参考博文:http://blog.csdn.net/phdat101/article/details/52442738
<nobr aria&#45;hidden="true"> </nobr> <math xmlns="http&#58;&#47;&#47;www&#46;w3&#46;org&#47;1998&#47;Math&#47;MathML"> </math>正常情况下,使用tf.initialize_all_variables()初始化变量,在完全构建好模型并加载之后才运行这个操作。生成数据的主要方法如下:
<nobr aria&#45;hidden="true"> </nobr> <math xmlns="http&#58;&#47;&#47;www&#46;w3&#46;org&#47;1998&#47;Math&#47;MathML"> </math>1:如果需要利用已经初始化的参数给其他变量赋值,TF的变量有个initialized_value()属性,就是初始化的值,使用方法如下:

# 原始的变量 
weights = tf.Variable(tf.random_normal([784, 200], stddev=0.35),name="weights")  
# 创造相同内容的变量 
w2 = tf.Variable(weights.initialized_value(), name="w2")  
# 也可以直接乘以比例 
w_twice = tf.Variable(weights.initialized_value() * 0.2, name="w_twice") 

<nobr aria&#45;hidden="true"> </nobr> <math xmlns="http&#58;&#47;&#47;www&#46;w3&#46;org&#47;1998&#47;Math&#47;MathML"> </math>生成tensor,序列,随机数的一些方法:

# -*- coding: utf-8 -*-:
import tensorflow as tf
import numpy as np

#生成0和1矩阵
v1 = tf.Variable(tf.zeros([3,3,3]), name="v1")
v2 = tf.Variable(tf.ones([10,5]), name="v2")
#填充单值矩阵
v3 = tf.Variable(tf.fill([2,3],9))
#常量矩阵
v4_1 = tf.constant([1,2,3,4,5,6,7])
v4_2 = tf.constant(-1.0, shape=[2,3])
#生成等差数列
v6_1 = tf.linspace(10.0, 12.0, 30, name="linspace")
v7_1 = tf.range(10, 20, 3)#just int32
#生成各种随机数据矩阵
v8_1 = tf.Variable(tf.random_uniform([2,4],minval=0.0,maxval=2.0,dtype=tf.float32,seed=1234,name="v8_1"))
v8_2 = tf.Variable(tf.random_normal([2,3],mean=0.0,stddev=1.0,dtype=tf.float32,seed=1234,name="v8_2"))
v8_3 = tf.Variable(tf.random_normal([2,3],mean=0.0,stddev=1.0,dtype=tf.float32,seed=1234,name="v8_3"))
v8_4 = tf.Variable(tf.random_uniform([2,3], maxval=1.0, dtype=tf.float32, seed=1234, name="v8_4"))

v8_5 = tf.random_shuffle([[1,2,3],[4,5,6],[6,6,6]],seed=134,name="v8_5")
#初始化
init_op = tf.initialize_all_variables()
#保存变量,也可以指定保存的内容
saver = tf.train.Saver
#运行
with tf.Session() as sess:
    sess.run(init_op)
    #输出形状和值
    print tf.Variable.get_shape(v1)#shape
    print sess.run(v1)#value
    #numpy保存文件
    np.save("v1.npy",sess.run(v1))
    test_a = np.load("v1.npy")
    print test_a[1,2]
    #一些输出
    print(sess.run(v3))
    v5 = tf.zeros_like(sess.run(v1))
    print sess.run(v6_1)
    print sess.run(v7_1)
    print sess.run(v8_5)

运行结果如下:

(3, 3, 3)
[[[0. 0. 0.] [0. 0. 0.] [0. 0. 0.]]

 [[0. 0. 0.] [0. 0. 0.] [0. 0. 0.]]

 [[0. 0. 0.] [0. 0. 0.] [0. 0. 0.]]]
[0. 0. 0.]
[[9 9 9] [9 9 9]]
[10.        10.068966  10.137931  10.206897  10.275862  10.344828
 10.413794  10.4827585 10.551724  10.620689  10.689655  10.75862
 10.827586  10.896552  10.965517  11.034483  11.103448  11.172414
 11.241379  11.310345  11.379311  11.448276  11.5172415 11.586206
 11.655172  11.724138  11.793103  11.862069  11.931034  12.       ]
[10 13 16 19]
[[4 5 6] [6 6 6] [1 2 3]]

按照结果也是比较容易推测这些方法实现的功能