深度学习（四）：RnnAttention实现中文邮件分类（含代码）

此代码不是学习视频的相关代码，是自己根据学习中文邮件分类后自己实践写的代码。

主要内容：实现中文邮件分类--也可扩展至多分类

主要方法：Rnn+Attention

RNN部分代码如下：

class RNNAttention(object):
    def __init__(self, text_length, nclasses, embedding_size, hidden_size, attention_size):

        # feeds需要的参数----中文
        self.input_x = tf.placeholder(tf.float32, [None, text_length, embedding_size], name="input_x")
        self.input_y = tf.placeholder(tf.float32, [None, nclasses], name="input_y")
        self.seq_len_ph = tf.placeholder(tf.int32, [None], name='seq_len_ph')
        self.dropout_keep_prob = tf.placeholder(tf.float32, name="dropout_keep_prob")

        # embedding层
        with tf.name_scope('embedding'):
            self.embedded_chars = self.input_x     

        # (Bi-)RNN layer(-s) RNN层
        self.rnn_outputs, _ = bi_rnn(GRUCell(hidden_size), GRUCell(hidden_size),
                                inputs=self.embedded_chars, sequence_length=self.seq_len_ph, dtype=tf.float32)

        # Attention layer 注意力机制层
        with tf.name_scope('Attention_layer'):
            self.attention_output, self.alphas = attention(self.rnn_outputs, attention_size, return_alphas=True)

        # Dropout
        self.dropout = tf.nn.dropout(self.attention_output, self.dropout_keep_prob)

        # Fully connected layer
        with tf.name_scope('Fully_connected_layer'):
            W = tf.Variable(tf.truncated_normal([hidden_size * 2, nclasses], stddev=0.1))
            b = tf.Variable(tf.constant(0., shape=[nclasses]))
            self.fin_out = tf.nn.xw_plus_b(self.dropout, W, b, name="pred")
            self.predictions = tf.argmax(self.fin_out, 1, name="predictions")       
            # self.predictions = self.fin_out

        with tf.name_scope("pred"):
            self.pred = self.predictions

            # 计算损失
        with tf.name_scope("loss"):
            losses = tf.nn.softmax_cross_entropy_with_logits(logits=self.fin_out, labels=self.input_y)
            self.loss = tf.reduce_mean(losses)

        # 计算准确度
        with tf.name_scope("accuracy"):
            correct_predictions = tf.equal(self.predictions, tf.argmax(self.input_y, 1))
            self.accuracy = tf.reduce_mean(tf.cast(correct_predictions, "float"), name="accuracy")

Attention部分代码如下：

def attention(inputs, attention_size, time_major=False, return_alphas=False):

    if isinstance(inputs, tuple):
        inputs = tf.concat(inputs, 2)

    if time_major:
        # (T,B,D) => (B,T,D)
        inputs = tf.array_ops.transpose(inputs, [1, 0, 2])

    hidden_size = inputs.shape[2].value  # D value - hidden size of the RNN layer

    # Trainable parameters
    # 可训练的参数
    w_omega = tf.Variable(tf.random_normal([hidden_size, attention_size], stddev=0.1))
    b_omega = tf.Variable(tf.random_normal([attention_size], stddev=0.1))
    u_omega = tf.Variable(tf.random_normal([attention_size], stddev=0.1))

    with tf.name_scope('v'):
        v = tf.tanh(tf.tensordot(inputs, w_omega, axes=1) + b_omega)

    vu = tf.tensordot(v, u_omega, axes=1, name='vu')  # (B,T) shape
    alphas = tf.nn.softmax(vu, name='alphas')         # (B,T) shape

    output = tf.reduce_sum(inputs * tf.expand_dims(alphas, -1), 1)

    if not return_alphas:
        return output
    else:
        return output, alphas