一、简述
TensorFlow 是由 Google大脑 开发的功能强大的深度神经网络开源学习软件库，TensorFlow 屏蔽了底层的CPU/GPU等，并提供API。TensorFlow支持求导、支持递归神经网络（RNN）、卷积神经网络（CNN）和深度置信网络（DBN）。支持Python、Java、Go和R语言等。
TensorFlow 分为CPU和GPU版本区别可参考如下文章：
https://blog.csdn.net/sinat_36458870/article/details/78783587
二、TensorFlow的安装
2.1 安装Anaconda
网址下载：https://www.anaconda.com，在windows环境正常安装即可。
在“运行”->cmd命令行界面中输入：
conda -V
如果显示Anaconda的版本号，则表示安装成功，否则安装失败。
2.2 创建tensorflow
conda create -n tensorflow python=3.7
image.png
激活：
activate tensorflow
关闭：
deactivate
然后在pycharm中安装tensorflow(记得指定版本号为1.15.0)，安装时间比较长，需要耐心等一下。
image.png
2.3 第一个程序Hello World：
import tensorflow as tf
if __name__ == "__main__":
print("==========start tensorflow===================")
msg = tf.constant('Hello World!')
sess = tf.Session()
result = sess.run(msg)
sess.close()
print(result)
print("==========end tensorflow===================")
输出结果：
==========start tensorflow===================
WARNING:tensorflow:From D:/zhangzh/python/lance-tensorflow-demo/lance/tensorflow/demo/HelloWorld.py:8: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.
2019-12-03 10:34:20.288735: I tensorflow/core/platform/cpu_feature_guard.cc:145] This TensorFlow binary is optimized with Intel(R) MKL-DNN to use the following CPU instructions in performance critical operations: AVX AVX2
To enable them in non-MKL-DNN operations, rebuild TensorFlow with the appropriate compiler flags.
2019-12-03 10:34:20.292028: I tensorflow/core/common_runtime/process_util.cc:115] Creating new thread pool with default inter op setting: 4. Tune using inter_op_parallelism_threads for best performance.
b'Hello World!'
==========end tensorflow===================
备注：要删除引号和“b”（表示字节，byte），保留单引号内的内容，可以使用 decode() 方法。
三、TensorFlow程序开发
3.1 基本加减乘除：
import tensorflow as tf
if __name__ == "__main__":
print("==========start AddMatrixDemo===================")
matrix_1 = tf.constant([100, 200, 1, 50, 2, 1000])
matrix_2 = tf.constant([123, 23454, 54, 6, 657, 5])
matrix_add = tf.add(matrix_1, matrix_2)
matrix_subtraction = matrix_1 - matrix_2
matrix_multiplication = matrix_1 * matrix_2
matrix_division = matrix_1 / matrix_2
sess = tf.compat.v1.Session()
result_add = sess.run(matrix_add)
result_subtraction = sess.run(matrix_subtraction)
result_multiplication = sess.run(matrix_multiplication)
result_division = sess.run(matrix_division)
sess.close()
print("add result:" + str(result_add))
print("subtraction result:" + str(result_subtraction))
print("multiplication result:" + str(result_multiplication))
print("division result:" + str(result_division))
print("==========end AddMatrixDemo===================")
输出结果：
==========start AddMatrixDemo===================
2019-12-03 14:08:06.320353: I tensorflow/core/platform/cpu_feature_guard.cc:145] This TensorFlow binary is optimized with Intel(R) MKL-DNN to use the following CPU instructions in performance critical operations: AVX AVX2
To enable them in non-MKL-DNN operations, rebuild TensorFlow with the appropriate compiler flags.
