- save_graph.py 保存运算图以及训练参数
import tensorflow as tf
x1 = tf.placeholder(dtype=tf.float32, shape=[], name = 'x1')
x2 = tf.placeholder(dtype=tf.float32, shape=[], name = 'x2')
w = tf.Variable(tf.constant(2.), name = 'w')
ytmp = tf.multiply(w, x1, name = 'ytmp')
y = tf.add(ytmp, x2, name = 'y')
sess = tf.Session()
sess.run(tf.global_variables_initializer())
print (sess.run(y, feed_dict={x1: 1., x2: 2.}))
saver = tf.train.Saver()
saver.save(sess, 'model/test')
- import_graph.py 导入运算图和相应参数
import tensorflow as tf
sess = tf.Session()
# 导入运算图
saver = tf.train.import_meta_graph('model/test.meta')
#加载相应参数
saver.restore(sess, tf.train.latest_checkpoint('model/'))
graph = tf.get_default_graph()
x1 = graph.get_tensor_by_name('x1:0')
x2 = graph.get_tensor_by_name('x2:0')
y = graph.get_tensor_by_name('y:0')
# restore之后不需要执行variable初始化
#sess.run(tf.global_variables_initializer())
print (sess.run(graph.get_tensor_by_name('w:0')))
print (sess.run(y, feed_dict={x1: 1., x2: 2.}))
- 想要调用保存好的模型，只需要get输入palceholder和最后一步operation即可。
