L2正则化:
reg = tf.contrib.layers.apply_regularization(tf.contrib.layers.l2_regularizer(1e-4),
tf.trainable_variables())
loss = tf.losses.sparse_softmax_cross_entropy(labels=labels, logits=logits)+reg
BN:
BN:
bn的layers实现函数:
步骤一:
添加tf.layers.batch_normalization(input,training=is_trainning)层,训练的时候training设置为true,测试的时候为false.添加位置:激活函数前或着后,目前两种加法没有定论谁更好。
步骤二:
训练时需要更新BN中滑动均值和滑动方差两个变量(根据各批次均值和方差计算得来)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
train_op = optimizer.minimize(loss)
步骤三:
保存BN中滑动均值和滑动方差两个变量。保存模型时默认保存的可训练变量,因此需要单独添加滑动均值和滑动方差两个变量进行保存(也可直接保存全部变量)
var_list = tf.trainable_variables()
g_list = tf.global_variables()
bn_moving_vars = [g for g in g_list if 'moving_mean' in g.name]
bn_moving_vars += [g for g in g_list if 'moving_variance' in g.name]
var_list += bn_moving_vars
saver = tf.train.Saver(var_list=var_list,max_to_keep=5)