Tensor、Op对象不可变(immutable)。.Variable()构造方法创建Variable对象,包含Session.run()调用中可持久化的可变张量值。Variable对象初值通常为全0、全1或用随机数填充阶数较高张量,创建初值张量Op,.zeros()、.ones()、.random_normal()、.random_uniform(),接收shape参数。
Graph管理Tensor对象,Session管理Variable对象。Variable对象必须在Session对象内初始化。初始化所有Variable对象,把.global_variables_initializer() Op传给Session.run()。初始化部分Variable对象,把.variables_initializer() Op传给Session.run()。Variable.assign()Op,修改Variable对象,必须在Session对象中运行。.assign_add()创建自增Op,.assign_sub()创建自减Op。不同Session对象独立维护在Graph对象定义的Variable对象值。Optimizer类自动训练机器学习模型,自动修改Variable对象值。创建Variable对象时trainable参数设False,只允许手工修改值。
import tensorflow as tf
my_var = tf.Variable(3, name="my_variable")#创建Variable对象
add = tf.add(5, my_var)
mul = tf.multiply(8, my_var)
zeros = tf.zeros([2, 2])#零矩阵
ones = tf.ones([6])#全1向量
uniform = tf.random_uniform([3,3,3], minval=0, maxval=10)#三维张量,元素服从0~10均匀分布
normal = tf.random_normal([3,3,3], mean=0.0, stddev=2.0)#三维张量,元素服从0均值,标准差为2正态分布
trunc = tf.truncated_normal([2,2], mean=5.0, stddev=1.0)#不会返回小于3.0或大于7.0的张量
radom_var = tf.Variable(tf.truncated_normal([2,2]))#默认值0,默认标准差1.0
init_global = tf.global_variables_initializer()#所有Varialbe对象初始化
sess = tf.Session()
sess.run(init_global)
sess.run(add)
var1 = tf.Variable(1, name="initialize_me")
var2 = tf.Variable(2, name="no_initialize")
init_part = tf.variables_initializer([var1], name="initialize_var1")
sess.run(init_part)
var_assign = tf.Variable(1)
var_assign_times_two = var_assign.assign(var_assign * 2)#创建赋值Op
init_assign = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init_assign)
sess.run(var_assign_times_two)#2
sess.run(var_assign_times_two)#4
sess.run(var_assign_times_two)#8
sess.run(var_assign.assign_add(1))#自增Op,8+1
sess.run(var_assign.assign_sub(2))#自减Op,9-2
var_idpt = tf.Variable(0)#0
init_idpt = tf.global_variables_initializer()
sess1 = tf.Session()
sess2 = tf.Session()
sess1.run(init_idpt)
sess1.run(var_idpt.assign_add(5))#0+5
sess2.run(init_idpt)
sess2.run(var_idpt.assign_add(2))#0+2
sess1.run(var_idpt.assign_add(10))#5+10
sess2.run(var_idpt.assign_add(20))#2+20
sess1.run(init_idpt)#0
sess2.run(init_idpt)#0
not_trainable = tf.Variable(0, trainable=False)#不可自动修改
with tf.name_scpoe(<name>),名称作用域(name scope)组织数据流图,将Op划分到较大有名称语句块。TensorBoard加载数据流图,名称作用域封装Op。可以把名称作用域嵌在其他名称作用域内。
import tensorflow as tf
with tf.name_scope("Scope_A"):#建立Scope_A作用域
a = tf.add(1, 2, name="A_add")
b = tf.multiply(a, 3, name="A_mul")
with tf.name_scope("Scope_B"):#建立Scope_B作用域
c = tf.add(4, 5, name="B_add")
d = tf.multiply(c, 6, name="B_mul")
e = tf.add(b, d, name="output")
writer = tf.summary.FileWriter('./name_scope_1', graph=tf.get_default_graph())
writer.close()
graph = tf.Graph()
with graph.as_default():
in_1 = tf.placeholder(tf.float32, shape=[], name="input_a")
in_2 = tf.placeholder(tf.float32, shape=[], name="input_b")
const = tf.constant(3, dtype=tf.float32, name="static_value")
with tf.name_scope("Transformation"):
with tf.name_scope("A"):
A_mul = tf.multiply(in_1, const)
A_out = tf.subtract(A_mul, in_1)
with tf.name_scope("B"):
B_mul = tf.multiply(in_2, const)
B_out = tf.subtract(B_mul, in_2)
with tf.name_scope("C"):
C_div = tf.div(A_out, B_out)
C_out = tf.add(C_div, const)
with tf.name_scope("D"):
D_div = tf.div(B_out, A_out)
D_out = tf.add(D_div, const)
out = tf.maximum(C_out, D_out)
writer = tf.summary.FileWriter('./name_scope_2', graph=graph)
writer.close()
graph?run=-5.png
graph?run=-7.png
参考资料:
《面向机器智能的TensorFlow实践》
欢迎加我微信交流:qingxingfengzi
我的微信公众号:qingxingfengzigz
我老婆张幸清的微信公众号:qingqingfeifangz
网友评论