安装 CUDA 及 cuDNN
需要确定自己电脑是否是 N卡 并且显卡支持版本在下面的列表中（具体也可去官网查看），如果不是就跳过该步骤。
CUDA CUDA™是一种由NVIDIA推出的通用并行计算架构，该架构使GPU能够解决复杂的计算问题。
cuDNN 专门为深度神经网络设计的基于 CUDA 的加速库。
安装 TensorFlow
TensorFlow 需要 Python3.5.X 版本（大于 3.5.1）
我已经安装 3.6 所以为了多环境方便使用 Anaconda 安装 TensorFlow。可以到 下载对应的版本后进行安装，安装完成后：C:\Users\zylia>conda create -n TensorFlow python=3.5Fetching package metadata ...........Solving package specifications: .Package plan for installation in environment E:\Anaconda3\envs\TensorFlow:The following NEW packages will be INSTALLED: pip: 9.0.1-py35_1 python: 3.5.3-3 setuptools: 27.2.0-py35_1 vs2015_runtime: 14.0.25123-0 wheel: 0.29.0-py35_0Proceed ([y]/n)? ypython-3.5.3-3 100% |###############################| Time: 0:05:20 100.51 kB/ssetuptools-27. 100% |###############################| Time: 0:00:07 101.84 kB/swheel-0.29.0-p 100% |###############################| Time: 0:00:01 121.63 kB/spip-9.0.1-py35 100% |###############################| Time: 0:00:16 106.15 kB/s## To activate this environment, use:# > activate TensorFlow## To deactivate this environment, use:# > deactivate TensorFlow## * for power-users using bash, you must source#C:\Users\zylia>activate TensorFlow//如果没有安装 CUDA 使用 pip install --upgrade tensorflow 命令(TensorFlow) C:\Users\zylia> pip install --upgrade tensorflow-gpu
测试
安装完成后进行测试：
E:\Anaconda3\envs\TensorFlow\python.exe D:/PycharmProjects/TensorFlow/Test.pyTraceback (most recent call last): File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 18, in swig_import_helper return importlib.import_module(mname) File "E:\Anaconda3\envs\TensorFlow\lib\importlib\__init__.py", line 126, in import_module return _bootstrap._gcd_import(name[level:], package, level) File "", line 986, in _gcd_import File " ", line 969, in _find_and_load File " ", line 958, in _find_and_load_unlocked File " ", line 666, in _load_unlocked File " ", line 577, in module_from_spec File " ", line 919, in create_module File " ", line 222, in _call_with_frames_removedImportError: DLL load failed: 找不到指定的模块。During handling of the above exception, another exception occurred:Traceback (most recent call last): File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 41, in from tensorflow.python.pywrap_tensorflow_internal import * File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 21, in _pywrap_tensorflow_internal = swig_import_helper() File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 20, in swig_import_helper return importlib.import_module('_pywrap_tensorflow_internal') File "E:\Anaconda3\envs\TensorFlow\lib\importlib\__init__.py", line 126, in import_module return _bootstrap._gcd_import(name[level:], package, level)ImportError: No module named '_pywrap_tensorflow_internal'During handling of the above exception, another exception occurred:Traceback (most recent call last): File "D:/PycharmProjects/TensorFlow/Test.py", line 1, in import tensorflow as tf File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\__init__.py", line 24, in from tensorflow.python import * File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\python\__init__.py", line 51, in from tensorflow.python import pywrap_tensorflow File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 52, in raise ImportError(msg)ImportError: Traceback (most recent call last): File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 18, in swig_import_helper return importlib.import_module(mname) File "E:\Anaconda3\envs\TensorFlow\lib\importlib\__init__.py", line 126, in import_module return _bootstrap._gcd_import(name[level:], package, level) File " ", line 986, in _gcd_import File " ", line 969, in _find_and_load File " ", line 958, in _find_and_load_unlocked File " ", line 666, in _load_unlocked File " ", line 577, in module_from_spec File " ", line 919, in create_module File " ", line 222, in _call_with_frames_removedImportError: DLL load failed: 找不到指定的模块。During handling of the above exception, another exception occurred:Traceback (most recent call last): File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 41, in from tensorflow.python.pywrap_tensorflow_internal import * File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 21, in _pywrap_tensorflow_internal = swig_import_helper() File "E:\Anaconda3\envs\TensorFlow\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 20, in swig_import_helper return importlib.import_module('_pywrap_tensorflow_internal') File "E:\Anaconda3\envs\TensorFlow\lib\importlib\__init__.py", line 126, in import_module return _bootstrap._gcd_import(name[level:], package, level)ImportError: No module named '_pywrap_tensorflow_internal'Failed to load the native TensorFlow runtime.See https://www.tensorflow.org/install/install_sources#common_installation_problemsfor some common reasons and solutions. Include the entire stack traceabove this error message when asking for help.Process finished with exit code 1
这是由于 CUDA 版本升级，CUDNN 升级为 cudnn64_6.dll，将 cudabin 加入 PATH 并且复制一份该 dll 改名为 cudnn64_5.dll 即可。
2017-06-01 17:48:32.402786: W c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE instructions, but these are available on your machine and could speed up CPU computations.2017-06-01 17:48:32.403261: W c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE2 instructions, but these are available on your machine and could speed up CPU computations.2017-06-01 17:48:32.403602: W c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE3 instructions, but these are available on your machine and could speed up CPU computations.2017-06-01 17:48:32.403991: W c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.2017-06-01 17:48:32.404366: W c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.2017-06-01 17:48:32.404748: W c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.2017-06-01 17:48:32.405161: W c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.2017-06-01 17:48:32.405527: W c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\platform\cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.2017-06-01 17:48:33.651581: I c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\common_runtime\gpu\gpu_device.cc:887] Found device 0 with properties: name: GeForce GTX 960Mmajor: 5 minor: 0 memoryClockRate (GHz) 1.176pciBusID 0000:01:00.0Total memory: 2.00GiBFree memory: 1.65GiB2017-06-01 17:48:33.652016: I c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\common_runtime\gpu\gpu_device.cc:908] DMA: 0 2017-06-01 17:48:33.652197: I c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\common_runtime\gpu\gpu_device.cc:918] 0: Y 2017-06-01 17:48:33.652404: I c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\common_runtime\gpu\gpu_device.cc:977] Creating TensorFlow device (/gpu:0) -> (device: 0, name: GeForce GTX 960M, pci bus id: 0000:01:00.0)b'Hello, TensorFlow!'
