备份与恢复模型
基本操作
1
2
3
4
5
6
| saver = tf.train.Saver()
with tf.Session() as sess:
saver.save(sess, saved_model_path)
with tf.Session() as sess:
saver.restore(sess, saved_model_path)
|
部分恢复模型
首先通过 tf.trainable_variables () 获得新模型的所有可训练参数，然后把新增参数剔除。新建一个 saver 对象，并把要恢复的参数传进去。
1
2
3
4
5
6
7
| restore_variable = tf.trainable_variables()
<do some modified jobs here>
saver = tf.train.Saver(modified_restore_variable)
with tf.Session() as sess:
saver.restore(sess, model_path)
|