学习TensorFlow--变量

import tensorflow as tf

#定义变量

var =tf.Variable(<initial value>，name)

get_var=tf.get_variable(name, shape=None, initializer=None)

#两者的区别是tf.Variable能自动处理命名冲突，而tf.get_variable直接报错

#常量:tf.constant(value, dtype=None, shape=None, name='const')

con_var = tf.Variable(tf.constant([1,2,3,4,5,6,7]))

#output:con_var = [1,2,3,4,5,6,7]

con_var = tf.Variable(tf.constant(1.0, shape=[2, 3])

#output: con_var = [[1.0,1.0,1.0], [1.0,1.0,1.0]]

#constant_initializer

con_var = [1,2,3,4,5,6]

con_in_var= tf.constant_initializer(con_var)

x = tf.get_variable('x', shape =[2,3], initializer = con_in_var)

#output:x=[[1,2,3],[4,5,6]]

#当定义的维数大于原有的维数时，会重复输出最后一个值
x = tf.get_variable('x', shape=[2,4], initializer=con_in_var)

#output:x = [[1,2,3,4],[5,6,6,6]]
                      
#初始化0/1:tf.zeros([row_dim, col_dim])/tf.ones([row_dim, col_dim])

zero_var = tf.Variable(tf.zeros([2,2]))

#output:zero_var = [[0,0], [0,0]]

#定义相似变量：tf.zeros_like(var)/tf.ones_like(var)

zero_similar = tf.zeros_like(zero_var)

#output:zero_var = [[0,0], [0,0]]
                      
#常量填充：tf.fill([row_dim, col_dim], constant)

fill_var = tf.Variable(tf.fill([2,3], -1))

#output:fill_var = [[-1,-1,-1],[-1,-1,-1]]

#产生序列：tf.linspace(start, stop, num, name = None),也可写成lin_space

lin_var = tf.Variable(tf.linspace(10.0, 12.0, 3, name ='linspace'))

#output:lin_var = [10.0, 11.0, 12.0]

#注意：区间的开始和结束就是序列的开始和结束

#产生序列：tf.range(start, limit, delta, dtype = None, name = 'range')

range_var = tf.Variable(tf.range(3, 18, 3))

#output:range_var = [3, 6, 9, 12, 15]

#注意：最大不会超过limit，且不包含limit

               
#随机数：tf.random_type()

#tf.random_normal(shape, mean = 0, stddev = 1.0, dtype = tf.float32, seed = None, name = None)

#在正态分布中产生随机数

ran_nor_var = tf.Variable(tf.random_normal([2, 3], mean = 0, stddev = 1.0))

#output:ran_nor_var = [[ 0.90410209 ,1.26647866, -0.2628631], [-0.79755908, 0.35044804, -0.36429995]]

#当seed设置整数时，每次产生的随机数相同

#tf.random_uniform(shape, minval = 0, maxval = None, dtype = tf.float32, seed = None, name = None)

#在均匀分布中产生随机数

ran_uni_var = tf.Variable(tf.random_uniform([2, 3], 2, 4))

#output:ran_uni_var = [[ 2.81662631, 2.69823909, 3.59916234], [ 3.6251266 , 2.05515909, 2.72103643]]

#tf.truncated_normal(shape, mean = 0, stddev = 1.0, dtype = tf.float32, seed = None, name = None )

#从截断的正态分布中输出随机值

#在正态分布曲线中，包含三个区间(μ-σ，μ+σ), (μ-2σ，μ+2σ), (μ-3σ，μ+3σ)

#则tf.truncated_normal()的取值范围为(μ-2σ，μ+2σ)

trun_var = tf.Variable(tf.truncated_normal([2, 2], seed =2))

#output: trun_var = [[-0.85811085,  -0.19662298], [ 0.13895045,  -1.22127676]]

#tf.one_hot(indices, depth, on_value = None, off_value = None, axis = None, dtype = None, name = None)

#indices：张量；depth：每一维的维数；on_value：指定数值，默认1；

#off_value：默认0；axis：要填充的轴，默认-1

one_var = tf.Variable(tf.one_hot([0, 2, -1, 1], depth = 3, on_value = 5, off_value = 0, axis = 1))

#output:one_var = [ [ 5, 0, 0 ], [ 0, 0, 5, ], [ 0, 0, 0 ], [ 0, 5, 0 ] ] 

#one_var的维数分别对应indices的维数和depth，on_value值对应的索引值即为indices每一维对应的值

#再来一个例子

one_var = tf.Variable(tf.one_hot([[0,2],[1,-1]], depth = 3, on_value = 1, off_value = 0, axis = -1))

#output:one_var = [[[1, 0, 0 ], [0, 0, 1]], [[0, 1, 0],[0, 0, 0]]]

#tensorflow中的变量不能直接输出，需要建立会话，有两种常用方法

#（1）

sess = tf.InteractiveSession()

#（2）

sess = tf.Session()

#初始化单个变量

zero_var.initializer.run()

#初始化所有变量

sess.run(tf.global_variables_initializer())

#用tf.Session()建立的会话没有eval()方法

print(zero_var.eval())

sess.close()

#占位符tf.placeholder(dtype, shape = None, name = None)

place_var = tf.placeholder(tf.float32, shape = [2, 2])

y = tf.matmul(place_var, place_var)

var = [[1,2],[3,4]]

#采用这种形式在运行完程序之后会自动关闭会话，否则要手动关闭

with tf.Session() as sess:

(缩进)sess.run(tf.global_variables_initializer())

(缩进)#feed_dict是字典类型，用于初始化变量,place_var只能接受数组

(缩进)y = sess.run(y, feed_dict = {place_var: var})

(缩进)print(y)

#output：y = [[  7.  10.]，[ 15.  22.]]

参考链接：tf.truncated_normal与tf.random_normal

TensorFlow API