tensorflow 加载部分变量

tensorflow模型保存为saver = tf.train.Saver()函数,saver.save()保存模型,代码如下:

import tensorflow as tfv1= tf.Variable(tf.random_normal([784, 200], stddev=0.35), name="v1")v2= tf.Variable(tf.zeros([200]), name="v2")saver = tf.train.Saver()with tf.Session() as sess:     init_op = tf.global_variables_initializer()     sess.run(init_op)     saver.save(sess,"checkpoint/model_test",global_step=1)

当我们保存模型后,我们可以通过saver.restore()来加载模型,初始化变量:

import tensorflow as tfv1= tf.Variable(tf.random_normal([784, 200], stddev=0.35), name="v1")v2= tf.Variable(tf.zeros([200]), name="v2")saver = tf.train.Saver()with tf.Session() as sess:    # init_op = tf.global_variables_initializer()    # sess.run(init_op)    saver.restore(sess, "checkpoint/model_test-1")    # saver.save(sess,"checkpoint/model_test",global_step=1)

神经网络训练时,有时候我们需要从预训练的模型中加载部分参数,初始化当前模型,例如加入CNN有6层,我们需要从已有的模型初始化CNN前5层参数.这可以通过saver.restore()实现.

之前我们已经介绍可以通过tf.train.Saver()的保存部分变量的方法,即需要保存的变量列表,同样的,在变量初始化的时候,我们可以对需要单独初始化的变量分别定义一个tf.train.Saver()函数,这样就可以单独对该部分变量初始化,例如下面代码,saver1用于初始化变量v1,saver2用于初始化变量v2,v3:

import tensorflow as tfv1= tf.Variable(tf.random_normal([784, 200], stddev=0.35), name="v1")v2= tf.Variable(tf.zeros([200]), name="v2")v3= tf.Variable(tf.zeros([100]), name="v3")#saver = tf.train.Saver()saver1 = tf.train.Saver([v1])saver2 = tf.train.Saver([v2]+[v3])with tf.Session() as sess:    # init_op = tf.global_variables_initializer()    # sess.run(init_op)    saver1.restore(sess, "checkpoint/model_test-1")    saver2.restore(sess, "checkpoint/model_test-1")    # saver.save(sess,"checkpoint/model_test",global_step=1)