这样就安装成功了
第一个小示例
import tensorflow as tfimport numpy as npimport matplotlib.pylab as plt# 使用 TensorFlow 定义一个随机数构成的 2 * 20 的矩阵，并将其赋给变量 aa = tf.random_normal([2, 20])# a 是一个 Tensor("random_normal:0", shape=(2, 20), dtype=float32) 对象sess = tf.Session()out = sess.run(a)"""out 是一个 Numpy 数组，是类似下面结构的二维数组（具体数值随机生成可能有所差别）[[ 1.12049484 -1.74728727 0.45890146 0.36703163 -0.15883297 -0.75236785 -0.92546529 0.52383661 -0.76638812 -1.83311749 1.89417374 -1.08472025 -0.45552248 0.15569194 0.31735975 0.01950244 -1.07766128 -0.79500329 -0.8666485 0.39396518] [-1.57124853 -0.08938342 -0.55253559 -1.17106044 0.40585777 0.39375332 1.37915421 0.19739513 -0.75434619 0.15933491 -0.75306702 -1.60337615 -1.46187127 0.92681426 0.59408826 0.44740802 -1.06977224 -0.42683071 -0.09084698 0.63764465]]"""x, y = out# 将数组中的元素分别赋值给 x、y# 使用 matplotlib.pylab 绘制一个散点图 x 对应横轴 y 对应竖轴# 其实就是取出 x y 种索引相同的两个数后分别当做 横轴和竖轴显示成一个点plt.scatter(x, y)#显示绘制的图像plt.show() 显示效果如下图所示：
第一个完整数据流图
import tensorflow as tfa = tf.constant(5, name="input_a")b = tf.constant(3, name="input_b")# tf.mul 被弃用 需要使用 tf.multiplyc = tf.multiply(a, b, name="mul_c")d = tf.add(a, b, name="add_d")e = tf.add(c, d, name="add_e")# 上面的代码只是建立了一个数据流图，并不会进行计算# Session 对象在运行时负责对数据流图进行监督，并且是运行数据流图的主要接口 还有一个 InteractiveSessionsess = tf.Session()result = sess.run(e)# a = {Tensor} Tensor("input_a:0", shape=(), dtype=int32)# b = {Tensor} Tensor("input_b:0", shape=(), dtype=int32)# c = {Tensor} Tensor("mul_c:0", shape=(), dtype=int32)# d = {Tensor} Tensor("add_d:0", shape=(), dtype=int32)# e = {Tensor} Tensor("add_e:0", shape=(), dtype=int32)# result 为 23# SummaryWriter 改为 tf.summary.FileWriter# 在 Pycharm 测试的时候发现 run 无法生成数据 可能是 Pycharm 的 bugwriter = tf.summary.FileWriter('./graph_data', graph=sess.graph)# 关闭以释放资源writer.close()sess.close() 其相当于下图的数据流图：
我们保存一下上面代码中的图信息，使用 TensorBoard 来检查下是否是我们想象的数据流图的情况，在上面的代码中：
# 在早期版本下面的方法为 tf.train.SummaryWriter 改为 tf.summary.FileWriter# 在 Pycharm 测试的时候发现 run 无法生成数据 可能是 Pycharm 的 bugwriter = tf.summary.FileWriter('./graph_data', graph=sess.graph) 这块儿的代码就是用于生成图信息文件的（与计算无关，只是计算可以去掉该部分代码），执行完成后会生成一个文件：
然后在命令行中执行：
(E:\Anaconda3\envs\TensorFlow) D:\PycharmProjects\TensorFlow>tensorboard --logdir="my_graph"Starting TensorBoard b'47' at http://0.0.0.0:6006
在浏览器中访问 localhost:6006 然后选择菜单中的 GRAPHS：
可以看到和我们自己画的图逻辑上是相同的。在该页面中可以点击图查看相关依赖等信息：
