transformer 源码

训练时：
1. 输入正确标签一次性解码出来

预测时：
1. 第一次输入1个词，解码出一个词
第二次输入第一次输入的词和第一次解码出来词一起，解码出来第3个词，这样依次解码，解码到最长的长度或者<pad>。就结束。
训练时，全部输入与预测时一个一个输入是一样的

1. 需要传入词向量

    def __init__(self, hp):
        self.hp = hp
        self.token2idx, self.idx2token = load_vocab(hp.vocab) # 这里在实际的需求情况下传入自己的词典
        self.embeddings = get_token_embeddings(self.hp.vocab_size, self.hp.d_model, zero_pad=True) # 这里作者使用定义的变量训练的词向量，在实际的生产过程当中，我们可以使用word2vec、bert等

2.position_encoding

def positional_encoding(inputs,
                        num_units,
                        zero_pad=True,
                        scale=True,
                        scope="positional_encoding",
                        reuse=None):
    '''Sinusoidal Positional_Encoding.
 
    Args:
      inputs: A 2d Tensor with shape of (N, T).
      num_units: Output dimensionality
      zero_pad: Boolean. If True, all the values of the first row (id = 0) should be constant zero
      scale: Boolean. If True, the output will be multiplied by sqrt num_units(check details from paper)
      scope: Optional scope for `variable_scope`.
      reuse: Boolean, whether to reuse the weights of a previous layer
        by the same name.
 
    Returns:
        A 'Tensor' with one more rank than inputs's, with the dimensionality should be 'num_units'
    '''
 
    N, T = inputs.get_shape().as_list()
    with tf.variable_scope(scope, reuse=reuse):
        position_ind = tf.tile(tf.expand_dims(tf.range(T), 0), [N, 1])
 
        # First part of the PE function: sin and cos argument
        position_enc = np.array([
            [pos / np.power(10000, 2.*i/num_units) for i in range(num_units)]
            for pos in range(T)])
 
        # Second part, apply the cosine to even columns and sin to odds.
        position_enc[:, 0::2] = np.sin(position_enc[:, 0::2])  # dim 2i
        position_enc[:, 1::2] = np.cos(position_enc[:, 1::2])  # dim 2i+1
 
        # Convert to a tensor
        lookup_table = tf.convert_to_tensor(position_enc)
 
        if zero_pad:
            lookup_table = tf.concat((tf.zeros(shape=[1, num_units]),
                                      lookup_table[1:, :]), 0)
        outputs = tf.nn.embedding_lookup(lookup_table, position_ind)
 
        if scale:
            outputs = outputs * num_units**0.5
 
        return outputs

3. multihead_attention

def multihead_attention(queries,
                        keys,
                        num_units=None,
                        num_heads=8,
                        dropout_rate=0,
                        is_training=True,
                        causality=False,
                        scope="multihead_attention",
                        reuse=None):
    '''Applies multihead attention.
 
    Args:
      queries: A 3d tensor with shape of [N, T_q, C_q].
      keys: A 3d tensor with shape of [N, T_k, C_k].
      num_units: A scalar. Attention size.
      dropout_rate: A floating point number.
      is_training: Boolean. Controller of mechanism for dropout.
      causality: Boolean. If true, units that reference the future are masked.
      num_heads: An int. Number of heads.
      scope: Optional scope for `variable_scope`.
      reuse: Boolean, whether to reuse the weights of a previous layer
        by the same name.
 
    Returns
      A 3d tensor with shape of (N, T_q, C)
    '''
    with tf.variable_scope(scope, reuse=reuse):
        # Set the fall back option for num_units
        if num_units is None:
            num_units = queries.get_shape().as_list()[-1]
 
        # Linear projections
        Q = tf.layers.dense(queries, num_units, activation=tf.nn.relu) # (N, T_q, C)  C为num_units，本实现中未设定，故等于C_q
        K = tf.layers.dense(keys, num_units, activation=tf.nn.relu) # (N, T_k, C)
        V = tf.layers.dense(keys, num_units, activation=tf.nn.relu) # (N, T_k, C)
 
        # Split and concat
        Q_ = tf.concat(tf.split(Q, num_heads, axis=2), axis=0) # (h*N, T_q, C/h)
        K_ = tf.concat(tf.split(K, num_heads, axis=2), axis=0) # (h*N, T_k, C/h)
        V_ = tf.concat(tf.split(V, num_heads, axis=2), axis=0) # (h*N, T_k, C/h) 
 
