一、Dense层

方法一： from tensorflow.contrib.layers.python.layers import initializers dense = tf.layers.Dense(units=1, activation=None, kernel_initializer=initializers.xavier_initializer(), use_bias=False) sequence_input = dense(stack_sequence_output) 方法二： first_token_tensor = tf.squeeze(self.sequence_output[-1][:, 0:1, :], axis=1) pooled_output = tf.layers.dense( first_token_tensor, self.config.hidden_size, activation=tf.tanh, kernel_initializer=create_initializer(self.config.initializer_range)) return pooled_output 以上两者的区别.