2019-12-03 14:08:06.322350: I tensorflow/core/common_runtime/process_util.cc:115] Creating new thread pool with default inter op setting: 4. Tune using inter_op_parallelism_threads for best performance.
add result:[ 223 23654 55 56 659 1005]
subtraction result:[ -23 -23254 -53 44 -655 995]
multiplication result:[ 12300 4690800 54 300 1314 5000]
division result:[8.13008130e-01 8.52733009e-03 1.85185185e-02 8.33333333e+00
3.04414003e-03 2.00000000e+02]
==========end AddMatrixDemo===================
3.2 TensorFlow的张量
张量，可理解为一个 n 维矩阵，标量（0维，如4,7,8等）、矢量（一维，如[1,2,3,4]）和矩阵(如[{1,2,3,4},{2,4,6,8}])等都是特殊类型的张量。
import tensorflow as tf
if __name__ == "__main__":
print("==========start tensorDemo===================")
# 标量常量定义
scalar = tf.constant(100)
# 向量常量定义
vector = tf.constant([1, 2, 3, 4])
# 创建一个 [6, 8] 的零元素矩阵,即所有元素的值为零
zero_tensor = tf.zeros([6, 8], tf.int32)
# 创建一个 [10, 10] 的零元素矩阵,即所有元素的值为零
one_tensor = tf.ones([10, 10], tf.int32)
# TensorFlow还支持等差排列(linspace)，正态分布随机数组(random_normal)，截尾正态分布随机数组(truncated_normal)，伽马分布随机数组(random_uniform)等
# 定义正态分布随机数组[10, 10]矩阵，平均值mean=2,标准差为stddev=4,随机生成的初始种子值seed=100
normal_distribution = tf.random_normal([10, 10], mean=2.0, stddev=4, seed=100)
# TensorFlow变量,
# 变量一般在神经网络中用于权重和偏置。
variable = tf.Variable(normal_distribution)
# TensorFlow占位符
x = tf.placeholder(tf.int32, shape=None, name="demo")
print("==========end tensorDemo===================")
矩阵的输出：
import tensorflow as tf
if __name__ == "__main__":
print("==========start interactive demo===================")
sess = tf.compat.v1.InteractiveSession()
matrix = tf.eye(5)
print("单位矩阵[5,5]:")
print(matrix.eval())
matrix_v = tf.Variable(tf.eye(8))
matrix_v.initializer.run()
print("单位矩阵变量[8,8]:")
print(matrix_v.eval())
random_normal_v = tf.Variable(tf.random.normal([5,5]))
random_normal_v.initializer.run()
print("随机正态分布矩阵变量[5,5]:")
print(random_normal_v.eval())
sess.close()
print("==========end interactive demo===================")
输出结果：
==========start interactive demo===================
2019-12-03 15:46:19.350444: I tensorflow/core/platform/cpu_feature_guard.cc:145] This TensorFlow binary is optimized with Intel(R) MKL-DNN to use the following CPU instructions in performance critical operations: AVX AVX2
To enable them in non-MKL-DNN operations, rebuild TensorFlow with the appropriate compiler flags.
2019-12-03 15:46:19.352187: I tensorflow/core/common_runtime/process_util.cc:115] Creating new thread pool with default inter op setting: 4. Tune using inter_op_parallelism_threads for best performance.
单位矩阵[5,5]:
[[1. 0. 0. 0. 0.]
[0. 1. 0. 0. 0.]
[0. 0. 1. 0. 0.]
[0. 0. 0. 1. 0.]
[0. 0. 0. 0. 1.]]
单位矩阵变量[8,8]:
[[1. 0. 0. 0. 0. 0. 0. 0.]
[0. 1. 0. 0. 0. 0. 0. 0.]
[0. 0. 1. 0. 0. 0. 0. 0.]
[0. 0. 0. 1. 0. 0. 0. 0.]
[0. 0. 0. 0. 1. 0. 0. 0.]
[0. 0. 0. 0. 0. 1. 0. 0.]
[0. 0. 0. 0. 0. 0. 1. 0.]
[0. 0. 0. 0. 0. 0. 0. 1.]]
随机正态分布矩阵变量[5,5]:
[[-0.29164872 2.2333603 0.02524624 0.5131775 -4.4677563 ]
[-1.9241036 -0.04562743 -0.9280727 -0.8892461 -1.5311104 ]
[ 1.2640982 -0.81486684 -1.3741654 -1.6512661 -0.2854395 ]
[-1.7600229 1.3423405 -0.49197945 -0.6679723 0.16603552]
[-0.8282764 0.15621257 -0.8586982 0.9839028 0.5759689 ]]
==========end interactive demo===================
网友评论