动手学pytorch-机器翻译

动手学pytorch-机器翻译

1. 机器翻译与数据集
2. Encoder Decoder
3. Sequence to Sequence
4. 实验

1. 机器翻译与数据集

机器翻译（MT）：将一段文本从一种语言自动翻译为另一种语言，用神经网络解决这个问题通常称为神经机器翻译（NMT）。
主要特征：输出是单词序列而不是单个单词。 输出序列的长度可能与源序列的长度不同。
数据集采用 http://www.manythings.org/anki/ 的fra-eng数据集

1.1数据集预处理

#数据字典 char to index and index to char
class Vocab(object):
    def __init__(self, tokens, min_freq=0, use_special_tokens=False):
        counter = collections.Counter(tokens)
        self.token_freqs = list(counter.items())
        self.idx_to_token = []
        if use_special_tokens:
            # padding, begin of sentence, end of sentence, unknown
            self.pad, self.bos, self.eos, self.unk = (0, 1, 2, 3)
            self.idx_to_token += ['', '', '', '']
        else:
            self.unk = 0
            self.idx_to_token += ['']
        self.idx_to_token += [token for token, freq in self.token_freqs
                        if freq >= min_freq and token not in self.idx_to_token]
        self.token_to_idx = dict()
        for idx, token in enumerate(self.idx_to_token):
            self.token_to_idx[token] = idx

    def __len__(self):
        return len(self.idx_to_token)

    def __getitem__(self, tokens):
        if not isinstance(tokens, (list, tuple)):
            return self.token_to_idx.get(tokens, self.unk)
        return [self.__getitem__(token) for token in tokens]

    def to_tokens(self, indices):
        if not isinstance(indices, (list, tuple)):
            return self.idx_to_token[indices]
        return [self.idx_to_token[index] for index in indices]

#数据清洗， tokenize, 建立数据字典
class TextPreprocessor():
    def __init__(self, text, num_lines):
        self.num_lines = num_lines
        text = self.clean_raw_text(text)
        self.src_tokens, self.tar_tokens = self.tokenize(text)
        self.src_vocab = self.build_vocab(self.src_tokens)
        self.tar_vocab = self.build_vocab(self.tar_tokens)
    
    def clean_raw_text(self, text):
        text = text.replace('u202f', ' ').replace('xa0', ' ')
        out = ''
        for i, char in enumerate(text.lower()):
            if char in (',', '!', '.') and i > 0 and text[i-1] != ' ':
                out += ' '
            out += char
        return out
        
    def tokenize(self, text):
        sources, targets = [], []
        for i, line in enumerate(text.split('
')):
            if i > self.num_lines:
                break
            parts = line.split('	')
            if len(parts) >= 2:
                sources.append(parts[0].split(' '))
                targets.append(parts[1].split(' '))
        return sources, targets
        
    def build_vocab(self, tokens):
        tokens = [token for line in tokens for token in line]
        return Vocab(tokens, min_freq=3, use_special_tokens=True)

1.2 创建dataloader

# pad, 构建数据dataset, 创建dataloader
class TextUtil():
    def __init__(self, tp, max_len):
        self.src_vocab, self.tar_vocab = tp.src_vocab, tp.tar_vocab
        src_arr, src_valid_len = self.build_array(tp.src_tokens, tp.src_vocab, max_len = max_len, padding_token = tp.src_vocab.pad, is_source=True)
        tar_arr, tar_valid_len = self.build_array(tp.tar_tokens, tp.tar_vocab, max_len = max_len, padding_token = tp.tar_vocab.pad, is_source=False)
        self.dataset = torch.utils.data.TensorDataset(src_arr, src_valid_len, tar_arr, tar_valid_len)
        
    def build_array(self,lines, vocab, max_len, padding_token, is_source):
        def _pad(line):
            if len(line) > max_len:
                return line[:max_len]
            else:
                return line + (max_len - len(line)) * [padding_token]
        lines = [vocab[line] for line in lines]
        if not is_source:
            lines = [[vocab.bos] + line + [vocab.eos] for line in lines]
        arr = torch.tensor([_pad(line) for line in lines])
        valid_len = (arr != vocab.pad).sum(1)
        return arr, valid_len
        
    def load_data_nmt(self, batch_size):
        train_loader = torch.utils.data.DataLoader(self.dataset, batch_size, shuffle = True)
        return self.src_vocab, self.tar_vocab, train_loader

