Tensorflow中的變量就是一個放在內存中的tensor結構,用於在計算過程中保存數據,變量的數值可以保存到文件中,也可以從文件中讀取
1.變量的初始化
import tensorflow as tf
Weights=tf.Variable(tf.random_normal([3,2],stddev=0.35),name="weights")#聲明一個Weights的變量
print(Weights)#打印Weights變量結構
init=tf.global_variables_initializer()#初始化變量
with tf.Session() as sess:#執行session任務
sess.run(init)#初始化認爲
print(sess.run(Weights))#打印Weights的值
tensorflow中的變量必須被初始化,否則其內容是空的,以上代碼執行完後會打印出一個3行2列的矩陣,值隨機的,執行的輸出結果如下
<tf.Variable 'weights:0' shape=(3, 2) dtype=float32_ref>
[[ 0.01990979 -0.26959115]
[ 0.32198292 -0.09266231]
[-0.32708889 0.3107968 ]]
2.變量保存到文件
import tensorflow as tf
Weights=tf.Variable(tf.random_normal([3,2],stddev=0.35),name="weights")
print(Weights)
saver = tf.train.Saver()#聲明Saver
init=tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
print(sess.run(Weights))
save_path = saver.save(sess, "/Users/william/tmp/model.ckpt")#保存到文件
print("Model saved in file: %s" % save_path)#打印保存的路徑
tensorflow是可以將變量保存到文件的,要用到的是tf.train.Saver,以上這段代碼執行完成後,就會在tmp文件夾下身材變量保存的文件
3.變量的讀取
可以保存到文件,就可以從文件中把變量讀取出來
import tensorflow as tf
Weights=tf.Variable(tf.random_normal([3,2],stddev=0.35),name="weights")
print(Weights)
saver = tf.train.Saver()
with tf.Session() as sess:
saver.restore(sess, "/Users/william/tmp/model.ckpt")
print("the restore variable Weights= %s" % sess.run(Weights))
以上代碼把變量讀取出來,並打印出來,其輸出如下:
<tf.Variable 'weights:0' shape=(3, 2) dtype=float32_ref>
the restore variable Weights= [[ 0.01990979 -0.26959115]
[ 0.32198292 -0.09266231]
[-0.32708889 0.3107968 ]]
如果變量是從文件中讀取出來,就不需要初始化,只需要聲明就可以了