tensorflow模型的保存使用tf.train.Saver()
例子:
import tensorflow as tf
import numpy as np
X=tf.placeholder(dtype=tf.float32,shape=[20,1])
Y=tf.placeholder(dtype=tf.float32,shape=[20,1])
W=tf.Variable(tf.zeros([1,1]))
b=tf.Variable(tf.ones([1]))
out=tf.matmul(X,W)+b
loss=tf.reduce_mean(tf.square(out-Y))
train=tf.train.GradientDescentOptimizer(0.01).minimize(loss)
sess=tf.Session()
init=tf.global_variables_initializer()
sess.run(init)
saver=tf.train.Saver(max_to_keep=0)#创建Saver,参数max_to_keep ,这个是用来设置保存模型的个数,默认为5,即 max_to_keep=5,保存最近的5个模型。如果你想每训练一代(epoch)就想保存一次模型,则可以将 max_to_keep设置为None或者0,如果你只想保存最后一代的模型,则只需要将max_to_keep设置为1即可
for i in range(1000):
x = np.random.uniform(0, 10000, [20, 1])
x_mean=np.mean(x)
x =x/x_mean
y = 2 * x + 3
sess.run(train,feed_dict={X:x,Y:y})
print(sess.run(loss,feed_dict={X:x,Y:y}))
print(sess.run(W),sess.run(b))
saver.save(sess, 'ckpt/model.ckpt')# saver.save(sess, 'ckpt/model.ckpt',global_step=i)第一个参数sess,这个就不用说了。第二个参数设定保存的路径和名字,第三个参数将训练的次数作为后缀加入到模型名字中。
最后模型保存在:ckpt/中,会有四个文件:checkpoint, model.ckpt.data-00000-of-00001, model.ckpt.index, model.ckpt.meta
重载模型:
import tensorflow as tf
import numpy as np
X=tf.placeholder(dtype=tf.float32,shape=[5,1])
Y=tf.placeholder(dtype=tf.float32,shape=[5,1])
W=tf.Variable(tf.zeros([1,1]))
b=tf.Variable(tf.ones([1]))
out=tf.matmul(X,W)+b
loss=tf.reduce_mean(tf.square(out-Y))
train=tf.train.GradientDescentOptimizer(0.01).minimize(loss)
sess=tf.Session()
init=tf.global_variables_initializer()
sess.run(init)
saver=tf.train.Saver(max_to_keep=0)
for i in range(10000):
x = np.random.uniform(0, 10000, [5, 1]) # 这个值要小一点才好,太大了要发散(梯度太大了),所以进行预处理
x_mean=np.mean(x)
x =x/x_mean
y = 2 * x + 3
sess.run(train,feed_dict={X:x,Y:y})
#print(sess.run(loss,feed_dict={X:x,Y:y}))
#print(sess.run(W),sess.run(b))
saver.save(sess, 'ckpt/model.ckpt')
#test
model_file=tf.train.latest_checkpoint('ckpt/')
saver.restore(sess,model_file)
x = np.random.uniform(0, 10000, [5, 1])
y = 2 * x + 3
print(y)
print(sess.run(out,feed_dict={X:x,Y:y}))
输出:
[[ 5019.91789035]
[ 9799.95392019]
[ 7643.69704575]
[10588.81737293]
[ 8225.58717911]]
[[ 5019.9355]
[ 9799.988 ]
[ 7643.7236]
[10588.854 ]
[ 8225.615 ]]