rnn_script.py 文件源码

def RNN(X, num_hidden_layers):

    # reshape to [1, n_input]
    std_dev_He = np.sqrt(2 / np.prod(X.get_shape().as_list()[1:]))
    X = tf.reshape(X, [-1, sequence_length* 8*8])


    # Generate a n_input-element sequence of inputs
    # (eg. [had] [a] [general] -> [20] [6] [33])
    X = tf.split(X, sequence_length, 1)

    # 1-layer LSTM with n_hidden units.

    # rnn_cell = rnn.BasicLSTMCell(n_hidden)
    with tf.variable_scope('RNN', tf.random_normal_initializer(mean=0.0, stddev=std_dev_He)): #tf.random_normal_initializer(mean=0.0, stddev=std_dev_He) #initializer=tf.contrib.layers.xavier_initializer()
        # weights = {
        #     'out': tf.Variable(tf.random_normal([num_hidden, num_classes]))
        # }
        # biases = {
        #     'out': tf.Variable(tf.random_normal([num_classes]))
        # }
        weights = tf.get_variable(
            name='weights',
            shape=[num_hidden, num_classes],  # 1 x 64 filter in, 1 class out
            dtype=tf.float32,
            initializer=tf.contrib.layers.xavier_initializer())
        biases = tf.get_variable(
            name='biases',
            shape=[num_classes],
            dtype=tf.float32,
            initializer=tf.constant_initializer(0.0))
        GRU_cell_layer = [rnn.GRUCell(num_hidden)]
        # LSTM_cell_layer = [rnn.BasicLSTMCell(num_hidden, forget_bias=1)]
        rnn_cell = rnn.MultiRNNCell(GRU_cell_layer * num_hidden_layers)
        # generate prediction
        outputs, states = rnn.static_rnn(rnn_cell, X, dtype=tf.float32)

    # there are n_input outputs but
    # we only want the last output
    # return tf.matmul(outputs[-1], weights['out']) + biases['out']
    return tf.matmul(outputs[-1], weights) + biases