_score = attention_score
        x = self.add([x, attention_out])
        x = self.norm(x)
        return x

decoder:

class Decoder(tf.keras.layers.Layer):
    def __init__(self, vocabulary_size, d_model, units, **kwargs):
        super().__init__()
        self.embedding = tf.keras.layers.Embedding(vocabulary_size, d_model)
        self.rnn = tf.keras.layers.LSTM(units, return_sequences=True)
        self.attention = CrossAttention(units, **kwargs)
        self.dense = tf.keras.layers.Dense(vocabulary_size, activation='softmax')
    def call(self, inputs):
        x, context = inputs
        x = self.embedding(x)
        x = self.rnn(x)
        x = self.attention((x, context))
        x = self.dense(x)
        return x

seq2seq:

class Seq2Seq(tf.keras.models.Model):
    def __init__(self, vocabulary_size_1, vocabulary_size_2, d_model, units, **kwargs):
        super().__init__()
        self.encoder = Encoder(vocabulary_size=vocabulary_size_1, d_model=d_model, units=units)
        self.decoder = Decoder(vocabulary_size=vocabulary_size_2, d_model=d_model, units=units)
    def call(self, inputs):
        pt, en = inputs
        context = self.encoder(pt)
        output = self.decoder((en, context))
        return output

3.4 模型配置

构建模型如下:

seq2seq = Seq2Seq(chinese_tokenizer.vocabulary_size(), english_tokenizer.vocabulary_size(), 512, 30)
# build model
seq2seq((pt, en))
seq2seq.summary()

模型配置:

def masked_loss(y_true, y_pred):
  loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(reduction='none')
  loss = loss_fn(y_true, y_pred)
  
  mask = tf.cast(y_true != 0, loss.dtype)
  loss *= mask
  
  return tf.reduce_sum(loss)/tf.reduce_sum(mask)
def masked_acc(y_true, y_pred):
  y_pred = tf.argmax(y_pred, axis=-1)
  y_pred = tf.cast(y_pred, y_true.dtype)
  
  match = tf.cast(y_true == y_pred, tf.float32)
  mask = tf.cast(y_true != 0, tf.float32)
  
  return tf.reduce_sum(match)/tf.reduce_sum(mask)
seq2seq.compile(
    optimizer='adam',
    loss=masked_loss, 
    metrics=[masked_acc, masked_loss]
)
seq2seq.fit(train_dataset, epochs=10, validation_data=val_dataset)

模型训练结果如下:

作图:

plt.plot(seq2seq.history.history['masked_loss'], label='loss')
plt.plot(seq2seq.history.history['val_masked_loss'], label='val_loss')
plt.plot(seq2seq.history.history['masked_acc'], label='accuracy')
plt.plot(seq2seq.history.history['val_masked_acc'], label='val_accuracy')

3.5 模型推理

构建推理类:

class Inference(tf.Module):
    def __init__(self, model, tokenizer_1, tokenizer_2):
        self.model = model
        self.tokenizer_1 = tokenizer_1
        self.tokenizer_2 = tokenizer_2
    def __call__(self, sentence, MAX_TOKEN=128):
        assert isinstance(sentence, tf.Tensor)
        if len(sentence.shape) == 0:
            sentence = sentence[tf.newaxis]
        sentence = self.tokenizer_1(sentence)
        sentence = tf.concat([tf.ones(shape=[sentence.shape[0], 1], dtype='int32'), sentence, tf.ones(shape=[sentence.shape[0], 1], dtype='int32')*2], axis=-1).to_tensor()
        encoder_input = sentence
        
        start = tf.constant(1, dtype='int64')[tf.newaxis]
        end = tf.constant(2, dtype='int64')[tf.newaxis]
        # tf.TensorArray 类似于python中的列表
        output_array = tf.TensorArray(dtype=tf.int64, size=0, dynamic_size=True)
        # 在index=0的位置写入start
        output_array = output_array.write(0, start)
        
        for i in tf.range(MAX_TOKEN):
            output = tf.transpose(output_array.stack())
            predictions = self.model.predict((encoder_input, output), verbose=0) # Shape `(batch_size, seq_len, vocab_size)`
            
            # 从seq_len中的最后一个维度选择last token
            predictions = predictions[:, -1:, :]  # Shape `(batch_size, 1, vocab_size)`.
            predicted_id = tf.argmax(predictions, axis=-1)
            