2. Encoder Decoder

encoder：输入到隐藏状态
decoder：隐藏状态到输出

3. Sequence to Sequence

3.1 结构

训练

预测

具体结构：

3.2 代码实现

class Encoder(nn.Module):
    def __init__(self,**kwargs):
        super(Encoder, self).__init__(**kwargs)
    
    def forward(self, X, *args):
        raise NotImplementedError
    
class Decoder(nn.Module):
    def __init__(self, **kwargs):
        super(Decoder, self).__init__(**kwargs)
    
    def init_state(self, encoded_state, *args):
        raise NotImplementedError
        
    def forward(self, X, state):
        raise NotImplementedError

class EncoderDecoder(nn.Module):
    def __init__(self, encoder, decoder, **kwargs):
        super(EncoderDecoder, self).__init__(**kwargs)
        self.encoder = encoder
        self.decoder = decoder
        
    def forward(self, enc_X, dec_X, *args):
        encoded_state = self.encoder(enc_X, *args)[1]
        decoded_state = self.decoder.init_state(encoded_state, *args)
        return self.decoder(dec_X, decoded_state)

class Seq2SeqEncoder(Encoder):
    def __init__(self, vocab_size, embed_size, num_hiddens, num_layers, dropout=0, **kwargs):
        super(Seq2SeqEncoder, self).__init__(**kwargs)
        self.num_hiddens = num_hiddens
        self.num_layers = num_layers
        self.embedding = nn.Embedding(vocab_size, embed_size)
        self.rnn = nn.LSTM(embed_size, num_hiddens, num_layers, dropout=dropout)
    
    def begin_state(self, batch_size, device):
        H = torch.zeros(size=(self.num_layers, batch_size, self.num_hiddens),  device=device)
        C = torch.zeros(size=(self.num_layers, batch_size, self.num_hiddens),  device=device)
        return (H, C)
    
    def forward(self, X, *args):
        X = self.embedding(X)
        X = X.transpose(0, 1)
        out, state = self.rnn(X)
        return out, state
        
class Seq2SeqDecoder(Decoder):
    def __init__(self, vocab_size, embed_size, num_hiddens, num_layers, dropout=0, **kwargs):
        super(Seq2SeqDecoder, self).__init__(**kwargs)
        self.embedding = nn.Embedding(vocab_size, embed_size)
        self.rnn = nn.LSTM(embed_size, num_hiddens, num_layers, dropout=dropout)
        self.dense = nn.Linear(num_hiddens, vocab_size)
        
    def init_state(self, encoded_state, *args):
        return encoded_state
    
    def forward(self, X, state):
        X = self.embedding(X).transpose(0, 1)
        out, state = self.rnn(X, state)
        out = self.dense(out).transpose(0, 1)
        return out, state

def grad_clipping(params, theta, device):
    """Clip the gradient."""
    norm = torch.tensor([0], dtype=torch.float32, device=device)
    for param in params:
        norm += (param.grad ** 2).sum()
    norm = norm.sqrt().item()
    if norm > theta:
        for param in params:
            param.grad.data.mul_(theta / norm)

def grad_clipping_nn(model, theta, device):
    """Clip the gradient for a nn model."""
    grad_clipping(model.parameters(), theta, device)
    

class MaskedSoftmaxCELoss(nn.CrossEntropyLoss):
    def get_mask(self, X, valid_len, value=0):
        max_len = X.size(1)
        mask = torch.arange(max_len)[None, :].to(valid_len.device) < valid_len[:, None]
        X[~mask] = value
        return X
    
    def forward(self, pred, label, valid_len):
        weights = torch.ones_like(label)
        weights = self.get_mask(weights, valid_len)
        self.reduction = 'none'
        output = super(MaskedSoftmaxCELoss, self).forward(pred.transpose(1,2), label)
        return (output * weights).mean(dim=1)
        

