对比分析BN和dropout在预测和训练时区别

def bn_layer(self, inputs, training, name='bn', moving_decay=0.9, eps=1e-5):
        # 获取输入维度并判断是否匹配卷积层(4)或者全连接层(2)
        shape = inputs.shape
        param_shape = shape[-1]
        with tf.variable_scope(name):
            # 声明BN中唯一需要学习的两个参数，y=gamma*x+beta
            gamma = tf.get_variable('gamma', param_shape, initializer=tf.constant_initializer(1))
            beta  = tf.get_variable('beat', param_shape, initializer=tf.constant_initializer(0))
            # 计算当前整个batch的均值与方差
            axes = list(range(len(shape)-1))
            batch_mean, batch_var = tf.nn.moments(inputs , axes, name='moments')
            # 采用滑动平均更新均值与方差
            ema = tf.train.ExponentialMovingAverage(moving_decay, name="ema")
            def mean_var_with_update():
                ema_apply_op = ema.apply([batch_mean, batch_var])
                with tf.control_dependencies([ema_apply_op]):           
                    return tf.identity(batch_mean), tf.identity(batch_var)
            # 训练时，更新均值与方差，测试时使用之前最后一次保存的均值与方差
            mean, var = tf.cond(tf.equal(training,True), mean_var_with_update,
                    lambda:(ema.average(batch_mean), ema.average(batch_var)))
            # 最后执行batch normalization
            return tf.nn.batch_normalization(inputs ,mean, var, beta, gamma, eps)

self.training = tf.placeholder(tf.bool, name="training")

x = tf.nn.dropout(x, self.keep_prob)

self.keep_prob = tf.placeholder(tf.float32, name="keep_prob")

binary_tensor = math_ops.floor(random_tensor)
 ret = math_ops.div(x, keep_prob) * binary_tensor
 if not context.executing_eagerly():
   ret.set_shape(x.get_shape())
 return ret