CS 20I : note 2

TensorFlow Ops

这是阅读 http://web.stanford.edu/class/cs20si/syllabus.html slides_02.pdf 的笔记.

• 基本操作
• Tensor 类型
• Project speed dating
• Placeholders and feeding inputs
• Lazy loading

第一个 TensorFlow 程序

import tensorflow as tf

a = tf.constant(2)
b = tf.constant(3)
x = tf.add(a,b)

with tf.Session() as sess:
    print(sess.run(x))

5

使用 TensorBoard 可视化

TensorFlow 是一个可视化的工具, 可以展示运算过程中的计算图等信息.

import tensorflow as tf

a = tf.constant(2)
b = tf.constant(3)
x = tf.add(a,b)

with tf.Session() as sess:
    # add this line to use TensorBoard.
    writer = tf.summary.FileWriter("./graphs",sess.graph)
    print(sess.run(x))

writer.close() # close the writer when you're done using it

5

在本地文件,可以看到 graphs 目录中有一个文件.

启动 TensorBoard

打开 Anaconda Prompt, 激活到安装有 TensorFlow 的环境. 然后 cd 到 graphs 上层目录.
运行 tensorboard --logdir = graphs, 然后在浏览器中输入 http://localhost:6006/ 即可进入 TensorBoard.

在网页中点击 Graph 就可以看到一些东西了.

注意到这里两个是显示 Const, Const_1 , 那么如何给它们改名字呢?

import tensorflow as tf
a = tf.constant(2,name = "a")
b = tf.constant(3,name="b")
x = tf.add(a,b,name="add")
with tf.Session() as sess:
    writer = tf.summary.FileWriter("./graphs",sess.graph)
    print(sess.run(x))

5

再通过 tensorboard 可以看到 graph 对应节点有名字了

constant 的更多例子

tf.constant(value, dtype = None, shape = None, name = "Const", verify_shape = Flase)

import tensorflow as tf

# 注意 a,b 维度不同, 在运算时发生了 boardcasting
# 更多关于 https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html
# https://www.tensorflow.org/api_guides/python/math_ops#Arithmetic_Operators
a = tf.constant([2,2],name="a")
b = tf.constant([[0,1],[2,3]],name="b")
x = tf.add(a,b,name="add")
y = tf.multiply(a,b,name="mul")

with tf.Session() as sess:
    x = sess.run(x)
    print(x)
    print('-----------------')
    y = sess.run(y)
    print(y)

[[2 3]
 [4 5]]
-----------------
[[0 2]
 [4 6]]

import tensorflow as tf

# 生成指定维度的零 tensor
# 一般定义: tf.zeros(shape, dtype=tf.float32, name=None)
v1 = tf.zeros([3,4],tf.int32)

# 值填充
# tf.fill(dims, value, name = none)
v2 = tf.fill([2,3],4)

with tf.Session() as sess:
    print(sess.run(v1))
    print(sess.run(v2))

[[0 0 0 0]
 [0 0 0 0]
 [0 0 0 0]]
[[4 4 4]
 [4 4 4]]

Variables

# 官方文档: https://www.tensorflow.org/programmers_guide/variables
## 创建变量

import tensorflow as tf

a = tf.Variable(2, name = "scalar")

b = tf.Variable(tf.truncated_normal([2,3]))

# 初始化变量

# 一次初始化所有变量
init = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init)
    print(sess.run(a))
    print(sess.run(b))

# provide name and shape
#v1 = tf.get_variable("v1",[1,2,3])
#print(v1)

2
[[-0.89090335 -0.62117159  0.95811218]
 [ 1.0819943   0.74583757 -0.73409617]]

import tensorflow as tf

# 变量初始值为 10
w = tf.Variable(10)

# 赋值为 100
w.assign(100)

with tf.Session() as sess:
    sess.run(w.initializer)
    print(w.eval()) # 但是结果仍然是 10, 为什么

10

import tensorflow as tf

# 变量初始值为 10
w = tf.Variable(10)

# 赋值为 100
assign_op = w.assign(100)

with tf.Session() as sess:
    sess.run(assign_op)
    print(w.eval()) # 这时的结果就正确了

100

Placeholders (占位符)

TF 程序通常有两个阶段:
- assemble a graph
- use a session to execute operations in the graph.

import tensorflow as tf

# 创建一个 placeholder
a = tf.placeholder(tf.float32,shape=[3])

b = tf.constant([5,5,5],tf.float32)

c = a + b

with tf.Session() as sess:
    # 用字典来初始化 a
    print(sess.run(c,{a:[1,2,3]}))

[ 6.  7.  8.]