4. 实验

#训练函数
def train(model, data_iter, lr, num_epochs, device):  # Saved in d2l
    model.to(device)
    optimizer = optim.Adam(model.parameters(), lr=lr)
    loss = MaskedSoftmaxCELoss()
    tic = time.time()
    for epoch in range(1, num_epochs+1):
        l_sum, num_tokens_sum = 0.0, 0.0
        for batch in data_iter:
            optimizer.zero_grad()
            X, X_vlen, Y, Y_vlen = [x.to(device) for x in batch]
            Y_input, Y_label, Y_vlen = Y[:,:-1], Y[:,1:], Y_vlen-1
            
            Y_hat, _ = model(X, Y_input, X_vlen, Y_vlen)
            l = loss(Y_hat, Y_label, Y_vlen).sum()
            l.backward()

            with torch.no_grad():
                grad_clipping_nn(model, 5, device)
            num_tokens = Y_vlen.sum().item()
            optimizer.step()
            l_sum += l.sum().item()
            num_tokens_sum += num_tokens
        if epoch % 10 == 0:
            print("epoch {0:4d},loss {1:.3f}, time {2:.1f} sec".format( 
                  epoch, (l_sum/num_tokens_sum), time.time()-tic))
            tic = time.time()

#测试函数
def translate(model, src_sentence, src_vocab, tgt_vocab, max_len, device):
    src_tokens = src_vocab[src_sentence.lower().split(' ')]
    src_len = len(src_tokens)
    if src_len < max_len:
        src_tokens += [src_vocab.pad] * (max_len - src_len)
    enc_X = torch.tensor(src_tokens, device=device)
    enc_valid_length = torch.tensor([src_len], device=device)
    # use expand_dim to add the batch_size dimension.
    encoded_state = model.encoder(enc_X.unsqueeze(dim=0), enc_valid_length)[1]
    dec_state = model.decoder.init_state(encoded_state, enc_valid_length)
    dec_X = torch.tensor([tgt_vocab.bos], device=device).unsqueeze(dim=0)
    predict_tokens = []
    for _ in range(max_len):
        Y, dec_state = model.decoder(dec_X, dec_state)
        # The token with highest score is used as the next time step input.
        dec_X = Y.argmax(dim=2)
        py = dec_X.squeeze(dim=0).int().item()
        if py == tgt_vocab.eos:
            break
        predict_tokens.append(py)
    return ' '.join(tgt_vocab.to_tokens(predict_tokens))

embed_size, num_hiddens, num_layers, dropout = 256, 256, 2, 0.3
batch_size, num_examples, max_len = 256, 5e4, 10
lr, num_epochs = 0.005, 300
tp = TextPreprocessor(raw_text, num_lines=num_examples)
tu = TextUtil(tp, max_len = max_len)
src_vocab, tar_vocab, train_loader = tu.load_data_nmt(batch_size = batch_size)
encoder = Seq2SeqEncoder(len(src_vocab), embed_size, num_hiddens, num_layers, dropout)
decoder = Seq2SeqDecoder(len(tar_vocab), embed_size, num_hiddens, num_layers, dropout)
model = EncoderDecoder(encoder, decoder)
train_ch7(model, train_loader, lr, num_epochs, device=device)

for sentence in ['Go .', 'Wow !', "I'm OK .", 'I won !']:
    print(sentence + ' => ' + translate_ch7(
        model, sentence, src_vocab, tgt_vocab, max_len, ctx))

Go . => va !
Wow ! => <unk> !
I'm OK . => ça va .
I won ! => j'ai gagné !