        # Multiplication
        outputs = tf.matmul(Q_, tf.transpose(K_, [0, 2, 1])) # (h*N, T_q, T_k)
 
        # Scale
        outputs = outputs / (K_.get_shape().as_list()[-1] ** 0.5)
 
        # Key Masking
        key_masks = tf.sign(tf.reduce_sum(tf.abs(keys), axis=-1)) # (N, T_k)
        key_masks = tf.tile(key_masks, [num_heads, 1]) # (h*N, -T_k)
        key_masks = tf.tile(tf.expand_dims(key_masks, 1), [1, tf.shape(queries)[1], 1]) # (h*N, T_q, T_k)
 
        paddings = tf.ones_like(outputs)*(-2**32+1)
        b = tf.equal(key_masks, 0)
        """
            然后定义一个和outputs同shape的paddings，该tensor每个值都设定的极小。用where函数比较，当对应位置的key_masks值为0也就是需要mask时，
            outputs的该值（attention score）设置为极小的值（利用paddings实现），否则保留原来的outputs值。
            经过以上key mask操作之后outputs的shape仍为 (h*N, T_q, T_k)，只是对应mask了的key的score变为很小的值。
        """
        outputs = tf.where(tf.equal(key_masks, 0), paddings, outputs) # (h*N, T_q, T_k)
 
        # Causality = Future blinding
        if causality: # 是否忽略未来信息
            diag_vals = tf.ones_like(outputs[0, :, :]) # (T_q, T_k)
            tril = tf.linalg.LinearOperatorLowerTriangular(diag_vals).to_dense() # (T_q, T_k)
            masks = tf.tile(tf.expand_dims(tril, 0), [tf.shape(outputs)[0], 1, 1]) # (h*N, T_q, T_k)
 
            paddings = tf.ones_like(masks)*(-2**32+1)
            outputs = tf.where(tf.equal(masks, 0), paddings, outputs) # (h*N, T_q, T_k)
 
        # Activation
        outputs = tf.nn.softmax(outputs) # (h*N, T_q, T_k)
 
        # Query Masking
        query_masks = tf.sign(tf.reduce_sum(tf.abs(queries), axis=-1)) # (N, T_q)
        query_masks = tf.tile(query_masks, [num_heads, 1]) # (h*N, T_q)
        query_masks = tf.tile(tf.expand_dims(query_masks, -1), [1, 1, tf.shape(keys)[1]]) # (h*N, T_q, T_k)
        outputs *= query_masks # broadcasting. (N, T_q, T_k)？注释有误，将C改成T_k
 
        # Dropouts
        outputs = tf.layers.dropout(outputs, rate=dropout_rate, training=tf.convert_to_tensor(is_training))
 
        # Weighted sum
        outputs = tf.matmul(outputs, V_) # ( h*N, T_q, C/h)
 
        # Restore shape
        outputs = tf.concat(tf.split(outputs, num_heads, axis=0), axis=2 ) # (N, T_q, C)
 
        # Residual connection
        outputs += queries
 
        # Normalize
        outputs = normalize(outputs) # (N, T_q, C)
 
    return outputs

4. feedforward

def feedforward(inputs,
                num_units=[2048, 512],
                scope="multihead_attention",
                reuse=None):
    '''Point-wise feed forward net.
 
    Args:
      inputs: A 3d tensor with shape of [N, T, C].
      num_units: A list of two integers.
      scope: Optional scope for `variable_scope`.
      reuse: Boolean, whether to reuse the weights of a previous layer
        by the same name.
 
    Returns:
      A 3d tensor with the same shape and dtype as inputs
    '''
    with tf.variable_scope(scope, reuse=reuse):
        # Inner layer
        params = {"inputs": inputs, "filters": num_units[0], "kernel_size": 1,
                  "activation": tf.nn.relu, "use_bias": True}
        outputs = tf.layers.conv1d(**params)
 
        # Readout layer
        params = {"inputs": outputs, "filters": num_units[1], "kernel_size": 1,
                  "activation": None, "use_bias": True}
        outputs = tf.layers.conv1d(**params)
 
        # Residual connection
        outputs += inputs
 
        # Normalize
        outputs = normalize(outputs)
 
    return outputs

5.normalize

def normalize(inputs,
              epsilon = 1e-8,
              scope="ln",
              reuse=None):
    '''Applies layer normalization.
 
    Args:
      inputs: A tensor with 2 or more dimensions, where the first dimension has
        `batch_size`.
      epsilon: A floating number. A very small number for preventing ZeroDivision Error.
      scope: Optional scope for `variable_scope`.
      reuse: Boolean, whether to reuse the weights of a previous layer
        by the same name.
 
    Returns:
      A tensor with the same shape and data dtype as `inputs`.
    '''
    with tf.variable_scope(scope, reuse=reuse):
        inputs_shape = inputs.get_shape()
        params_shape = inputs_shape[-1:]
 
        mean, variance = tf.nn.moments(inputs, [-1], keep_dims=True)
        beta= tf.Variable(tf.zeros(params_shape))
        gamma = tf.Variable(tf.ones(params_shape))
        normalized = (inputs - mean) / ( (variance + epsilon) ** (.5) )
        outputs = gamma * normalized + beta
 
    return outputs

6. encoder-decoder

            with tf.variable_scope("encoder"):
                ## Embedding
                self.enc = embedding(self.x,
                                      vocab_size=len(de2idx),
                                      num_units=hp.hidden_units,
                                      scale=True,
                                      scope="enc_embed")
 
