tensorflow实现Variable，Tensor，Nump

import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
import numpy as np
weight = tf.get_variable(name='weights',initializer=tf.random_normal([5,2], stddev=0.01))
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print('------------------打印出已经初始化之后的Variable的值------------------------------')
    print(sess.run(weight))
    print('----------weight的类型------------')
    print(type(weight))
    # Variable转换为Tensor
    # Variable类型转换为tensor类型(无论是numpy转换为Tensor还是Variable转换为Tensor都可以使用tf.convert_to_tensor)
    data_tensor = tf.convert_to_tensor(weight) 
    # 打印出Tensor的值（由Variable转化而来）
    print('------------------Variable转化为Tensor，打印出Tensor的值--------------------------')
    print(sess.run(data_tensor))
    # tensor转化为numpy
    print('-------------------tensor转换为numpy，打印出numpy的值-----------------')
    data_numpy = data_tensor.eval()
    print(data_numpy)
    print('------------------numpy转换为Tensor---------------------------')
    ten = tf.convert_to_tensor(data_numpy)
    print(ten)
    print(sess.run(ten))
    # tensor转化为Variable（其实是Variable继承Tensor的结构，但是没有值
    print('---------------------tensor转换为Variable(需要重新进行初始化)----------------------')
    v = tf.Variable(data_tensor) # 此时Variable继承的是Tensor的结构，至于Variable的值，需要重新进行initialize
    sess.run(tf.global_variables_initializer())
    print(sess.run(weight)) # 此时输出的weight和v的结构是相同的，但是值是不同的。
    print(sess.run(v))
 
    # Variable转换为numpy（也是使用eval）
    print('---------------Variable转换为numpy（也是使用eval）--------------------')
    data_numpy2 = weight.eval()
    print(data_numpy2)

image.png image.png

输出

image.png