            # `predicted_id`加入到output_array中作为一个新的输入
            output_array = output_array.write(i+1, predicted_id[0])
            # 如果输出end就表明停止
            if predicted_id == end:
                break
        output = tf.squeeze(output_array.stack())
        output = self.tokenizer_2.detokenize(output)
        
        return output

开始推理:

inference = Inference(seq2seq, chinese_tokenizer, english_tokenizer)
sentence = '你好呀'
sentence = tf.constant(sentence)
inference(sentence)
# 输出
# <tf.Tensor: shape=(), dtype=string, numpy=b"[START] hello ! [END]">

四、整体总结

效果还不错!训练一定时长后能够正确的翻译,好像相较于Transformer逊色了一点,但是毕竟这个模型结构比Transformer早两年;


目录
相关文章
|
7月前
|
TensorFlow 算法框架/工具
[seq2seq]论文实现:Effective Approaches to Attention-based Neural Machine Translation(上)
[seq2seq]论文实现:Effective Approaches to Attention-based Neural Machine Translation(上)
51 1
|
7月前
|
机器学习/深度学习 自然语言处理 算法
[BPE]论文实现:Neural Machine Translation of Rare Words with Subword Units
[BPE]论文实现:Neural Machine Translation of Rare Words with Subword Units
47 0
|
机器学习/深度学习 自然语言处理 算法
【论文精读】COLING 2022 -Event Detection with Dual Relational Graph Attention Networks
图神经网络(Scarselli et al, 2009)已被广泛用于编码事件检测的依赖树,因为它们可以基于信息聚合方案有效地捕获相关信息(Cao et al, 2021)。
187 0
|
机器学习/深度学习 存储 自然语言处理
PESE Event Structure Extraction using Pointer Network based Encoder-Decoder Architecture论文解读
事件抽取(EE)的任务旨在从文本中找到事件和事件相关的论元信息,并以结构化格式表示它们。大多数以前的工作都试图通过分别识别多个子结构并将它们聚合以获得完整的事件结构来解决这个问题。
83 0
|
机器学习/深度学习 算法 数据挖掘
A Generative Adversarial Network-based Deep Learning Method for Low-quality Defect ImageReconstructi
本文提出了一种基于生成对抗网络 (GAN) 的 DL 方法,用于低质量缺陷图像识别。 GAN用于重建低质量缺陷图像,并建立VGG16网络识别重建图像。
153 0
|
机器学习/深度学习 存储 自然语言处理
Bi-SimCut: A Simple Strategy for Boosting Neural Machine Translation 论文笔记
Bi-SimCut: A Simple Strategy for Boosting Neural Machine Translation 论文笔记
|
机器学习/深度学习 自然语言处理 数据挖掘
【文本分类】Attention-Based Bidirectional Long Short-Term Memory Networks for Relation Classification
【文本分类】Attention-Based Bidirectional Long Short-Term Memory Networks for Relation Classification
163 0
【文本分类】Attention-Based Bidirectional Long Short-Term Memory Networks for Relation Classification
《Autoencoder-based Semi-Supervised Curriculum Learning For Out-of-domain Speaker Verification》电子版地址
Autoencoder-based Semi-Supervised Curriculum Learning For Out-of-domain Speaker Verification
80 0
《Autoencoder-based Semi-Supervised Curriculum Learning For Out-of-domain Speaker   Verification》电子版地址
《Multi-Task Multi-Network Joint-Learning of Deep Residual Networks and Cycle-Consistency Generative Adversarial Networks for Robust Speech Recognition》电子版地址
Multi-Task Multi-Network Joint-Learning of Deep Residual Networks and Cycle-Consistency Generative Adversarial Networks for Robust Speech Recognition
109 0
《Multi-Task Multi-Network Joint-Learning of Deep Residual Networks and Cycle-Consistency Generative Adversarial Networks for Robust Speech Recognition》电子版地址
《Investigation of Transformer based Spelling Correction Model for CTC-based End-to-End Mandarin Speech Recognition》电子版地址
Investigation of Transformer based Spelling Correction Model for CTC-based End-to-End Mandarin Speech Recognition
97 0
《Investigation of Transformer based Spelling Correction Model for CTC-based End-to-End Mandarin Speech Recognition》电子版地址