                # key_masks = tf.expand_dims(tf.sign(tf.reduce_sum(tf.abs(self.enc), axis=-1)), -1)
 
                ## Positional Encoding
                if hp.sinusoid:
                    self.enc += tf.cast(positional_encoding(self.x,
                                        num_units=hp.hidden_units,
                                        zero_pad=False,
                                        scale=False,
                                        scope="enc_pe"), tf.float32)
                else:
                    self.enc += embedding(tf.tile(tf.expand_dims(tf.range(tf.shape(self.x)[1]), 0), [tf.shape(self.x)[0], 1]),
                                          vocab_size=hp.maxlen,
                                          num_units=hp.hidden_units,
                                          zero_pad=False,
                                          scale=False,
                                          scope="enc_pe")
 
                # self.enc *= key_masks
 
                ## Dropout
                self.enc = tf.layers.dropout(self.enc,
                                             rate=hp.dropout_rate,
                                             training=tf.convert_to_tensor(is_training))
 
                ## Blocks
                for i in range(hp.num_blocks):
                    with tf.variable_scope("num_blocks_{}".format(i)):
                        ### Multihead Attention
                        self.enc = multihead_attention(queries=self.enc,
                                                        keys=self.enc,
                                                        num_units=hp.hidden_units,
                                                        num_heads=hp.num_heads,
                                                        dropout_rate=hp.dropout_rate,
                                                        is_training=is_training,
                                                        causality=False)
 
                        ### Feed Forward
                        self.enc = feedforward(self.enc, num_units=[4*hp.hidden_units, hp.hidden_units])
 
            # Decoder
            with tf.variable_scope("decoder"):
                ## Embedding
                self.dec = embedding(self.decoder_inputs,
                                      vocab_size=len(en2idx),
                                      num_units=hp.hidden_units,
                                      scale=True,
                                      scope="dec_embed")
                self.dec_ = self.dec
 
                # key_masks = tf.expand_dims(tf.sign(tf.reduce_sum(tf.abs(self.dec), axis=-1)), -1)
 
                ## Positional Encoding
                if hp.sinusoid:
                    self.dec += tf.cast(positional_encoding(self.decoder_inputs,
                                                    num_units=hp.hidden_units,
                                                    zero_pad=False,
                                                    scale=False,
                                                    scope="dec_pe"), tf.float32)
                else:
                    self.dec += embedding(tf.tile(tf.expand_dims(tf.range(tf.shape(self.decoder_inputs)[1]), 0), [tf.shape(self.decoder_inputs)[0], 1]),
                                      vocab_size=hp.maxlen,
                                      num_units=hp.hidden_units,
                                      zero_pad=False,
                                      scale=False,
                                      scope="dec_pe")
                # self.dec *= key_masks
 
                ## Dropout
                self.dec = tf.layers.dropout(self.dec,
                                            rate=hp.dropout_rate,
                                            training=tf.convert_to_tensor(is_training))
 
                ## Blocks
                for i in range(hp.num_blocks):
                    with tf.variable_scope("num_blocks_{}".format(i)):
                        ## Multihead Attention ( self-attention)
                        self.dec = multihead_attention(queries=self.dec,
                                                        keys=self.dec,
                                                        num_units=hp.hidden_units,
                                                        num_heads=hp.num_heads,
                                                        dropout_rate=hp.dropout_rate,
                                                        is_training=is_training,
                                                        causality=True,
                                                        scope="self_attention")
 
                        ## Multihead Attention ( vanilla attention)
                        self.dec = multihead_attention(queries=self.dec,
                                                       keys=self.enc,
                                                        num_units=hp.hidden_units,
                                                        num_heads=hp.num_heads,
                                                        dropout_rate=hp.dropout_rate,
                                                        is_training=is_training,
                                                        causality=False,
                                                        scope="vanilla_attention")
                        ## Feed Forward
                        self.dec = feedforward(self.dec, num_units=[4*hp.hidden_units, hp.hidden_units])

# Final linear projection
self.logits = tf.layers.dense(self.dec, len(en2idx))
self.preds = tf.to_int32(tf.arg_max(self.logits, dimension=-1))
self.istarget = tf.to_float(tf.not_equal(self.y, 0))
self.acc = tf.reduce_sum(tf.to_float(tf.equal(self.preds, self.y))*self.istarget) / (tf.reduce_sum(self.istarget))
tf.summary.scalar('acc', self.acc)

7. train

            if is_training:
                # Loss
                self.y_smoothed = label_smoothing(tf.one_hot(self.y, depth=len(en2idx)))
                self.loss = tf.nn.softmax_cross_entropy_with_logits(logits=self.logits, labels=self.y_smoothed)
                self.mean_loss = tf.reduce_sum(self.loss*self.istarget) / (tf.reduce_sum(self.istarget))
 
                # Training Scheme
                self.global_step = tf.Variable(0, name='global_step', trainable=False)
                self.optimizer = tf.train.AdamOptimizer(learning_rate=hp.lr, beta1=0.9, beta2=0.98, epsilon=1e-8)
                self.train_op = self.optimizer.minimize(self.mean_loss, global_step=self.global_step)
 
                # Summary
                tf.summary.scalar('mean_loss', self.mean_loss)
                self.merged = tf.summary.merge_all()